From feb48f0ec3080cad34b0650f4ff0caa9bc3fedcd Mon Sep 17 00:00:00 2001 From: Morten Hjorth-Jensen Date: Mon, 21 Aug 2023 13:50:43 +0200 Subject: [PATCH] updated --- .../_build/.doctrees/environment.pickle | Bin 167952 -> 209056 bytes .../_build/.doctrees/exercisesweek34.doctree | Bin 0 -> 37724 bytes .../_build/.doctrees/intro.doctree | Bin 52265 -> 52579 bytes .../_build/.doctrees/week34.doctree | Bin 0 -> 778590 bytes .../_build/html/_images/week34_101_0.png | Bin 0 -> 16238 bytes .../_build/html/_images/week34_103_1.png | Bin 0 -> 23846 bytes .../_build/html/_images/week34_74_1.png | Bin 0 -> 38196 bytes .../_build/html/_images/week34_86_2.png | Bin 0 -> 66200 bytes .../_build/html/_images/week34_86_3.png | Bin 0 -> 17003 bytes .../_build/html/_images/week34_93_0.png | Bin 0 -> 24893 bytes .../html/_sources/exercisesweek34.ipynb | 302 + .../_build/html/_sources/week34.ipynb | 4962 ++++++++++++++ .../_build/html/exercisesweek34.html | 603 ++ doc/LectureNotes/_build/html/genindex.html | 17 + doc/LectureNotes/_build/html/intro.html | 19 + doc/LectureNotes/_build/html/objects.inv | Bin 810 -> 873 bytes doc/LectureNotes/_build/html/search.html | 17 + doc/LectureNotes/_build/html/searchindex.js | 2 +- doc/LectureNotes/_build/html/week34.html | 4364 ++++++++++++ .../jupyter_execute/exercisesweek34.ipynb | 327 + .../_build/jupyter_execute/exercisesweek34.py | 147 + .../_build/jupyter_execute/week34.ipynb | 5975 +++++++++++++++++ .../_build/jupyter_execute/week34.py | 2589 +++++++ .../_build/jupyter_execute/week34_101_0.png | Bin 0 -> 16238 bytes .../_build/jupyter_execute/week34_103_1.png | Bin 0 -> 23846 bytes .../_build/jupyter_execute/week34_74_1.png | Bin 0 -> 38196 bytes .../_build/jupyter_execute/week34_86_2.png | Bin 0 -> 66200 bytes .../_build/jupyter_execute/week34_86_3.png | Bin 0 -> 17003 bytes .../_build/jupyter_execute/week34_93_0.png | Bin 0 -> 24893 bytes 29 files changed, 19323 insertions(+), 1 deletion(-) create mode 100644 doc/LectureNotes/_build/.doctrees/exercisesweek34.doctree create mode 100644 doc/LectureNotes/_build/.doctrees/week34.doctree create mode 100644 doc/LectureNotes/_build/html/_images/week34_101_0.png create mode 100644 doc/LectureNotes/_build/html/_images/week34_103_1.png create mode 100644 doc/LectureNotes/_build/html/_images/week34_74_1.png create mode 100644 doc/LectureNotes/_build/html/_images/week34_86_2.png create mode 100644 doc/LectureNotes/_build/html/_images/week34_86_3.png create mode 100644 doc/LectureNotes/_build/html/_images/week34_93_0.png create mode 100644 doc/LectureNotes/_build/html/_sources/exercisesweek34.ipynb create mode 100644 doc/LectureNotes/_build/html/_sources/week34.ipynb create mode 100644 doc/LectureNotes/_build/html/exercisesweek34.html create mode 100644 doc/LectureNotes/_build/html/week34.html create mode 100644 doc/LectureNotes/_build/jupyter_execute/exercisesweek34.ipynb create mode 100644 doc/LectureNotes/_build/jupyter_execute/exercisesweek34.py create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34.ipynb create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34.py create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_101_0.png create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_103_1.png create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_74_1.png create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_86_2.png create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_86_3.png create mode 100644 doc/LectureNotes/_build/jupyter_execute/week34_93_0.png diff --git a/doc/LectureNotes/_build/.doctrees/environment.pickle b/doc/LectureNotes/_build/.doctrees/environment.pickle index 52148b4e5bbd85bdd554d10616211a5e47622d27..d85932bded7dfda18e5a474a812a8c5848ed51bb 100644 GIT binary patch literal 209056 zcmd?S37lNVRWIyC`#PgtUa!@THL}$+mSlO6EyttP@@PYn?IgB5-08kE-L0PPb}t%@ zZDn8Vq_Yrr^0Gi6VdomaT4D058_-0qFh1ed$D7^- zsCw;b@r~0g6dMka7#<$(v{#kgR)yNNn$6(dH5F(~{7A9V^y-st4nMv)S9TkXVh)l{ z6f1eJUT95>x9a6idtIT~tTpzGj7-n+li_L|3RzWZ)n@T}*zSr>dm$v5@H)@6m*%Qv zm)|dJX|-3>+-70CSse#DjZR@pdsW@5RU5@-wLZ%~@^J7Yx$1PSR+t4Rjp3O2f1B^puQ`PEJ*&BB&Zh5weUmP#yiMth0Xx$yJ zxaXnXPJ8dk>Wo+S^3KGpbD!%J>fYquEdjs27KFl~mo4`{#AFEEh!evrN|HX1Fj;iKDm zuGQWk?4?m{)pK6CSUES|XiZKQFVL?psDfjTeQVk+!T{aCZi`K_=37T=wQ>>qIp8*3XOxuLfDO8p zyyJIUC+N3OzT0XRs?h4XW7T@ot2hsq@L9+{BszSDET=YGFHRL;(hGLpzHLyU|YS=D!bAOPaVQyJrwpK;AfJQ)M8Bq1e3^7U&T7=DkI-a+VYj{7|1+>djBx%ps&t+92L5&!-wul@fd39BIrJzRA3B8% z?M*%bU1wflcj3;$n&XAFg>{AXg$;#`g-wN>g)N0$&lc`@u5fqZp2D8Oy@mSEpG7dV5mIJ|8iagffivdg!c$Shl*3f;;V&C z8g8M~GxRCGT9&Kkz3fD}nuA`J9mN2O@tK~qlpzC-42E$RgXYGB{xSbzaped=w)-`OaiE{ufJ0do*tFH<$-TCt#*<7(FKkhoRwho_*+h_}_ES zJ=^R|6n+_wy7;A!wg(zbw@I?YDHD@xKywA=8@D{AeCf6Ko$}5Xy&0!E>6{K%m3YFrM@Lq(Ya1g zh@mNPgBtV7b*DmEk?StyYp;N^~v*WPphbGGn8+;Xuw3xV_Km5_JdAGk-MgCAHOV0e3%?_=PG?PKVM z->dL??ezyduNFF-fl9hExgMTG9qRNNGoE*DSIj0>XkU8mn;%8P@~ngL1zK{;+a2;x zSm`0L$nSZ_-_V(8F9~~S;X~+JrS(`ipY8CYu&?r?&GM1bYx&vr@|oT(xyenkNwkCV z;}^-tfkg)@=LXyK27&ImK~`@NfWQrQ=nVqRa)aHn!5saL+ukGF2K)v+kQ?mP8w4GR z8;t4=f-c1k4(bhpZov(X$Od``;U>pqlYq^@^0~ne$OfV927Q2AJe<%X3e8IsT12rq zo6sT(jw@Sqv3};plkKAYfFBpb$8d7T>sUv23LnP5roE<%iMPE%xCp^a++JLo$O^Bu zHPK$iD{z_sJB1InSJUF9Q$_m8V_Pgzn{Vso%ZdR%y#0H zasSaXK8eWR&#)IZ?$$lG(`d9;vsZ&(6;rHO_u%K3A&ybNSPzn&!aZFUrUlH2ZhO&r zw~Q%^{R;ea3FA9liIajrr+qboU(}5w;`!Zbx2(JVEFzLqVL;7@MYnL3hxO1jkS%4X<$(YkBrvH z?WL{eWcKcCqc|0^L92}LjTL!s(#6^urU|b|YGM(zuV9%_Lk`EXBtgHHxG?7ZO<^+-Ht=K>px+S}=ux zuPGqZQKoQ5sCL-C&v__Udr&r${URXPUPVi9vh4=NLJDFz|3rN8I$f+3;T2DLyGCH8 zh_TM%W29W17-93@HGIzqIkH(uaSoxW#)$9BQm5r`B$lqa1$ z1Hi|R(+DxZ-=X^iVV;W}@nFTYTcI#vLj*~s2`|CgwpPrYo9#R&+O6{c9xqN+D8vP~ zg6OV`ehpKd!Vt#~u}lxuv({fUVWEV7jpw}@jA-o2;CbP_g&z+7&11D6s-n&=?Ycck zWAM25By<~$`jX}4;vxT#(wUYu5r@Q}M=>kO&~o|~@gT2`*qx-Y4B@i+IP`(71^Mmr z2DIILvPDjqLSv2c^|;HPYWwOZ4|o!bb!rnoj>Kd8(qNnuW93yP7i0etvBN+MDhiif z4QQ|`rZKPMwL!;K`g%h^ z#P1?k2SnN|8a+X*PZ}~Snj$Ee{9$r@1y2S9acorO>P4}o6vKj_!xh6YhY)17 z9(z70-k+%>zE+u{SrQJyGQ{+YPvI4}9I`-B!Ek(CEkZ)Vx`Rn7ct(-K@ZX3V2LHxB z34Ek+2w88ABb1Flny1^*6yg5ngpfUlwU z*-M|>f=NvY?v5ZJPQTM>Kh`t)GE6L~|G=eF_XEz7<|CI*GtdadswLr<_Ka{4^_H~H z#AzFECzAn=z%ImgnU9|0mI(XA41mo--|-3!YrS}e#Y%I$RfC)3!RHjddEn}EjI@Cv z44aHJ>%?A#2Py*C*pLz~fZAfMAkO{;qSyH%?I%4O65z+=!qz;5K`|~M=GR`w01OAy zn7>21maLL?nXvg-AFob`C5;F^hV;1aDzxNJVRIN=R6ZIPyJPLEe*3>lc{o6Y;Epm_$5K>j$x(i>nCract#GM=Wu1e`&8>L|${xH>8^Q(xW~yz@)5{$Eua5n=dd1-{ zhhg<)IGRHvj}6~?+l!yL`RR*8+n#ys@w1OVes*MPyR-H2$8Wx6>$ZEHkrC(STbyYR zD_RjTpDAEfNijeQyZgkJU0k*d?e@bhp1>l?nJHj@+eL7_HtX zEtbp9gy+-{(EtsK!&fkw;va6)nQ_CGL=$$_DvLH1XR7W_h}L8xB6{u%;na#AMgoew zi${VGzggMZq@O}J$UDV~!-0`yp}JU+`@!fL&>W(-_mg>6n%f2Y%>7AWnbcxC1c7qB z3pHo2GwGHa-gbwqk0@}fqhxw`h`k#;W256I@>^2OK{B>zW3n>(^}xvX!^S{q%a-j9 zS=E7R?f~tHiLVG$f$Dj*5$l`Uz2V;ioqPE?baDY*$Y zPosg4GNoU@k_G32d~wnl3Xllwn{C|3rh<(azZYzMVrZ*#TOyqJdh52K;aj(D-6qIh zbg=#F`Tb=cdXb(+h{kPYn?{|bhUObh-x5igWEYQ}mFThLhpH{)k@!X?A5K-93IM_a zBM+s*m1q2mgw^3{lZ{dNO)`lg*(B&igfaTTaD!&4p{VN|fG;iqABMDc~?fKP9PC+lWZRwN|4rq`V$&OWZ`e<1M$S zKNjbM=<{%Klt5zYLY&8To{f^$%!b~j|7q{*B3u-P$IYFHmVDiMm3-q-Y$6Kr2^4jB z*yh|qV;gZuk0P{4VER3zI*5Fs#*o?r{Wk*;B5x8mz)RZFP4cg>p)WN3%1sGFU|gN^ zs6wF@8y$g?kb+`QL))Cad-pn9x0+HHTw|6Z$uJ@wnZ%4E$ToO`mgXR~91N8heI`Zk z!?zWugk3?>v4e#k@@m{Nzb8lqDh|8Va03}`L+72n_dECxSfyzo(r=V02E9ksN(m`j z+|Rg2OLk8Q*99{w9PT2NG38)*Bc;%Q>7K6^^A4p4W$Ueq7$(L1i`kXdsUe@r4 zb>ZBJAeNt9NE-=xKQaOJtDZMam1^y_Ps$;W-ebDE;Y#WJYOHKc+HVYZFH9k&3+v_E{$3Pz$fvP}Jvw^)@WZ2r@#@i| z50s0PiY#9%Ad7!5MK;>G}ga} zQgM)vf|Wtg!oemJhEA~@;K)E$N%G4I+RlqY3eMw@*xwp@>?lmBvi&f!PtKivxJ_rZiyVYLX%H8oJ>hO)_Cc&P{LRM^!e~|>Cs&xZ5RuPG zA?c?d#|YugP`C#v#!o+v4nhuEg97dspQh=b{<g-ttd35GsS7EXh(UAK^5F;S9~avfG4p(gdLr0uEvp_8&AM>f}|eCf^ai{&Dzgw zSb412pqx|$K&!1rClFvw@H6@dIuJZkh9#k<4Z(xN5in$2s-?Xo*kfnQS|1RoKs@A_ z6w4U)TE1>J1rKxeDw0#fP5A(KqK|}h)+`NJ2#7Aeb#M0NTW+%jENCv zHM~j|-7{>qNjCF;G1z!+U%5jzTP2%`uzlk1oSUFnG39!O16o#+7?32;kPgBm5}|+t zhSBKrjq+Vi&IpF;=R!d?%O=QQK}@_pJzj<%fQd!^j-s;${4eeVr6n{j26|K2=Wei}8oZ)tETqft42JIn59C{p{2{DI!V^XQS*I#?kEOx~U{-9Y zkrW0>NfO@+BORnEDB7yMlzRkxh7R{Ju?!WBlJ{2RbDC>|rL3|7dJ;|zb(shw;hJZ_ ze=M@IMBM-~uP{@%5dO|!Mvb{CCg=X>E-w{TmqL~fxl$aS4RGN zRphT%<1dBGYodR?Hu~r5@Mr1u;y?aMa=nc0`b&Rdd6FU>+X?0dI=U178%KG(C8#|(;CM0=??hJsz8Vn~Y4F-6EudsrT0 zo0ZmPkt`3f%}xs$oR@a%=Ky{g&R+O#w?lx16Je8$f<&HHR{ms+bGLMMOHQ;*MQy@~ zin4`HB##gu)GM28Qhq02MQ#usq9WH(VRx-)!@J$^WJP2~A+b(8_P-zZ7mS_K1zM{I zATWtwlxoZ(pG0Ko1<&=Pi-kwRm-z>1Sph#8hkDuXmr*2EMDV-;ka_1&wpb-t6b`nZ z#T*i61q2=`r?!gyc}2oC;&t`lWyC^RI6RhnqU@|)`E;j!wA??6 z8oq;Lu>L7L!VW|36apP^<0t|eB3^i|@WzmE_(Bjc!O=bPkI|7sDO$+BU&$dwk&Ohe zaqLRkeWxI!{|&DvFpaktA(TbOr;oQ6!w2MK_f>uxN%+O$lm4UdWLr>qp|F%vxZ6wM z6mSytTVat)vHB86xOQRD;s5(?1nI=0LmV-~ph_bwSWxvu)@D;i(=z4TML*KbFbEQlK6l z@Ckniza8*=hy3SV@}GBh-UdX%w14Wd1HOWJo!l7Cb;hm=&fx|}@Q1;D1u26bj!<72 zoH~>bi?o2l!T`^LvK%2bRNWX`5GDf!Uvu^i+Ha2iddWTSa%_Ou@Xr&6zeJ-seZM~T z7CVETv8%P;r$e21A@t)+W|2Y<1D#YR+xK^8Ub^;L2d>@B>#z1%O zOMT~V^^d#g4;4c7wVi6qJJmlTPXxD#^A3mwoZg|n&isVa3jL;>8Qx37nDLvw7ERj=e2okBmZtF)jh#!$0maF1 zEJo?ig&bHpR$7m<_rf8C?{IBP>BJOy?zI2l=`A$dZ`tE);o%T@TU-2Lo4&+Bqv0L5 z)0fkTr;}4g-;!J-7qapV8WE<8RBA*xQ)naU;#4-B*ozY0brB@PN=ssSMuh#gyQpzx z!uO%z;?VhQuTCu_F)`AC5k7`P&oH8Ixd`XB!a6Dan*aIh_&l~@Md>&2Ikx%*qJ(Oa z5~!Ilu%Dn!WVrdta^bcp^>D7;iYT| zD_$T#MHIQQX=yL(aBSD>*P}}Qgm|r3q&VR`<>xMHD*NtCX_=dm|0E?te!u@}l}>*YcT zUF>4)Yl>N1*=m$tF<0wj@4r%MeifR(a?a+jjW?&@-W&QUAzMynH8R#PDmx7*9ShT_0Z{r5o>3AV@<5qi-*mn>Sd=Xz1I~0`M)EK5SuZZw2 z6fF?r_3Ohgt-CY+!EfwDB0P%n?#R!e%*Y|=w!PpC0(BH6zD3lS$BqvU1|33i9p_NB zj@s3E=Qs*8CZN4UBuQZD@h z>W~B*APafB6aQL*dLZ%-N+rbZjcBtp%pt`aW7wx^FGhV@Y8&dJH6h}TrCagStQEmL z(rkwD4HU71MA><|T0T$WA1F?WQdCV8C3FrVDN$t2JAAS$QWY_(3mXEZyYSPc2k{^1 zq;xkvNws_M3AXOzkNf%K0sh#>9|y$8*lM>`I)s0g4&y(NlR}I0$a(#K?0@4V21zAg zRng^|rXjrZv}Nue6+dmm{XmR+iV4oc{cWdEM7s_HZsO3I;2h_Y$SxauoD+zSQ&|E6 zkxHjV;W2uXC7ZOCY%nGn1^VWZY+ILNr#KO=$mJH1`b3pzLyQdGW2su!QZ0!|MUlIC zq;kR{tCSp!BCB%?GkERD%Pfwy^N$Q{I{ty&m%k0WZ;y3T2s8p;!tMCoW13Jk$~|;vCU2t~DFQ^s@XSYw~$(O)iUR zlA?F>Xma;H7dKKkNZv$Q1}d+BlS(I%rkN|&Fcj@aLVdMDT|2z|VCdxj(QQ%PSkho= zKVU8G@|d&~rkh9F)&9&Gj8IWIw5QDBFIscBGR7gr(B|Rr=KW!jVo@65;R>#;@apGL z$2T9={*AJo?iH5heH0UL}_8@X9v{#MWX0l8Y!0j&#h&@GA28Pz2=eqnp0HBQiymI2W_fUrNb{E9+hma zhRpZhNBQoNM~a&^gohS5=5W#|ay5@wgT66$lP2Lg3vu$is>kXJ)p-*+8k9LabTgF@He2OSkpXOSff~t|UAFhkj1W}DAH(}0Oed3(rZu4+Hcrdui0L5FpbIz_iqHHHTXP%$y4bSap zNMmi#1Gb6z3B|tWj$@cyBNnk9lNVXw_eBLRy&ZpJ9PRMOJNe^X{P7;~;df>hMa1;x zk!F-`i2ww9oFkNb?%$O`y_nVxeBa7;Cm5U9*iKR@J@RyLumpjmh>$&?$Fl!CJ0JM9gw274#V|@|8VF5xpTCxDlsfaD!Fe4~!1dd2Ol&1mD9&?Y~-K?%xR0 z!@D&TQNy$WP}K-acan}pNqk-xuyuIinCfG z_KDyog$PRr@gIMRSt=#sAOAh$ANMIGsFEpuN>D6)93P3Wls?(@$!EJfDgA!82SWeX z9H8_lnuEQIGybbjjbBH z5*5%zH214ug_-(CW2Sr(e-o1UrXXPfO34zFTMk~J>#4jw4h~>bxX!$?z7B%Hp+Z9U|xkS3AjOH4mIFbS(0cg<0GqUxuDdw z_t^ry4gm(|bak?c3hQ0EU2+rVYLBseDF?sy`V95626L9&)yA&UJKJ5x3~89%VM}5c zUR1&I40eN{=m9G5Ri!hl*m0y|3Zj~99%T!uAYo*0 zhZk_77%=)D>UZ(EJatlS_kbSjtZWQnTC?1Tc_5HE>tY^2F)y4eLup)^8dzq0@P1a7 zuKLT8%rxeNGArXwf&~|idDkp>QjoZe)5UfBaq})t)ro~7&KHunbWN2bj@Lvmp1Uc- z_ZDLnq=oPL3@X?Wr&&an!DYy4mJb5)o-9q|S0&8|Z!%^~!{%9A60$qLj^XXroo%Or zjb~6x6ji!W&y9-f3hZ#!EVlaU+KQq3Wn=NuLboY{uw$_zfFu3{VuC-lKPAlxA2#Mm zL+yjMB>Ilp^}>5b{dJ5)WdWYpKX1au=}y;3-DMC=EYKnNjxh@w z1mChH(RT<2DQAjibCEMecM(L}>4V5r~820?~6{rak=|X8+&qxc0d7?mbH)SZx#)77W^7@RA7`gj%MuM&yTdadvG-gZ# zanhDV-$7i3t?pJC^?>-s@Mu4bKV+bO$XJ-PK&{T8B&5nsq<<~Zp?QljD;hL!vL(@X zXf_-}F_q#4$C<`8utnV8;g>*<43*s)F;xH6Sh%!MZOEW_4!Fz4n2R8`R7dQWjXBZ~ z`>-vEIARv@nQ=HmtXf{LTab`k%oF0VT@$W9GL5CY$f-tB85PjM21RUD(WMQG_H|~_Y+iXen9k7*;6dP0(JIN`EJ!5ztF%~5) zJgYKD3s!hhP)S&DoHph|L*=9`iN2$9Z6jAaS8Qg>$TzMym`OPQHepj@&X$bW)5c<^ zMeMqa%4aslr-EO?TH}H-R~mjTTM~W8FV=G+er)U+WAtOjqNK%WL8h2TDS5edkg&M; z5#wLiuy~IxiN0fTdwA?HzUPPSgfplhR;^E#t1}`>GluhBv9Ox?>x}3pjTMp>(H$9a zGjP(59yTDTCK^j5A~+~e35gW=6ILp}ZOp0$|8Lro=sWzkPzu356xTyUZa*y#>f(69 zc3h}8!@Ir-pr?I_q5fNA;nPBWi%2t=5Zq^Tf_t&Clm+v=eXHTSOkat9-IzZO-@mja z(RX}rJ%m#^*(NT4(RTHnxStmphfQ8&oL4XF*V^K?3|;Z%OA*_rpUYcau7kb8m`M%n zrM4vc4))eZ3&mW)q3ql9!Wjyaf`W^ekI>bZFODrC)4s(p-(f6%+U~qHGei7mCI4aS zP!@M6;Ns|P5%niH$0-JvyqZjG!C<}I zSgf>Qt;!Vv zgi@YB1m9>?L3y&cdDE*QYnW4Kqo`ZbU<+8YykCooJ2T=oRF*AUoyznV3yuq6?<28%sOxvj}Zi7ZG+E`H1Q>%v&K}czH1(I*TWP; zw%esP^`q2@sdOSdR(8}F!sKy_G>~|#?2+(TSxKH%I%52*n(ZADB)TensH@g(m3~cl zR_P$FGN9r!RP`(G<#5})s@(Qm4SA?IXDoh7rMIsM&m_@(3qF%sW-@dGRww107mRt* zAn|NT$c{L{sP;PNeDf)fk_wQgLl+f^)Gy>4=ts-uXvfIC%~-_r$l=hX_j$^fH_hkgn3| zv4l*IKQU%YgYb`RNu-9*+M9SOVdy)=>>^|0aKtTVj>y;F?$DY;+_&xUP3u&U2N&z& zCMj-#YJGH`hu-7cDf#QD0rc^A*m}NatmCxS^Ace_zK+<{hOAkkqpZ+THdIlTRYZO| ztB4ykI1d|dv4;3t)$kHBR=#U2k!E<`79_g*NquyaPF8DT!o^y9s$QMJy@g%eB)uWy zdDDvCtf(~KRB@APA(L63No6!&nB*wkV9b?<$+fm5X2RlBk--Y-*T4DM(ztf&L~c!irNEL0rux-@vPj zx{*{)4P~_><9&+u4B5Mlb&(dbEjX6#^URaDOOrM8ob`EDaY`Eg-)YR9hSc-6BxLV= zn=J`B{Ik9EbNII&BW^f}qK_0^NiMJ+wX69Wvr&Dc$Lu0db4|FU!G)9bd`W zT0NGKviE6Y#x#5Sgdowi=l4Dow)zb25T!Lh5TJ;3bnPi4@J(Yu(;_gCp*u~<;STsw zI`NXEPW%mHUNoiuwJizRiND6U^*Y$K?l^C5ICV@uMXD=KH_AsKN^Zh94X*5$*S$Uy zte_&CMZijkZ>=(BNCRuREr~c-7Q>|+ak-ybkdR#T%PYGv3~o~}fCe%+vRxgqd0Vq3 z?K9ST+8%#n80N-eK#aSB-Eh!Vm(>Y*vM)AfOtYx_1c|O=ZqJw1cb)c{B6i)Xi3Ly) zuMz7FXSCI<(!T$_RLC3oR%O&puG2TM(rMOI7|^`2-qHekR|X|kNWN^VCYDLTNWtqV z*OyN`BJpUX7vd>{$I_&c-Zf@fgZc?u60+MqW=ld29&B*^96W3{#)j1u-HeMBda&r> zXyWn6tw6Dv$n<44#CI6$Gp!+xL{@bqO*p9VlA&wL#p2TWv{5^P6YZ zumyfM>XlbpbyVP@d>YmJ4K0An?243&APXbviCEHLGx$|w)uc6pJ8`VvmyPmlRM|p% z;E1kBxd>%bhNFa(?vEPtt69TGY)Qne!D1wlgAs4OTab`k?6(bV6BkIi5cgMgD<9i+ z`n6t?&U3$&>Rp$f2(v5x)fla010`V-vn#x$?23Of{#DJ){y~uF>fZEzNx3mLOE~8n z&i>=aW3xrqRy-c8UDcbVi>B`OCM;d3&SaQel|A3c<|-9sbw7~gehe6Mq(QOLmW1qU z%LIw8ox*nf-^{hd=%5B_`Bfcpnk@mfIodI9cN>eCvYTL|hbxZJVT?Ql;suWyatW)> z-Nrm=$nCTxks3Mc;cP7mwID=`beNpTmejjiVhe`rtg&2a;aW!}S^~Z(pcC(63HcQ- zGv-DE>QO=BGLCj5%FT+7zx2kngeR-z*-CYqBA>AuHR}A*Hh!C7dxf#0(!w@^xO#X~ zWY({W-M}75AWqkkI|E}^Ck>m=7_+KjI%`WpcE0m$7QK$oy|fGXe7hYk2}EEOH#~av zTD5_DDd1XDZpBG#Q&jV~67Nt|c-%vW;CdIaFN$)il?G*wHDcG__3?Mupnk$w#c2)d z(JlrRs3=T=4pbzPE1)u=BHDh%q68m8Ls=Ab^))2pLS%ReqyGDiMbZrKeYPaxhG*ek zOAnogGz$`vi#~Y`8_&#C)vZ$J#6$_FGUA?Iw^nR=Uo=*7TGK=Q-zb0Cs?T4zwi`&w z5&J`9jx@{pye)~oW47rSU4@8|=z!d;Pq;FtdoxDuJH`U0MQwA2E^FkadYDeY94fFS zWLSO6m?aIif3+pici3(?)T+asLZJi|wprZEHtpgb!-;osk+Kb@fk0`f3X$qn6~^}=gB+DnM2C4N|r-kQ6{ z%0Nz13=eSjN+uR;Av6Bl#5-w z=?bpu3aagAnLnk=FHTw~q!Mg2a-4j_2Sh9Io%BR1n#OO_i z_QS?%NDJ+48C+H(<-iD<<{A?Wlw%2znGYJXsR90Twj|jZ@f2{ldjkSk*+bJ*Cs@IWg7WY5* zI491dcwIeLAXoK#(VK}FXlXj!F1vQ1H;W6BJ#o9NnY3ND$@o__t6MKfTt<&>VAN@d zDxHoWZH=h3tv}}ezQQHxIKs}70gi(|CN1xFv)eh%a|2S>v!0akiBdN1J`TU zZPO^uFkC>~5g~9T0`EywtvW^*+QrN&;W^vSpBig6t^J_JSjZD!p%5C%*qY$V zdRgq{PmCGStmKbax#sSyZEDMniq=Q-a94wm%f3zZo!Mv>*k?$HSA}u10iij)u z4bzq6c)cdc8T+;|XBtHRW=leLp?_hNyE$XlUFh~l>(~my+T%p4?i?!CQ9ycbc%w79 z`Rfe#O@n<}&9rx9W>_NB5~ybk+gp&3T=4v0K|<<+=Wh!Ve!S($%Uc3&8#07y)q2x;1i2H{x^v8>@QQ~*ELA1g z;ufc)Q&s7iFxc{(G5E>7fdo$sw!9`O*z$VgU)2o%HG)Lfk){u^txXTYQ?A#_E)*=1 zv0NOmY8KpBq_ul99)EtuSoV|}Z*R_Ug?z!G$b|xjI4lJY!-#9*LD-)(W=VtN1GaG3 zj{Krxcq~cB$+?YKy@Bq{7>&;ui=G}0+Ruz}h+7}Pp^8S*yzpsbrZhZ0VM{`eg}=p+ zc5}pRfg2okCUD9RYiGBH^LrSvqK;O=bB5|4jYUce)f$8?nWt>XQv{;~XX%^9+-Mkm z!YAk+QRJUfrBe+==UleZw|Jeo=1`ymU5GQ2&wv8FoAb!r4MBgDE zJcC4!>1wsoNZxHsc+RkW#8{-Xu&v?DKH^E=UtBVlkjDFq#=K|<{emrtz9W>Cr|u9- zuGX5kvViYvr87U2t?FF}Z%c|U+}nolGO}MbRz+H5hn16hI2NnQBF^mzS;BjjBK}Q6 z`MW)!nFr#HbV zW!qf4wl|J9$(tB|dOLo~Hxs!THDLU!nx(B2BrM`jw(I7BeQv`)r|UYT<)~Y%KV}8* zG!`|v^J@x5vd@IzhjisIfrl;o;4oa&|J{NP#C|d3nZ|6%C#ruqBZi9_#JDYr}hkx%X^GHvJS} zqCqPP)toIEwP%dwOpDsJ@-|@ZQwSS`Y%2IAtT|_mnbPn(FGyU*L2czR_Z;oXCXHyZ zXRP`684HoyR_ZoAesry~FBxHAa2P4+&(d>rPRJ&IkIyn{EpL4Qa z&EcR&JvQ~s*O-y}Z^pu=Meau2TkmtL^cE@(=X0y}o22vJA2()AgYwsHNyMSF@EWD# z$^F-YgydqL>Ct-pT|bKJ5jlF6<7C(~nh(>Z4cpJZ8f!SM{cJ#pHi+yoZ<;rg5GeX5 zV}3NF`3FJbG7h&lKT>S8=t5T-a}g{-Hmm1tj{y^%4b#2C5N%l3CuhEWbLPC{T-1Lg ze=~-f>V7Ur+E-g^%%29?fGr8x!&V9sJx_(na6;;gJ1g>06@3Y$NnbIaG}Rl7?7hZf zr|o<<`)R1eH?1TTzC)aqz`@e71xX|L-NyWB$nLf!5l7Zy_?06#4+$0|Bp3Y}!G}i6 zQ&mLy3)2m{l*3VK(T1yk8c$+Ymi{$1wX!h|X-#cg#tkK)>M&WQ+AJ=|6Y&c%@<;3~ zA%I*oW>K@YNkQT=4*f5pxbXhbQ)iq|2?E9Fq#PalQ@+Evy~S85X>q%SZ|y>T{cOm) zz%M#!tHXcdojq?d=1_y}Sz8jad%lj5@3qH%Qx|-t`yRzMMSI5emyJbEi|dwdxF#=o z7bXq-A2#Mr!}No;B>Il&Cc0duC6boOSvy>;6jA*n?_G#O*}WMf_$6Zj(;~Q8T%{s7 z6qEH7I7CGa;!s(0ElfHV{zYS^G~oWwmPFry+k2XiWuBaE7Lc>DA15}bG-lm%^Hjd& zfZKG%#w+p(sb<@p4)`U~ViWieV^yU!f%`N5fzO(d6}1BL$axJn+f26fp; z);WCkg-NH2zGKX}W+&gWB@wq13%5*qc0A)-kdR#T%d@-wV5M2F)@FrG(nxdUz%J<~ zLo3K z!5=iJV2@L&&TvuvsNXH%6NmboV;%;ce9RNSBuWhINn-`39Y^lVO~@)_{M=mW1rThuL;|?fN;R6-M2!0T@$)$r&Vyh$Uh-+_*rFu2uWrnnhG*55L>y0xAzjW2Je*jNkX-a@ zR=5t~e$MejPKt;ri5Sv$x>P1MvdrC-&FQC%^_|w7uEQzzDrfo;rxZmJoYD?JWGSDJ zbMoWHTxo{%V}eB28KLKazLf`x4IJ)sD@}JQ9&Cv{W6l4Lv5aXozX}O?Dlf_~oLH&t zlg6BA+Wu`@60*bmrY#BCZ5i_(bXyA&k_*m3u^=IJvCMWCgSfy29F5DHXOufwX?W8U zD39%XID4E!_3AYCk56Mu17A}Bl9pP>+{&!@gGaD~C~|@;@SJ_sYJ=9(+ws2F6!YXm zE%72!?~u7lpOmV!-1t|uF0(|C=sG}kKPqRNs&pd?-%zzD zs`5lRjd9STD~fqr^C*@zRz&hBrm4S8?nssnOcK8|d*CKwVPsj&y$T;6KTnFi5T zTN1J--^?iY+PPf3&z(Y8XDW(W_$^~|%9#DM*epSk8UKDbw=7F?2u~PupkZ*-mPBd{ ztT)e*+$WR}-VzZ-Lx0S$%ot0Q78c6=6C4Ez-HM;JEGgXAG-gG^=Se}L>oC@HKWm?# zn>35^3Ql#VLX$_cLB+CA*t0-q(ELsGquzNBiQJQ7I*BaFqcHV?AOrhuV{N7dc3+SM zMN(%Wb#@|ie>@UoV$6uCuRL&x@mc2WLI! zk$9ce_OkWqtTHGhzF z8wH83gF^4uL3@v&%mEi|J5&@G8*j&1_8^^GgzHZ~*+N&#XT5qIDK-1CI&}`UDkwe~ z&q}{UT8!+Ajg^>o)YzLr#dDSwrKtE4XSt0pBK8ecRO}O?T-9VCs)~KmC~}`M;~M(+ z*piT=$enCJyys9Ct}&+?@=w^3=sWV;fho>^;uCP!m%JBxtM2tU>l(u zJ6w13MJm%Cz_o|X&G%PNH$4w!VVgMD*~p>UNH8hk#B%tm|14^kJ*CWM`>V!arZwBK z42sDJJxH_l4>by37dNj6HDG}&)DTWsIP4t0=|59PvV_!wj~WZ18TChON%Y;EwuLsa zFAT6Byx>mP%0Y4D!0Cw^i@|$r3jfPk1!+xTTY@S0yVHJkbEo~p(lDRM?<}-1r0rvakARSC|DyB<49}rm>;1*-vG)yzL zB>LWYS1W7J{aU3&?J1*m+*q8nXbohP1!A-=;P%ENt61FYmVfZeGe1^PbCzM948c$=5i;p`_SpwL;G5XvHnhHi`{r=;Z#< z=tb70pA9$SpEX7?*(6B`i5u}BPTGino$;?~=J#qr;xY!H?l?lpvbdTr*s9uJJx?X< zkx(i^3j3?LtI5BcCWytvQoHO(@YwOw#wtq*RC{*@_ml*ZiT$WtD4B>a8+F{e&^Q8p7|lB_X@^``Ac&9b{UgVvPz^^@|cIm_@w- z;wp5nmnX%8e%F{84Wv)ml1L4b^|sx{Gu0Wlo);zc9E^dqr7SAiy{_}B&c)79DT|@0 z-L)0t^-sp~rNs+(v@(C9%hhryQ4+7Xw>)Y0{~wGw(m?w=LE^Bj&%!KEKT z4hQuqI-(R^ymW8I;N5L3VA>wFIdX{yioQ8-d0lUzF;pu{sJ*yMA3B4lHZ)%}enG3wim_yT3%4{BIjn$aeJdPs4 zI@E(0bXgCukU$Uqv`J^~SPwil2YNtD-@A^uKUXIW(=RiYL$jzyZAru}%3^sWS6Dn* zSdfrh^vMld&rz=&mOV5SVUE$K_bzSN*4|>Q;Iy{3A;V!X@FoVKgsD9tS?5j0{Af1w ztSyPYqqYCAS49qR9oukNN#XL3y4U35MRem3%3h!2^d3q+ZoeqDfU|6bGX<)It*b8@ zv!g-vhqfeALuJ3@+UMem33w40?;PcK_8-RbrA2HLGOcsxXm$sGxITu? z0*f=U>N+`L8|pj89BH6^OOUvXORm*NDmgl)gOhDydW=G#J!RG3cul_+Rs#q!h5XQz z7*ToBTIxDuUNog|wk08Z$_7EA=W8iHYZwlBo_v=mIPr&!%>iR!(sqj=rI@(Up-HH2 zJ8H~|2F?ApB>E1`GQN>RhedtBkQ9xDNDIkw|IQ5^j)Y>hlg4alaO7-B#Nn`5C&;A+ z50VxnBo}>JY7Csfj8S)*Uan9nKG}*N71kfKfxO39t!WKpFt~AAa-=vO2|Min#h4Y% zNIJG8;_z4?ERBRWek@2xE?!|9LM1M+pJASK=d#yTV@rs8V@apHHF__ZF!o9pd zGR7y_+DN>_y}aK{+ROU`<6qSb>vMv{Wps|NJLF-iLlBxa2%y*Q>TS>6l)L%A7^@(q z;t`LYR5-=u)5K}Q9n>q6JgEO|%$A1C-`kRqo$YTK++Od&UAG?>qEySMmqw9ITq7

pOsuuAvDzjbSL=;FU>n>Gj5JX%MclCDC^XH=puyEo@mL zSz*pDI7d4MZ?Cb4Y5U(b$gA;ri~*-afLA8PfA<*kq#<{gEs4G(x6yZpaEY(TK}npM z+Fe^QY7@rdrA2L1;2aTubgnk!FZ7XCCe_CqH|9vg>~UKXsWGz-&TJ5Oe3so=II<{~ znDc7L`oPWVk?rtqjTo{w7)zHHvW=N(x`9-1msPn_5KD-BzSfu_4Y5}V5|`2WxP>!D zvgi4paa8Z(Ft`YEx;URC2azQ6Dr`x< z$xj(`r~&umwj^X%`!UA8*S^WR(NGcHE}u(o!sva@Sg^F{t3ha{ zPispz@J*CSqjq+>TvHv}gnf|j8uO)@)VBqRt^;?^W48B+>R3pwr45d0s);3Ls|J^d zY*lddu-?E@L|l7*p@?x3OOxazr%NNn$li2azg8^wWhPMv3|DK5;=?2itZAr*(yvde?98B2$zAp!p0|&f%@qCb|Oq*U* zXtIU^rotvEB}!aa;Ho2z^ty28ZQd(K9gyCTWYaxmto*ct(!mU_wil9%Ek3c`r${c6 z!Nhe3F5eX&_cyaFx#GPfYzLk&W?!@5qk=@&v8(sx6K)U|0r<~j#YVg7v_p#`L`e0j zx5}{;zWEw6icMoRq($*YMv=J{o~78YC2q5VTNR(AG3H5Q)-;4Fwj|^jbB;mowRgFD zAM(*e!K@;xE>@>dw%Mz1cP5K;dg_cN4F>+*#;QpR{O-(zFIyIw9)WBq zf&tNR^c{QmYG)2bib9 zH!Z|YW|H`gp^2*^Lw+M@CTw^9nlW!0njf%gh+v(N5&0EkE;X$G%$7tPYYR_LI*UBvTab`k@XFJIgye!Z z+AK&&E;*;rsXxe1R& z4;o9HJQitsZm-AY8#v?Wq~t6dc2*>Lj|Yv}(E!+IOG0+E7u%AM-JQ+glDfNdDO{kt z&s^T_e(%v@1)ZACnt2m-cf0c-);h?so}!YnqxCvn=t--eX*z%A-zb%M$$aUL!%^VX z#wtj*PEtkUDDY=VqrfYSe^oQWX9S6^qd@N$nJa=L()xlU`iQ&z`;A3Rso3_)@JO@Z zB<${2C3W}r8MC2j`n|R!WOsiTgVpQBV)h6QPt=>Z-~wg*Y2!wizrXc_8pPZ=YVb1k z3os1dWl%q5tctXtq8eTxOK?sL{0AFRfh^fj76tW$HQ|3V=2Jucd=#aP2N_7{wou7f?iXB@luq$C#{6h-{goimbvW$3 zQ*;{Vew;~ny4q@R=gVnMQDPHDVC4pxC`s<9@(Xe{8AaVI7%56s`Hq<>@xnZ9&p~{ZIdkodj z8S5r3RNFG;R`kfO*njjzpcF)GIL1oyT$Q{({?o=xY5@L(EeYA#-fv4nPP1%AJ(y-K zNJuVt5o|$1>SET`_arV51G#W{V<7h(EQ@oWA~W}7y*lAe6wAdXu5hCQ34Af}Y0BC| zX=$pRAa2kOs>MYdj?!O+L6HA#Y!As@hjft`1o>uC5ajQTe^u)Xe=A6s4`$ZW$>syq zX>3$;oC&c|U;V|n?>|R79<5g2*qa4_rtkekvF4;lxqpCh{Y+!k(8%i-Fx? zti+Uksl7MT2xUbb2~;N}S;Gtpk`*a1EUTEH0#R`ozTwFgAh1tDf7qCD4gK3}Nyy=N zhz+RMk+L-yD&Y_}(NW3eICV;SG#9&MF0>k)Y#zLosF+;sA;5v)3G2}WaS)O`dLImVT zjM>v*eUB}PzQejcyy{n}9(Pm3X>50C!{B|=ShTd@t;zi@!Q7y zXlVVWEs4IP^!s8#nHpQb-x{kc ztp&W8f|V>Uiq?nnhVGA)H|vX=72=Z32ys>4cPt^c@^xd@HDmcpTM~UYmf#{w*vtgt zF3~<6f5=d-cKWqK2};srfvDgn0!>0r_6lQGG-#IElIS}$n?_r3>t%K4@K9M4r0&fa zt~-ncOxyD|XWS@<;d~h#fD`JH>@a3Z18&%sL>xGaD3J^_@w(Q6gydq8?cgA};Bdz! zaq)HApo<^wn!UW?u4_&e^HW@c|ESw&RFzvHBi*C)uVFCjshGk18=jo)l>VJRzQZ5i z<&XaqAITkoB#j5Nu2^)%6{RcjA2!qW(z-Y4p~gWD%j>&9t2AT$<36Q9DhP@5o8!0~ z*Ts!lwSrHvDJWjXh{|s4fE7H%vxRFBtLBNd^ND&M+b9S%BEb@868K)u*ozHxeC_3$%%i9H|o~JvA6RANFQ(l=n(ITdig9!6CXVv|dvD|5)zDXQl z4LD9NyDCu)<17-@-!NuPL-p5eN&G5aWCisJPCq+{dIN>S&?lVJj~s}Ms6hr%ybbg9 z*T$ly1y0EzDo{`iyk#5b^J*Uq!z~i6uNm{Bq4gD85`9N&kQ1;`(JC3Kgy)RX@|&VA ze9xxB_8LF+8ayQ-6h=-YLQ9N!(GXf_OQP=x-9iY7L$~NgA_?<|*x#lz;^<$;XmeQ0 z4@>(JV|lx=@M(M8R)M8BIE#)Z7Sz184eqH2Vgz9Bm1!^xwn%`p#tdoz-)c*u?*MNF zgudet-pdt0jl|V5d?iXOiMc_k-(rX#HWoiE#6ybXBWfo3qbh@~@~>bV#z-W_r;It& zFg{^RqVE{*KGdp1FQ}gs?#ehrVOFx}-E=vMivy0c4eTS~Ixw#utK4BpgYkdLST$+! zzaulrvWX+MybqQQMH5{gqQM$iG*pCuXzG+324$oz%oua4*+SEnMBi=UnlKc;KfFK0 z&#{ksNpm%1RDaZ1$h4?ln-mleGm)fs8wO@1ct32+lm_0rZAtVUymdGybdHV%(bd$+ zE>m(72J92Yf~5s)eTHs?0B3wDM1pG@-n1&A`YmI2G^l>VmPFs7N*mv4&()A&`-ZWQ zX<=KCDdu;mqhWaGszm9pjelK3>1(zm`i|0Q(4kDK~hU)WWtmr9xaIkMQ`F6?F%nPUMZ)nb+hrIMt@EH~y|vzR5eB>HYK z+rt_~Cyt(WsNxjTe3kuoN8Nvq8i(1dYzMoH6_K{@XJyTyYPpdWwXzL_uQF{#D1(9u`$`@5hmG0PU_WI`qVKTZ9n1|H{9;l_i~pSwGSUv_Bf^Go z(O5lc4dKpk9*7zQ@s&+y5vz2Ic*>Yx%_3%ONyIHeZYRzk9%L@s%O_gpvez7^&4@glhxQpU6B`T1I`k&UM4BUbm9jU`U0ZmseFDWA3ll@qnUgCA!${@9oezmI=DB=H4H68g5_ z?+Z#@w*~FbLp)k=n+{e0$50gp{UNNn^WsQnXm9S`j-~kbf>h3;yX=KjJ@JeMO*?oAMI88;p6>z`NF##J>Y|*3>3J zqHDJ+eYVROTKfHN4|ZUTZLbe{lj}@Xt5om?6yWW%X?xrHu(!Ec;w`AT zs`NB@szJiF-0JUTqHR}-1W_Al_Wd8AGC)`!hT+Kn)a+bx|f$c z7#ZDQOLAJ2H{nFbh?MWJeO+L?A0lI2ghcx)2HWR5inq{S5xrKGT?gha`Y1ZxN8b|O z)vGOW4;jNhu=#-$uJ(f?4>ynZ z@oLj+jNlAf4SOa#hKse?%0#EVfOOk=s6Ke5KNR>XA^jJ&s6xSU5^`_@6N5?HCwFEV z0&gTl6W+?gm3wDa;2pc@R(9Ah-9mQPmZi>1(zmIziFBJ9dp>5aJyD->E5)bCvS}NS zrjw|JC$?a0e#uzgwAif6R71Y9v+-||Do{Ak1F&)braSFxRK-eN3f&jlcGJ#{54~>r8TCTC|D8e;ara?WLdGS zFiA)D>#rEIrrFh>2@;njU)7(@79idjo2>XNR`eYgvTe=%V`NhsqDYcitzg~JfS^x9&pAP=>&<9?z(eDgCr(%j;+T|F4 z^1UjaNe@bEO>0F8C6trhsbu`ic|l2rl@@fBrE-*+;d{b@%=N2yPTv(IdOnBjK2&Tr zX#$~gGJffG%I~6r3!MISfcysMcq@kl;yNV^P#%S9L~TnN4E6)Y+D!}gov7R`WWy3G ztkzB0U$no4a$?By%ad$4vm<1qGy{<%qS`~kB6_be$J&Up$CiYgBiJ5HI*HGPoT+$R zvmhb4m}jQ4T@7wOjr=pVOLA-u&^jMw56;w45P*#}@0{@v4bWX4vsc*$pEQOetqo=~ zjX;vn6~dls7H1cQBx3Ic&URCClEQREIS^Hh+0+d0oGppILw+ModHbrU2T^efs6a?C z6S_DF>e9+CI^@jTnt^?%vAAi0b;L|EA^6Su{Bk@e;8$@=!n8w)<@3gzX;{9^mPFsN z-0+|RAQ#7?ver=uyES6ue$808w8(A9MBt)a2*I3!SC|?o@%orCKN?=YVoRd$c-?!- z%eQjWnN|h&Wbz3DI@{>Ws8~jr>nJVbDr25C%lO{+tz|4dcmWsl;%N)#&2Y8k9 zMTD+JU5#kxv?0LQo-kHQT5NZQ*k*TRl}mbjp|Z8vW&zzXB9vkSVTPi#fuqK}YBq4h zmPFre;NH*%cJ4s|Qjsq{OY5iVOeNSW%~NHSa#h?f6WMu)tmrR^5?e*nSWRiI;$CSL zJF#XL$>$_y7Geflsu;BcF}rR8gxQSJJf1XWS~HJ|Es4IHhZCB|u0782YJJ)*&$V|j ze{%-<-Npi^1=^87@5061A;aNHTHP`alN2SI?=nvs~^k#z;ze$JuXvQ=RXGYuuIzh=yw2J6RcN%S37+3R)# zRyw@KMJv3}$%zcq^EYQezh*3OT0oUfx|?%Z1j87A1J6*Ikyi(KqY}-p7;~nf`DeBy z;%HigLS;0Sa|kU+NG?8Po1R5nU_v%km&JousHj!;S9dSWR4l;z;T(7^SHw!&dNMd^FR{WH2N759pG5%G}6gLVI zT`}l=BI&^Crsvf}DBNpMmAHsX+xlZx@O{RjCXWZ2g4=^A>B}4;til|H1Hj^>w6=SU zS<$q9r!5KD^>#6E-HP;D_mP49Rg_J*5XMJh3uXN=0trNPLQh(2_!201 zCv5TmA7d6ZnE&0DMBib)?P0O2AM^lhFAD$A@2$AZLR2?Ssybx!CWD>X+NUK%`?d@x z7YA|#ey-o68(gH>7ej$NAq&7UW>SOudRr2y;kNdDZ&EHYIUEel2Yk?nP9HuHt2C6N zJ!AWzvCL^ZvZG!|q5<5PLeW0yEGMK195m)k19YDt(RB&g^9;ZnPu8oDbXrt{lGm|> zlWBC!owqe3R5I2>dW7IM6Te*DO8vK-LcR%cot9n}mFBvVt4M zKoXLg28a^G5lsxxbx95b&hrn*WJbgct^$K#O%yuw*BRWOHdaVl zaCc-HkVs_oXCViO=2kO=PYDuTSA#td z?L2^@7{xM9Z4Na&>_Fr(FXmAUJji3G?!GP9AC30^(Y$+D8>lax9OM2E#+pov`-`b8 z1WAl+VG@{bzwzqfR<{@p{9V`R=z!%>d@5L#v}FA|W7ahz_#0aiay0lWTM}|z#HRTD zwJ!1xzlR0TD5Wtv-$Wm=HQl(47Dar<{pyECUlM&oe4{or5jm~d4PdxIMHv5VQOUia z`g#dDZ`TQyEslfBS|ginNk}W=b*+g$d|Q@Q#OY2JBqSH>Y`3e33*i3D<;{6pbIOB* zfiu0Z&0}JhIl6K$jVub1*q(?%v;$ADwYrNxcJs$w{Be)?NVYeUFrLJAFWQ&x!+#hO z{noa}jDOsx)U`(9O9aKPlUMIUnPF$AGgr}G9eu>z`xV9#r&PCABaxI(H4+!f8KQz8 zXEvTOX2b7Qt&qg5B?-Mo;(0--=YGY8Q{Hs3Laqc3Yg45N6y_`Dq8*{#tr4r~CyixH zi`GW0yP3Nz&IVDf2cNr=#4W7qAaVPEF+&<|KWazdfq#h?yFXts1WDYqJx(5J!|%x#@+HWjSTA&6qxj07(JYja3gy#WcRy26_ z+LGuyJR6STh6Zv-;GyS$5gc^AI#9Pp3{}BcxU^7hMDju1BZmZq)tNiL;*XsD!u&7^ zm}kt82H1oxiM|82O#mheDbQVm$PziorGEC%l^M&p_w4$N}i=gz(FmFr({YGOJHK5;MOCmMU*52oa;H+vxT%dx4vl0r; z#yrq&jTpmUGL|kah8x2ZsSQz$3-LkbPC+b84w8s{$e1Awv7Z+tx;m!4Us2x|;IJq8 z%5!H3R~v1kF_h{69jd#PbEpC6S3QY#!b>B?nEsKm3e#e`C&ZK@z~XNA$o=P8y7&l# zF~mg09}+W+5qDJBlNR`YV9d1!{O4>*$gcWXww+#QEro?2_Bi`g2aw%V6iA{Yoq3ek z4ufD#FIT7(QJz26oiB+J+sC(!)s)sgloApVZ=C&H5>pHXQp~I(X2d`eA_@Oy%(iAA z|6)s`?*?*I4k1q5AEr7`Qe}=**>;Cd`lA**Tr{P0QsgXqZ%MNW-EdpKR!I-ZAtvsH zQ>`ee2gP)Pf(ODERMZf$sDvQMwZ>9t7PZNiMBgoHpR}lW{faI#k;6F(QpM4!sTTy< zdiEPDE$sj_Dy@fa+>+HVy39+wSFOk#F`9(6_yfkwYeuuzmc->Wnxk}copNQ>B^!-3 zK{lF#vC=GzCVpL?mC+o!Ela{LjV$r&M!0wi~x7Y3=)u#=owq z`eRudFs2?D~4GKG91JfL7teX+Vu5b%3vg@T&NamJacx7OHuY9EiXv6KRvoTa#cR$~>*rz}ulMBH zMqFcsrnQkDfWIbU|H4g^8VRkXfm_MEd=}R~1R7EAErUj~tPyg=ptI2#Pg8PQI42r=}hPxe>?`&8FnQ|g)k`m2IsS4X+$c~q+(^}KWCS?9E`^N4cW+EecJUo{pxrKZ~hnHj+k>PfQ= zbo~DL`}j%z@jo~Iai7GWge3llB?&$E^GmjrmQS?GWv@A2#?r6Df-&^6%g=AWzF^-A z1^cKaJEcWgvGje#TwT8RTcu0Sco&*TqKmmDr2={<>*PAYmrUX>arZ6rM}w2jel~!KE1(~;!?7X@nX}P_Dy6h!_@0V*{-9(7+c4MkLO(!cgvwf7;cDMmHR(AQQTkYvM$Hv4SfXsb)d^G)K2atoXMXYaylL+Z%CgE8q^R zGRd9aNnU|pW***b%!5zj^&yEjT9VKw_BRMhUAu#AZt^XmlRMhL&In44HXF_%w_J7_ z-ZTnKV+FSJ_MN*T{iB`sC066F80#P{lv{Cs6W?o&&G&%&>?F9yNl@lKE4bI7{Uu|b zG-y9$OX3&sB5S5kn94FTD6SdL(1|}}PQGj`Oj?YT8Z*p|hQ%Kn|F(w37i>wS#=?5K zS)4S(w8L*%{r_Rie_Hh~$!*Dcc-?N#!y&MxrHsC_j>b1tdt7&zUEeUDO>n%wn9jsxHrK2~$CVhz{J3P=? zkk|(~7*BR$EJ#Q$CT$I#xPYU&=khwLgF)g#!=V_2v+It?EL?gU{7u$PK8E(&*&*%l z$2jTMsC5N?;beY>&^5{PfEgg;3FiPHyxK>p^UL_gdLmgxVZF{c{(-?SwWM_+no z^LrmY7rz#SyMm8PX6b8~?V3NYR!=-x0a)WZkQKPqygF9ok33U^C(7&P~;;txBvm>)cHl zr{|3YO@-6-C`qlbDW*h)O%1ZQ8MC25_GViWsUfqTD8!L?-XU}fIA@0hF^8^uqy##; zY%`WUH-B@+?xV&Mr^4}Gv-v&^oB-$#sD(eV34LEN)m!dFym+|_Nv652ZHB48_ah)t2E(rt?T}~B*Hll?f8hBqc=1T+b z4{b@v&i8r7zt=O^K;Vi66Gy}qYwM30tM3{MlnSfC%(U3)^5{gnm{@~=zI4^TZT!<3 z9RFrZA`XXzFE5t^+@mc>NG|4aYph4)^`~eGt;2UeLr37Nl+zpcZMw8!``CVaKOFb= z2GqY}#`xrMz?jx&w;BJwW+b-=5|?qn9XvUUn;k32CBT4%EVcNRfCz?6#I$By$=@6=M!GjLzATh+||i+{qD>yQBpP$%Q{+{w_ic?Zv2`-mH6G z=lRaI|F^trfs&&t(|JrLGn33CnFJD^O(r1|GM&jx-U%U@yf922NCF{ju;e%f?apcnX3O**Zu4N|L?!IZr#U!Z!8}zcbnlU6DB0)nM-g&G~3lU(Nzz<4vlE| z(FWz{X_!haS8@`JcpRCBR15A)`-VNK__?#97nigTOy!s_NH z=L^zF;TwkPx!N@#-iAR2~&;|U$I*9(Ibvz!*D7)R6b@kCX(^EktvT^ zt&+XGaLj7Tl+r`^3W=8oF&>6J>+#rZCN)|flkmTvp8RfvBHm?|#Ze51Neg}eKL%#C zmP(V2E!Bt_h0)Y7tX9uVkC2>?{uv7$cU$PVrDdU=&rbedW}>E80^l3@KH*Ass^67b z=1MIWsZN>C+4gnawgRboR(_JK{1hF2Mu%tU@GKpk!{KDU`IOu8PyMA>ex=K7tHv?g z#6Fm4G7cv*(erK-J^z?lPbQK-e&|w`N*5hyzCp4R|$+pYHeC7XmJ=t!I82MNd zlj>=D31*kjo42t-*E8OgWUYOF##%EKYl$m?Tgyl5$I*ZiAHF_VR2xuTe!rc01j=0Z z>P{~j*vFHK)`py;CA6-M)>m*ZUQcL;T$s%r#+A(r)rqC%XDeeVkg^bif%ncL6^HLJZJ*K0pjY0HlY7O&!m?HRkVh~;q7 zI6|F;2a|aWYE$`h_cV=n6MsZ~P+RS~A5p5YF;PA%rK27|AzfHVC+39Ftfg7?W-Z;R zEs%o49e!C<3uBp~Jl)^Do%XWCanY)K2SfF+wpw1Hn$=~0OpSw6otZ^n)|N)W=^y;E z2pp$5I}mENgt%#+n=3~du@yxfJW<~N3?Pg@nQC8`bZ&{E1Uujq0X zqe@(cYHMMMwj2s7i~X_)9F-<}iK82k>Se~_Ds2V4V$sYCv#2}IQ8+5W>Z&m8(H230 z;c~w$0*7JV79$!*{~X>z@JZToQG^%|br;bQiW=^)vB(XOmkbQu_1a2%h3TEXp5#mTJp;xa7Zibwp>%uD_0quyR^mk3Ql!3m!nu!!qitW z_@K4~3I=!jWf3?A9aq3XY07Wi_$!wBGo$i|wnAP}nSrraAyK#@kprG)L)x|q&%@fH zDDZsQFN?t8X|5kb!*X0_Ku&0@;1!S-yngIB9u6d0R3x6&mO(+{=YClPj>Oq^K`E4m zhcUk#P3rnf$*hy!$!N7N3#wgxPNn!by^XoVThT zb(OXjibw77%OY@((wr0OYIUj3P|yIWf!uh`X@V(NgWvY~mk8GDzuC<;6u^vfb}czU*uU& z>>t$=LhPVyW*C?A2K5IT*b%vA`Gj6phV?ORLwJR?2a67+CKQR5CdLU(ut_p2f>>l# zMI4BCs0H#7Z7CJRANI>4aKzW{#=vL35@-(n&tiFAA(k{~%hr0U^FkDk?#kqDchNl+ zM(hnQXdB6^H>~Z-l9rXQbJz}z=Jq4JrO;9gM%2zUr`U9*WY*c}o1yx~32m7b-+0z9 zi@<$jz`t+wz@sw7nyu_l13&hZnahLvEZgYcQ{Y{sv7;`TsrpO1w&aSxO!vzoaDS<@ zywTNctMfbKxk_6luO+|U3P=Y+ohm}hwPjHd>hsGYaD?0ofnDIZe_~t?Xe;6smsW(6 z!kajbaEUZw*#XHc6_WkhVknSY<(Ea^kZc{;-n(;$c+nQHVi+y0C?d^b*r5;_&6d#J zJBme+hmBAIE1=?`Zh9dfN#F$>7gLr-Lun(L*-$g2a+-Lu6Wpk6It2v!$w%>E^Z^$!|^OdjU5e>>Q2|0t-8()+QKWYbG=^{fxFJ! zedGw}g|Qdy*&kIK+)cBS+{X~#qphM>i05_*#Lz!d+rDxFcZLew-P$54z}@ASMexAw z9nE2U6w}{ok*fi2{QDTVfg?9(YoUO; z8T%iE5Tl!`X@YTw8@TarW87ZUR?sVMbGr;%!mur&05Rn(aC21HUeFdvf$fA}7JMH+%Jp3F`jKVbfTeQ+IK6RjT@<&M$S0* zFn$qj)x6?2yVAgkhDo-tkj)9!*(z8Ew8c@t+V7V|;9xDR3l{Bcg=zkwIM%afj8L*z z1E1$y4Cvdm)%ObMg>?a?&95*zA7=5xN$eJ52dSGic#ewjjoR`l2;bnBMc@cuuqS)4 zcmyk)#aC;_FfbTkhba(m)S9`=9`0vU@6}e>E2{Il3Q}hA8Obt-Wh4VBvjfp{RYdR6 zmQ6wQZoe!7N3>2X30LP(o!=R;$F)`RidemH6ApyVQxQ6@EsKKCV}4l#j?lPKC0sa- zdl#eilD2wY(HcMegafnlRm@)0mPovHm~RjZC<<>!y+8i1v8^B zE2WZAcdOJ(21atu%AmeD&PUHI`r%TT!aW?ukXe`|4r5ldDI|6P8d3qBtu2@W=uE#X z0ta;dxPbPAb|HF3G-MQGZb-dlVNlm=EAF*;hQw$ zz){@@=yIbWdq-Q*sM(&*W%INz5PIkkzGXyQJdDL_D7aXkMGG8;5sWZ)D2h3zZp2TW z9ecx&w!yr5!$22s&y9v{jYX3vX=_aLxuLPR$Qq*)o;5~1L#eUDUFNCo644f2ahC&r zSp@Db9qcYN5EISkqoXzOt3NZGw`eQm70wwFP8vp$5;@@MQsH@~lsT5iQGynz;5@D^i2}}X!Qxd3 zyU=a+A!>xRU=Qke?B`62^P+j@4)vnX3N!xbrRFXy9`i6hFR@BDA~={M6qq#~)2t35 z>c5ELS+=9836b1CxjtpwKk)-z@oVZ#fyF|UESvB1-~q+wa*E5 zsM8(KV*lQ32&<-10x*n6o$PBvnNp10Gc=UUh*fq*#>i8Q(wVjx1AC@ggj+a?MRk%e$0 zTPRF!CoCjdkhOO^-+BuM>qO}2*49a>S{L|b5xD1d?#>TKGs)|yq9qR`!@7N@?j4Nw z<=Sd_twCmWWh-(C)P-EmSKwfspd353dbLiQ*x5`DGC}HtY8#MngNeseDDHMTH5NhoYg~c{CER9cu`UMa((a^QIC8!?;Co zFq%%LMr-C{o+2;C{kXQNyyAW_hCD~>v>u@Ws`B;Kp^3i0v=G)d|9>)AdZPmSEz5wgbE1wCr>|uxK zVIjJs0_ax3eL-711>6&USp*Jlw`kC@?Jd%OY?X zx-X^ANi?lx=alG?;y0?if_U!+_FJeaY@V|*j1g@Ec!hCM*Dz~fzeiaE@uHOdGPVW| zm=~)sAJ7(2fqB1Q7JKq7Y*0 z^xRY8#Ln$09<>=PEXJw@tSZ!&L>>K`|1ORf0F4Srb! z?jPGv!9RL1EfgLUH;U0@ZS>63$d`TRUTqV4^_}gf;ydW@&C?`BHi4s}39F8DkG2Ym zBi-$nMc|HfiF`3|SKBGpHK)pqUFLCZV|jI%4feZ7XnSfk&Jxv8j%y39ILc#wS^R%= zl$W%P<<(Idx{~oeXKTM#Mf*kVrz>c`;Fm?g_WfmeY`O{ z_a}l*M#gr%wh_FRzr|fNq$acwVP>I4c)VdXXEE(l!CtE^qyqLTzbpa=d*Qgn^kj`Z z&$$@pA#L@&!n|<&B0O#btzU&WqAj2T@d3Xq0*5$^{a1`Uw!NmnffS;XhSRGxpBNo= z)O@ai5zV{|_bu9{@Cx@5Os;k5B0l=7x}2mIU#6n|Hf=c-)Nl05;x7;N`?XEs6?JM$ z^*bx+^UE9tu|?!^H8NVdpGqV+4g;%tESZTWRFulKX9K6L`h& zBCHBS*bx~ZPM!Dc)WOmATA?ESxVDT6(#QR>2pnn6`u5c8>i^U>g;&I>)Gl+D+LbDn zFKItr!SY4FECR=}YxfasF@yEfVt0o61KI8dw5J?gQqNu+)K}5`uI!NvBCN&5PAB7x zN;`2~rQ$kATRH{T*?w6Bj_aatxtNb4+6B7Y3z%|U%DK_i$j2CO z)HZ?FvUyQgROk>kNCz~cqN&jWA_viICR>LvbO=KSM{T@XMS8uqj0)0g{jvxUX*rL( zVJ|*@-6--U2FMVSn{D&mv#=WvU`E0l&|;ChS&RAC_C;rvZBJNE`?WwTo-8I?n=KgJ z=ZVsjK0IGel$PR&$2W%#mNFax&MATAxbD!uHo}N2#rl8zuh4bCFPIag6__ajUoxbIl#cS?P zt?Q@(dg|}K*@BPGX1=lkQO}(wbu5(bB*oKnC(aay9Souqhr?!=jba;H^JcUrHy(EPTf>g!{wLClCJW`kc0Dync19?6RlCA*>pL9}3+l zOsR+fEyhOUq+4Zl6n(R#o8!Z5gX#^R(Uw>F?Ej=+7J++16Mgkl2bde+>Sf04U$hnQ ziW#=Ir7vZYzyZTX6^0*bi=e>pJ+CZoFD{)R{;n;XcQ#I#aBW%g3LKA~J-L)DpR|&5 z3&k2m#EHFjXB&-Y?#2Ftb=G+5Wo2Cd511H_`tFO~f)_ zC-ws>_J7b8Q^Edsepv(#dGpp0BOgm*@wdASI<7NV^DhpnAG`&t1#B4{rErsq#0A=mqp-6^xH^;`mpz)!L!IqxzOQgKA9~QWZajKkvW{ic26Qqw;M7)3>e$(+D7qO z5c@0GhWn_Wo2JI2uRAQ2l3`=8?%tu4!v{92KCn$&R>cQ4`(+Wh53H;BK>u3AWH>}4 z9N0AiAsR5kMGM%);0K0qQ`x*2{cuJhadkJJ5-0YGjJBD)dd0e`SM(#W0;x%xW8}j$ zAVzA6@sWy}2 z`u>qVdfWP7QI(=3SorAgBPmKlC9A!djYSYuFM=@i2!~QAA1!yA>4pgt2G27ybedsi z>C|TC=+tb!iB2tMo=%-+fleJ}kxuPqiBE^=G|496V3 zVIHMZqj>|JCYf*M(_8rBS~^WhJOCE!h`g+1wUNvPBL=aDXa+Pywp!=|3~}_a*Ds|L z^9Cp0R5nBBBge$Si8q(cO_Cg(cuUzFwiA<(uRmvXE^zCmL|Q z$mA4|?_;tNV2k`y<+%n1^uXTIRmAg;gOPEh8za)N~X8zxD}Z!kF(`H&Yz6r(Cfh*1%w+q`4%Pdm>J3o+e!>D$;@0mlNvi)BlceA|7l<23 zrvWBO@`Fs0xnXB>DH5B*}lzBuRcg8M2Fqk65{XCN-?h{OsxPO$SuzT-C;%<_CFOwwuD3c`nhnOVUzse-Z{){9AkCw%P zN7Eo-*J~%p_4nEda{W;|L9YKHI|0|P>akqv(Fi>z_ke4o(4(74(qku+q(`1f68#P) zA$rN`l-?^Ror>-h$#yE(?z|9erwX=tCW&oXl0u(HI6==i!6ZGSeWA!m=2^}pnP)$f zWS;9KDR|t^39{AWk`&ipvJ+&h`Q3sgdBAoi$pbP>k_WtpN%DX%GD#lrQzprG|0GGl zdj$bIbl%%+6hn=9C zj&g#e`miL$O<%JUbklQof_$@WwqQGDkbH9kCrFd1Bt`a{?F4220w)OTA4`&Msw%un zD%=8vFF6D3TZF(XB}uwel7i)xc7j-5XD5i|CpZCxN>)oORVv_p zh0g&g{B#JvpI(O3N%*RAYvk{(_`Tt5{N5^*Tf`(Ow~I+quD~Qo{az+X>W3sL9&Gzefsg%AA(odd>hqVdm z7f4bt9kdg~^e87tuMbO7-1Id&K{q{TC+Nv-O@i$tQ7D$#397KKwiDD=Icg`Ut@2Sj zfwqeIsGVp=;iCN zWs+)0lS!%}Z)K8d$iHI}m7@7$CObg>H!P+F|OyrW@LYsXxLb z8T9K+l5Wp4NxF4R7x$8G{gPxK-~=f;#w00vpCrXCPjG^?{-Y$t^?9v=CBI$waDwbn zlBBrl9!`*Pjx!14IP|gGGab-pej6Be2z|CQN%~}%Bz)x)7k|W z(sU`4q}-KEl5$5RDY$&VP7s%`*a`Bu|KNlu$QAS3?U@-6cd!Ewnjyr^F-hXy&LoNZ z5R)X-^F+=>rdE2t%@mpXnWRj6B}pRTJEhWn$k9w?>FPxG@)Bh46xknSlCu9Rll08V zz2XKu6RmSs_F2gOw|)3^mdO5HCMo-GnWXIJ_KO=xqF#N=|2Pcyj!F&$kVz)jgZv1Smw@~_lN&%j&*Vms9T$si10Yv0xe4SoOl}7GW+u0Q{1}s4QF`R3 z_rc_F_0!wxz}I*&Zy1qe2Cw$G=t4+5a4otK?p=ymw4b2E19W(V4v*5|Njf}5ho8~m z89F>mhv(=p5k=3OOoy}Sa4sEI&|wuF*3e-c9oEy~N;>Sr!6;ZA2a|}06e(sSd1I)Q zOvR~^5K`QV9)SmL6S7E>B$6b_B}tM_k|d)fNm5CYypklTB}uYNk~EMc=^;tdMv|nH zBuO(#l75mTEhS01N|H2|Bu@SN1Um>^3guhN)(i&Bh&!*6O{|u9 zHjQcaaSG^49VruI;xsLv%@!kt(ezLj8*pxEKp5`EJ)C+^Sq90b@X)&uB|ABUE|1}X zAp~J7SJRt}RQ3qcQ<4}FFM^mBH)1KQTNfK=@3N*MPIU&0*vq@bOzT}%d$Mo^cI2ww zbIfXvV1jcho;Nb(Ys*$^B$~-)Hdn8dhODXCArre>P^XW%2~rMD;1_8I+glh#p~_S= z$tdZ(r74!A{vzz)O)kP2(V&nQ=~k=NPQjZF(v^}mIhM1Z_z<#*?I=U{*~?2nyGVGhLQw0gflwz97bb?#gXE4?=qvHkC{U_ zVth0%PM9Zn2`ukWcnXDh9&xHNwYEDCxv4u3C+g0_iMsP}qV7DLs5=iQ>dwQ7y7O?N z?mV2RI}a!7&clhi^Khc>Je;UI4=3u*!-=}{aH8%!oTxhwC+g0_iMsP}qVBwF%Tzyz z?mP^8m#j%9nYY_K2=!q2saR)TqQGWHtB^LNqs4^trq)EX5V0pHzw2@6WWal<;H|=V?wntq1l*FY)t4iCe#`eS{)Ng9TPeo6Do}* z9ylfx8Y@-Ya8hp6A+h;>p?2g9tZC`d zLXj?@p#lj+3QO{_eZmrY)s?NbopDsbNN?tQ>vLNItjAn|s z(R^|;u1+iEQ;`B3DUBNwS6S@|BT7r{@D`0FvKTknWzDcJ6;h?)2rVPR#pXmYor)m3 zOtftNk{ER!L|Y((tt7avKo~r{H^)%dl!h?iQUuUcMVT+qy=}M{!6vr|Rs+w*)fee% z2cnM03z3|Wk0987sc4i3AH&Q0jLZ;BDO!zj5JoJLgQ}vzG6z2igQCeIekqg=9!!ov z6KdqVH*x1}Rwr&D)B$NBawHiaHegI(B$yg8A(rhUJf=l;=9Dx7=@{ip2Y&Re*=zQG z^Mo}uU&0Q|bZO(jS#3AH{Yz7-Yiv{cN(-3p~xOzaj4{>@^QUnG@h1yHD?#&ytX;-1Q(szXVdl+Sm-q`eF?Am~`MgniYczd@wpHdrledaywJlR{r5A+URAmJrCd}Qy5!o~^YI6z zTys{;_~}Jof6J>X*D@&=)y^)tuD)t`rj%>SIiWkAdtyR;xvmMszg9@G=xyN=YyaNM zeksLjeeluDTYtRq)f8)`6pP+oZn5rayN|`%@T(iPoVB3&RkhYCDHgp=U1D9+)polS z>kn7I>*yh)*16CYq<617)6Mt4A~XG_XUme#Qmrn?L~m+WrUT!d^+SovoK+u~@Xv>9 zGs$9q;555SrKet+fZ7f&!LD8LzPCtiUcO<|HNT403I=WHgTa;Q4R<`hO=>fH!t}#O z_ie24GGwCm2r@Zq>q}eG_n+&itvCFk`MjI{*zmEF)z;udO_~}`zyAFb56^RO+Bp3; zZ+L3uv*(c8!Vw+KwQ?>u=S3Yp*a&rr%Ha7IO{;2ndZ!otaWp1Ox zr8pcjUx%Y*oM!xaW`T2TRyqQ;4BbY3=3Va_Yg_L$>c;%bd*1od&ez(gb{$NuSTgEP z7z##x=3Vc2Z3m^O0;60P(Z(MzrVI6g@5qexS?&4`^qCk{R%@PJwP zJfpTJTK}SvTBKIE=x-B|wfIXYQu&&j?Yx@I97ZQiHcy=d*4fTq=yNHK&j82oGqJlL zzWO7v0{*4tj#=jnWwVDOSly5|V8H?=S7M_^$L}-fJ0(wFPX)Bb@g0#XC+;l`E%_gW CM8KZ_ literal 167952 zcmd?S3z!^7buVsN@6~JdmMlNo8e3Y~nq674{E+-wX(icSy&}mr2FXM3Oz+N&W@pCJ zGg_^U36B7FGT{-1BqSl^4Iv@$4TYtn|1fV$jJ08KN+qyAduDNR(%$)2koxuv=^f%6He!a_VRqK zV)OgO?XC98y4@`1nzbCzaXZEB?bQvZUUN&$T4R=fi%fs|f$Tajr6Ac@pfb^$r7mPz>Q?=Su#mU)KyE5CvFXl=G;%+4b+OTs~ z`x3<4Y2SakHsds$f;BN~-Dg|HhBJBpcAsBQ3PRwJ%k~EzWwBU?0Kt7D_5&TC+++=6 zZxH>#Vr|;V)u9}@QWav!cb;!=Y`IP@%&Y(mqTE0aFg%+40t1Qf*V6YT*KIkjhi>PE zR(q4s7q`}Goz9CR+{E3E!1A z1Ha496oO?6BSFIPL2JI;hwfj=i8e| z;5nyiPe4D3v#Kk`K6P;g23M_`n{+@1CZtfSRP4qyMi@phVP2Y?1!8V%!Uf$H*$l|7 zX}cFICAW#54hq^}I_E-0(_n?DVpNnWO^A3AsW!&4&~P;0(;tgmh?l76z!*UhF!}mYC^cFs}E^eX}+NBjklEa81`_L(FYH#rfs4{bk z`-^uK*PSe`FK#GqEN&`pE^aCAEp9LFd%k$*3&nei_ZANn4;Jq$-d}v6_+as&;%M>c z3&mq-@G$=WNb%9)iRX(ai>Hc@6(29Y^o8QfUMQX~UMM~_eq%9*1}`t##ff6RSZFW3 z>@Z5j$>)nx#bSG7SYkQXslX2)i}HzKez4mbwhPm>YKc^TxX`H83$>XlzHQo5xgut0 zSY4>6RVZB=hLx=FpP}d3d7gXvEeWu`hbizkbF|3OFN>cKzZQIn(c&5D0_^y73CaoW zzn09`vK-BkC0O3K;*Y1!XP@1-`?*^>)5>J6Rt3LYb#>#@Sa zmnJRzba{fG(zqp6rhzAnM|-tXhhKoHA~y*R#N4o2*ultGtHf93P@`2mCHd2luwjIo zFly};ry8&nPGPK+Z=zSb&cGfm8g9T^wFfa5H5)DA=@r=1SvWjhfT^QFk)=VggbjYD z`1V$N6;E=x$r@Qi;fXDE`FCI{&+@;lThXBm~fkSu{F7 zHfo*X&$XA0)$%RjJBxP3_rf&e>!s&v`E&4+`RyV8a@kQp@CeMVPz18&g%;R!{}IfI1H}SdH&J=!2e!&;rV7~qWA$g>f)Dv zr9J32?I!gdPMMfoeVQvV-`JI@(3ej8&>81a$(gZglh)bbHERaGoOQ-=?P>T55aBr| zU#ym%hW8>Ih2yZ)cEcL2OgV7VJVu^Z7+IY!+I5VDJ3_yRzWQJ9P5ydc^4E7pzuxN6 zCO3P!^b9%->k;X#|1aCciruvVHTjXmMOw<@qiF_eX}&ikA9goxue2jNtadkm05 z51jUvG0fS*3$ZJu<}5l~z^Fuj=ly|uLv-*1YkUm%boo98ZqPo4Zt#5=UZ=fr%yH_0 z!|978!lZNZ4V{VhvS7p%-v?DMufRg~LWds(V}~EDk&i;7h@Y*O&(vYUO*YFW(E-Dc zx5>x8hWR4k1~;h^n;43ya zxLa-Dt2j3}s5bD`lp8!K8>kA)O%BT@KAXNKa)V>CL7+>%PI8O!gcea~P9?O6Vskd3 zMHHMTf)?-=a}6xWI>qSz#|;$$5`n;gN{^{R(zxa(2V9 zJFeSa!|nloRm@dl0fC<%KtP}fqY09o;=Nr~ctuPuc6-Ssy8_$K6EuFh3`2=I`LrO= zX)i<+iH0J%kmWk$gMY&#;&$kqzeL1p0m~Km+0+)UX1VFLU5a1~nMkZBg7=p;;3{Id z7XEH=vsSQY!{00y-{dBzo1K&Xt8jDuf5LI9zbw9`_%{FFSR{zx09HqEZD_ICUfya>X79P zYh^PzE_{OR)wG5s-F7LcP!#Lpr{b%?=~A@>_j1bFHv%n1uyPh3BbCy`2&@0T;d@8O zBg~=~7ZKxeM?9yH1}(?+us-yEsMCJ=mH$9}wt&zJH)YUAN*T+%KtP)Sr3kOl8U#x$SP6&1OBXr;odLM*IS1>L+Ehqa#J3o~(QnrY z4S*Se#G5%R(x<}uvAPcZun{5TxD%E?^;wDJ$|{1!h^TTT_2fyi5QF?3hL0aFd9EWK ztem#16mN4yv{9y_Rfn&UqTL^{u{Sd4+h$KNK2F z15E_XBND_lss`E(0|F$}}H(APPeUB33h zr^g(L#fESbFCfFVeYtPvgsr@`Y{T|16QO!qnNZwmA)vwXl5AebX$9h0B^H_#T@!}0 zTqDCySgaHeeDi+np=DzG;hOd$`kpLedqA|{m2LT}cu)L?o~)BkLecc#vG<~gYKk8W zo~#frdZ9~9Nko+hGM9cIp`lK-6FlU#GPuIh84Iya=sj;4E$6ZOgSXa*pF)(P+AK|$ zC``SyRlQh61Vd70b~t+1DSs#2#bx6476RG)y+3(!-}1Cc(b*M9nZKPvb0_qz$9TgD z7Wu&o%pGF)7_MFbz<*Z!zG6G*1xwnGqQ;wgazkXn8zKvRB5m#(Jweb+Dl!Y2LJ^q! zp?N<}efImtOu-K2U@tJpLqlnU5XUt98}TSNQ!3Dql<#t1#RP+hD8&UzRV+~P6F$b= zLPXnXV*yP0PX+~X?D*sxC9yXX!-Ai~6@!^WEU(sxJzp!{pJ^aaRh^<)5)Q%ug5#xU z@QPgtSfKD)U|-jW*ox3@f0FW_Q3x*hHv(|}zp>!}A1Q}UHkvuagfWYU9`YU(?@R^L z0mbb6*C}Wh{QLWhZw&tZrsA9ZfBQc{3;w*%*HHVymCtSdq!x;jjv%s3ztd^ItY`FP zyj4>Ft}CbRMQ$a{$FH2GFA)m*O2SX}jIbX*mbA~sX>0Gxk^+uEFT}o+hhE4n5&DT4 z0K0CU;}s~@M)3@bm1eG0hnwTT=M=uV@9J}~(S-@aZXL}!u~*@Nibybac7zKMZn0hv zXa53W;6jNuW}XiS@MCge>ki_HunP#DwKp&T!~QhpZ8feZt)%T5?BX?YwF$AL5mCH= z9*^A+E%{T}WyKJck8)z0seSEle@OO+BduXe?D~f~ z$iFJAA3vim%I6f%59qKk=+|L^5pkmZXIeN#c_35oM|lHl;#mhGu0CVBf(N;2a2R_lH-_Gu+w!IJbH$O;x_eB$z@lL_N{IXJ+4(NvtpI$)1Vmv ziqYO0YSeJ1F`IfV=wbK=s9g>|esebGzq=#!E`3L9bQu1YSk_^;(r&=H;tk>(LQUa0 zV6lnq!C)hRIF6-F=3;%0iW2;fM_{(8veFcu@(8$Y%Z@c}$r*3)nxuIRJ zc-i?2=g(glnc8jbJb(U{9Xod&v_?j(TXtB}4py`x^gL6` zZt(;bQPxZm8^t!F)b&}5H^nS$Fy|}q+?~R(^+c&su_hd=j)(?mP(Zti$rS&vo7RjS zv?Q9)vsOj4sajJFdqT7(6%k=*uM<|i@)YLg=SKV-RKB}>s+o|_gj;8#dUUDq z;2A3&Kat;(V2+ToMH`cr(XabTb{IMaO53;Zwn(eSYWXqRToPXqr~=hXXd~7)^@G9R z0-b~W95NiMSu@x(p#60V9k%Fq&^!4o@DT zhfNhb8r~?_`P9%(>-I!A@%7GKL&LZ2+PO=Rea^x@s^g89dFVw(8X+3jm0b#T?lmOu zHa$%wVUk|F?1Ds(dw#UmLPCb8Wb)xutr-G9Xkg@_6nN+v{~}>^tkz^@4E-jl#E@*_ zk0Qbt<6zjOS!!tK`RdMHVMz-e0~Mk=EdCf`HDpPW@CYPs_KH_nVaEASkJT_ydMny} zra4RjhavhYNuAn8x-?vGxy7N->(REvO>{W!*b)A*I3Glx&Bbv7iK&b3yln4~zlWRwk$B?{ zg-4+GrVm2oP2vW4NnN^G{uMOzI!(WFbAkzsi*p`Pgs8==BTy2(pxD#UF6;jL@3(gD z)TJ)C#w+((~%|Wa=FqN=AlVbGY+p1N@_8Q69!a@%TDt3k66Qq0* z2g7REMXK1)CF}kNEc^$o(lihmH=!v8qbDqtGIF1Iobia3?4A*>3uaU}+$9KO%7S?# zp^(9}FV#u~i}Gx;jaF5dNiqLocBOS{z-Q=(k@V^5)b{h$?Q^gfe#gc1?F{YNZDsFZ z?PtyPyoyjucmthQ;AXPR%KZ!lu@{RVw)B-bags76;8no~a}fZOjw3HBI829}E8i(? zUps6`5}tF+iIzc(M>@5hl#^HO8D>NH2H_306>?L&xvyZc7~RCap0lFbv7Xe{OD??#0~ zZ9bc;{U7wAEffj7_WTecXyy z*qI||&pyxzSlB4v_V%K%wLOhB?D5f)#~vF!hF6atf2dNToL>1_u~583dRm?$=@dx> zB!dCo_7jip3;{^RBJdIEKHTUlWfg4uIe~0)^r& zBuPVybJ!ge>6=(cB4-eOB;{0eUh>1}zA$=Z>u;r;79@6HW#G54u*n3|DV75q8OVmR z<+6gdb!$MuI*%TETSG594ppk|K87@oix-}4k=HEJupL*<;-TixwyKBp2q+o2hF8` z`*Y9I^iO|1Cx7TXa#<0!r+mg|1KK1NBv|^}iT9JP?~Xo&0in=N@TH`9D{@x~S0K!e zD*)WY0EQwxe#Ds`7vE51)$&w;W2e$!3lf7&f}eRYsIYQi@gkssa(L}wuf-B5wpX-` z3^_mW!{5Qe$J;V$huVmgT1*1Sek(Pg>J-58em9^)`wsrATP691L+Mk^T+y*%=Ytp5 zhFieNLEvJ1fgl-r;wRwpkP1DRsWmPVvGK+6S15U&@_t#CTjZ~J52PI0*FJS~#}@Bj zFPk+m$S`#wmk}{&PX9gjP~$hn_Kj$}9BSx==wRlNPq2hHRIvlXc>xxLP&{alFKBwy zSJ4}Cu=8lXj=r3>FCq-Zrh~TX@h_Y;@W%UkO6jgRVpa{}hUKyFG`bcBp;Vm|Cl(?( z4n9v}e+WA|%NFA7%T1@?I-#cyY(cOYz-sMfHmo{Paw)kJ0nl2@?f4y7=l_gqK?nUu zA+sdJw8?*vXaNJprC8d_{5^Kotc^Z_D!PXplTrm{ujR>Ri~lg+s3FTU*p&BiC;G_j zX33JlLO^usZTDwix&8K+%SmgMXyrMTXt4;E>#}^jUOc9_YT$yAcjixd#Gv>M>~bpJ zncryOGQ~^bEjGg&o@??r!;!;*0I}l&)ep_2>u6&FCa=rmLCX542=ql~7NelR+mlNX z^_{BP)JyTN+9Y~|oqU`vSxR0)1+mulatgu0m4}A4m(j*C1;!|ynzutJEY(1I%gQFs zj@X5q$QXh5;&)|r-plqtcXF5Zg|@G@q43&BX>HEe>m2k|H|70bSS#9e z6U+vBI=5)h0N!V73cS}YVrTA!;&z&g*VCK)KX9b?{~$;yLmHTGgjpkTEtr3%Mp{8ZTia$V6yw#3;eAbt-i@Q`Lf(QqmDavAsNa zC^muD2Y>o);hLQ+pT=+%PK111a4u047!xB7Jvh}YhG)=hi)`loqQCLnUT}wOwpun5 zVf)12xi~?wVoI6}3|dx_7?dQ?kWQ2&5}|+t%xLuaX8A5BXZWW2g`gu_WfSDDAST|J z&Q;(CU}BNK6Ox$=|BDAf`5WW|`f>}g$$3wm=<_x5cX*?*y~4BI4+A#Y-EFcRc{HRX z7`X5^kXjnzC?x-&8K;S0T>Ct2lC(tkv7X}t9BW2eZes+8)<%jY9ALqQs{cwzVgX%# zXKnC~|3JQF$sfXcBsiMm9U7`(_gD&i5NgGm8cAWGlqB)JP|~$D1w~u6m-C2#&wy~g zP}QJO^4^MArV1i7*ZND@&Ghx*|Zw@s}}jIj=w@=o{s+6i2mu~&vH}zhn7*W znHyzn*IW7v&65=A{EI&~&>5KE-#8N!{2S+Ff`7j>Fp)S#<9{1IZpF5F!ne$8>2w^T zP&LZFqjM5s*VLaYR^cg!4rnGsLInRIXYX~g0VIql7KSW#UH$jz0GRiFz0$(87|Yh! zAfJnQ9y2In5$)yTJO_4#N&zW45ELOh?O=I~ZB|;FMY24^Hajh3aPZh}Tm<+PID5gn z-A>aLPX$dj3lardS$UH&4zSW8C^^wGm2eYIRFo}LA_au_AYR#IOXznBHRJ}-X(VzT zL+q{>ZFsjEo~($>gh*@1 zSph#8r(@agmr*1Zg#X+H$h>nXTdWo=3J2TEVh)J20DO;>Q(MLUe5i-(#OvV`j1djx z&f&4#6J=*@!LyzAg6Be>)kZ-`Ok}jTp2OVNYC4_q-}TwSnX=%YCxU;Hr-tMTKV(E3 z%PZ{4j6LhZ|FIF`!Qp}6*xAZq(RjBw%!PB&{uAJeQ)b9zhQ~%{^Sl?l!(*@m!yLrK z`7dz_jsKF7PC6s!TTeRo7VvpgD)xF9IIPwJ%mkjm6F3=i-1RufVXp^BlNv z1Akd|lAqlWe&(qXEmSATISYH%IPd2(*WMgGTa2;l9W9p1vB7IXN5-(*SfYtEg72E( z=-VaEiVDaVXN#{7e(L(rPfa3PVb3@eE%(lJ2JhfJsdt=>u!G5+LZAa~97R9_#EUN! zzc(Npyx<2+a0ZV2V+`b=7cFGpui}uR$VP(KIKF^(-zmuGeZ%Vs+Ld2|P!^p-KG|Lh zACQyXS9@tB!52$Udyj%6VtxsO;&MviZZCsVz)93^f<_L=gF3LDSc-{*QF}xG-hT%pGM&^R~|K5lFlcXA1|b@paK1Ur^jIeow2t!CRG(xrPREz7wp}y2$s~ z3V|iU@d2+JFcf6RnzXA!<8P?FS_G=GfQHGUwC?t`STrHy7OQY?n*y^g4&IBuMaCc! zTOrjK;M8DnT|~wUq6t)2fX>2UP(P2-56CJf*OTi!SpLq!%*@O%_nSh`Bk+nx(9#`& z0i`!Wafg%bm5s@Kh{g_|1|_}2z90#tF zt1+&`Y9)7>{2>w?1voIlM893@2b5sTCutYvBAN~qZ5)jdv1(Y=_9o$^qe=n{43r8# zMdqylNvsbiN&AOJE_i2MiVNHG9fxSI^7o*CF*(1qFvAgWO4`o(S=+7lG6V`JUx*P6 zM$w+mvC+?cdgHwpZ~vX}&=Oxha%%GX<(GT8<@mdQ^&Rrdef;H#ho1U@U^MRM?_M%_ z!%Kqi?&9yBocN_T2H)N7efR$yI~ja;4}bU4JAV38!FTt1-@WbQZ;Ti>!Q=VA`o*oj zRpKwcxc~P8n)-z1@`gspkRBWT(|g|jbMN_~UkKYw`trv< z`mwjy%eO7M<{J9z51;$S#GR;>6AK^JD{Mz1d z%e&|g?}#)0<#P~eBlu37G(vCO8K%md1vA#0I^|m+%s1g+_A)O&0V(h`d{u7XKP0C1 zDr|;Q{Eb~XL5#dpG9zBoI-0f@dD0f5EkB8`$Tlt~FB0dZv7)3u7jtyvM0q1lq6@DQ z4#2inXpmC)xYPd3v)gIP-hRN^&gKruSKB?aOkd)((C{65=*wvY#mTdxZ)ssPaycvC zpbOqSVY)syN?=ICpKbB7{j6Lbs*H)V9!4Bl9AW$<*93`Pi)?ne2Iy$fW2EmM?mrU&iN&!A<4A!RPq( zFA^b?xbxix%X8O-6ayt_W$^63@VdP0Da7A-pZ|c*6W_nI{B?YezgSjd7e(j0=*9RK zLu#=Avu|at)+c^)L8$q)Xifn;gw4AIgT5!$ywzSJBFG{ZDzCi zXn{zcR~LJE!=B-@xU_MEU>V9_J^*_JP+=+rrcWGEK6L)6wr>>lShj-nQdb+pz% z&EA4_5(Q~-s^p~8M2za9DAk9ERe3eNo%SW(GpDTp*rKW{jU?MfuhQ?CT9 zVRX5cV*u|wZJGOf#7~=WKN#blqCfL+fBP8}4{kt#n>b@6IL9#((o6S%bqaA%Iz>(( zQpw0DJX&vZ&yE@QY;CM(6w;ZeXS=!-dBg$UP`~UFGF7MoXMmB$d)%wXjeE5$)+>t9 z%+o6?C@x94oG30icW35r&V6u<`>;IL2MV#w(}(;KY`=vIVji$WjZWJ-3su8ubE=z3 zS!XBg>a6&Kk0~EP`Y(=vQ0aDY)NQwQ0=t{yg!rQ^@gk0WoZuRpQB1GOFR~M%uzk^fw)($Efyz7L%~Ji03?8$%l0 z+t(WRc15hW6p)#xw`;ta)3;F3ekf0w!yho_a8-;$ik8g7;Vp-Q;;iB%`(ss<>vtNL zN>q*@D6$_hH|bvCzWk(dUslHYLP3vt`l54fK%_@HFLJ-$uidZcEKMPrd3tm3ICA)8 z1>@b;!#Gog6oDxkHlvLOa>oQ>(>NVqofaE*@Y+OTd~ANWB1+ufU)Sz$+4E1Xjp0af znR$AC-x2TJ9Q70_OcVA5&J?3U@VTOcGdH5wXDP*oGLB$1#=^HmM2Fawk>bApzH#3d z#QIL*n0fks-5IK)D7ttYr#xy@BEl;J9qrj%4VmvhkMi9k{~Grt2!GRGX29j7h{`DbcxA4af z{uttq+r-B>ZmZ$t+HU-fu|u(ic{m;NYol3yZVE3%DP0T|k7`uv2xU4^-Y+t#rGJh4 zbe~}8iosE=1I0t;>EUvUKu|H;*u)k3o_RdoCmtyXGY^j&1PVxv7Pk3VL!`1tC7#@b zIWP2*^YVbOiY=UyVnu2b$NDBN6A&p+-y5oYEf)_tqOQ$fY#s`eLjf!6NVwabhndLz~Xt!qTAcWL$ zvhWx#Zzl1H0MYkDEP4SV%36b6K!gXWqBwFh*=pb*3?-i-@+MATh!7cKVC1Qlci=zX z*bViFfBbpvANMFFh>|J3Mo=uj0w0O6lwaNT$s4*nDYv^l5c0onjEX9d`%FvDJl5lu zUt=vgisx#`%>NR8s4rNww`D}>2XL7Mms#R68*u4))I;N|#}}Zku?YIT1FSGpzo5;O zN8)D!67Loy3_vLxVzT4NWxBn^Ibh*PCC)md03Uo_oY&=oC{*tVsm?L4DeX%vmj9tG zmXzbBy(8l=_gQ8>8)ZK0GarU63Ao?a=1>9d_Y6t2L-vuy^*gKr-A11&&>OIFv(DBg zo2Z-IW!NP*VXnTR-M_RTZp_d)bD6W`p*DW4I@tb6n;`|We=sHS_joZ3mZPz0z+AK= z^C#2dv9C|7EZO;4(WBm5?$I$S+f$>HXxNydyk$`%vVdY_RAwx)1;C!MDDjE$hpIe$ z5GOpTwtGO2Wj3_5U-Q5r%maR(vo_`dl<2`h5){0ose${958n33(oI-d>NDA#pwA(@ zlc2%3YV)pW@ODAsDoz*IAI8O$I7TBpa2zP2?$YHmmdLk{U_5tIhVM~r5u~*$H)K$k zia5<8WCtz-PP2S6fJ0+6kzboMBOKFaOu=T%l!P44qYQ7ar-IGrP$3YtcTuH@N@fb| zaPTCy^6T1)p{r?kFD-OiGKlAtsscFTPar1v3y-IyIiaG>lY&~wltkZAyFqx*sHzN0 zR1x5bZPq4qoUS~KjJ1~(e}ed?AXV`1j6ij z$Io%nKQe^Uyuc_fL@oUO1P?M9l+^&&(XfFqfRgJ7tV>i_uhC{tf%QB8X<@zI+rXiP z1J-?WMShf_SSoV+x4dO-b~9Y_7&0Vhi`v zdl%|P$6@>-19d{XGiiZZlR=@kFgKC$wM>QPQEgTfXpWhZ=sPr9S=_<8sFm zs`7hVbSUF^{Cg|J5dYa) zH5;t@!!B^bBITpn94Z+9wke6eW4vj!Aa0cq^|@&u6qO{Dbz`?ijNsR_JC_#0%^6#8 zmyL558k7kumal3vqyY96Qxbg#Y}Ml>mulTUlMwKZJ!5#*E$`PlWOWAFqaj{Gs3a^n z)@bvgpt91GMBh=_=H^QmOU*293CF!=m`OOXF#)4FTQXvIYj-p4fVw^t<}(}PGlXBl zTH{V_t`z+4FeTA<{9+>~;>X6GF-GULyOI{8MVV3o1)}BFLBitVN$p=(uz1{*MBlO4 z6P!+q@A+Xn;T%c|)f$tP+Khi>haQFCPgtqEQkzu;{%1`|^d0^k=MbRrW_Z^(0ra#l zG1Nb&-TAao?+|GQ6N39}UT`lqma<@;w{I1E2h^45Piyn1;QJG%B>Il;ZAWq3A=|{Y z4$7{+75DQZc39^{#`zQ4LP(4A?HRhb$LmFGpMEZHb%hG{?`kutfc;TZ5`71I=M%+J zzGxxIvUEu}LqSrIf1TjNrk%ISKg3h~#p8B`$tsws)SL%fCT^_e5bACu5d+S4UeQICw??#&p)jRXB!rfd~9 zx#3h=5Q5|gwtvjr@OVeQEj;7#S#v`ZAep)=1WeVQ!hcH;9VsKuY z0|kR8O-b|}gSFxm2%T2Ots`71KVsVwo-;1rqur&nxU9?Iz(Sz}o{*00@f1R6P(|p~ z+Po+Tz0#CK-x1n1it|x*q)pCLErfa9I`WlDmvB*T)vsydCs9Vnk;Z$B=+A45AT6S| zW$diq2ik$~`#_=USR!cH-B_z4{d3wpDoFpdDT%%#z5h7f11Nk+QCA>zid_WXXjMUZ zvbayrsUvHcQ)i>78`5A6_>{J2(rUn68F9xU_bpqS%*wuTV38sSA_tiI7NiA-wPBsA z4WH2FSJ8&wH6;<(27^6Gxvj}Zi3~_cE}k>Z#V0QO+CtsVGW9+o-lWV>cw#E=367O5 zThSZ$n(B8a9xJ;$I967cXO$Lf|Ei*U*9sC>G5P$u;H=US)HbIYB~%=$;N)?KbXd6U zxf-&m*s0z9WK*FCy?tG9CW#8LdrW4T$-oU*lazCQi#AURBsZIqkOT2XMzz;D=UdM> zlvIE`9lDuBq<$gaKs{PEM>|IDxONxQBZm{3l1GvB2rmk90gWW=sE`kECo$_g1}0}*9eMC7Nlh`52vdDwW1CB)x`6)z!U z<-^)NQWWolf<#w8sjG8ps!h`SIbExZ2^VYasYY!E7nOB!lhlTc=NGhlpB7K)Ce;Hb zvmTQnYrZ(iQTj7&t`toE#FT_I2A?%0A=gc;8CSG!>XW;6%VRFNY5ozJBdrQHd+ADF z%rB_2m^8bjdxiCC^~yf2dD^#TTEw}3xF_on=ZN0mfH^o%nr#viqE=|LrzqGmQxa0a z=7H9Dw%L6)gt~_M~-mTryv{sej@Tb&o8) z=w8TpEKjoZH)yl0Kz*Gl2{~*xo05>mgB7ly#=|t?GaO#g&DdC>`->iqCZ3F39uu31 zbYEsg{4Q;IrnM|1kyRb_CNL_zWayf5v1U9WWBa5wql)^Co05>~H_xnL3jBT)g08h1 zsK7<}G-2;I&;Tx$D^f1}ER3ipVn~D4;F7jz(yGB-IM(m=jq+^5eGBw~Bf28xBIsMl z93`Z5yW0FJ+E6zo5!VI-OCpUCZ@wFlkX-zSY3>Md;a_uf^-?b8xdtihU6&sWvMb)N zjaIUP(w$C;*%e2VvMb)J{i}+ay+@Gf>fZEzNx3;TOIVj&>+s2wvDu<)E4Bxp*6w&p zaksZ%=|XiTgXF60B{!R|Rzs`%!6f(NliD09P<-5!gdA%hV>Elc(A>(k#OR;~YWdYH zahfdwwK>``ZvU*^#k9Cx=T#h|!x-=$Ygxggf?UF?^Xu9?DaieuDT&m`8AsUGqfiS% zv`B}^i)=}Cs3o>wxVEnD%`!l-YV8eFqQ&Qn0y^;_mXKeuNt+u5sP%%xRkU^^p$qsd zZ|RNO>rU4yv(?%(MLuISYQpnNTl;N>?Vz@x(vGqb#MOff1+!jNY!`bVes{W-+!-8S zm}EBZ(PmY_biXMHIr#RnTJ+j%KG_9)z61?-$sw?civXQQz2=taibj!JaT?ncwE}KV zJ6aPS_s~(e-X-jdqMT~grOYumc1v0xe}@(7<=P@nt58pLQ7B(TK@zktB9UAHkqHse z_A3@8_z)7xqM)lMAsH7U#Y?dI7qq*iDBep=NyHV;z`d3pI-4{D5|WEPc@3M-&D6qM zrPis5GEQa0J-u$NSoPkbE#$PShx)%!{<1ZXzhG@Qn3N;-Ms1E1&G|l45`D*P%L%$; z4wh(4Zq~yz;nJDN#MHeRqxK=~4yHwIYld#Yj#>|hCQLqH~E z&3sOqF$Ks!G9{53NMnEfEIH`N$wb9948(%dv`dxUyx^h)jF9^U-;tN9eoCi!ZT|8o zefb5}nQPYcYrSwbLkk4tVWN^7hCj=N*Lbv-5D`oKupGTLca4?7q|&P2F;xa`v2GZQsJh7^P~|0Ca7|C~of?ikNb-i#J`xRb76y{Vekr8uCk% z)(JOh^P~*=>rF|>LBCay==ql4hO@ISE-s@rI=1o)Eu37f#UM*=!gw9i?pRt&y)okk zoWVN#oJGJ&@U6$R8B)L+H6_t^ux>xya70!EmTV4|^676X(R!1it!RrOEws01a9N4$ z2P|lsYfNA$#}gtmC2ckpz$Z;f!~r*03CYD28*Bp-l8Zhqro^RLhaWjQdS=fZ!nRx3 zM~|BI?jKJL#8C+P#K^t6^6G@ z3`j^W-enp*_V&lbU7y|d$GY#|7<;Igh_Dq8m$LEfVgu+BVpmg_ANK+FLVr6+neJOoA=JiQA&hk^NJuU|HTO(n>{qQl z=2YQ5iKUxvw>2a3VQoRAMZ^jtVoQF5bR}u8*Cjb)AJpbdf#{b_ zNys7eK1R8lGiE%5_B_$RRuI-6r&% zBRfew%7%tyf&zr`Pw?vgOq*2&|36{yx~({S9vxlH@oXwPacnf6Cm4Uofc%GcFVh0D zCd1KdaFd-vvOz@LNdxZRv{_LE{!LR7a=?AVl!Tl)8TKohIeFL{kdR#P{9r&r%7W)_ z0}@`mon+HY*rb|FKkUTZY1$B{cxYgi|23a>aQ#8L|dTkO(wbgC*}2!bs~ z*7jy`NgYoNwiJ?rEr+y!RZ;v01&OY*?n7*A%fs-L8}*6}0gGfT8waeKMLQO0?cR*- zPeHry$@WK)IF|EVAzyGPa-qN>4oiW6FB9E zwXnOr=t_0F6QJ*>M{uKC4qDyK0^At`}8qq|0fHio@hn`~P73a%ZeDsv6R zEX1EPbPLw?$%AQ=S)kuU4fTMh5JCwN&wn=+uS^U2wke6eBeWGaD^(GB624EBlo+li z0wT>=<=Z*hF=9LVhS+r(T%bfe3hyV%V+gr~;*J?@o|Lg=nUd%`azkT&O&ZISkHGMR zg9Q1;U%%?@E)^3%>RhZ&z>c|I+bLgSRL8ZupVEZB?b>S=TtokB@5}^8aI;~cQM?Hq z$+}b+Ky)L7bwal9!`hrGSRXMZ(RZxZon&{wBts+@$!^C$)c7(X@{X5(e=n z({=OUA=~xN>Dm@aIqDy)k9qL_opx7K3K+>g6M`ce>A0MRI11+7#40@hTbmU{>i^4> zgdAC4X5e}q))?f?q>6Rav70T_4T-^{JZ5;-ukX$3MbUr2!<#{s?4-|82%d!H<)Ahz z3OuV!Nu-9yI8?qZxHp(b&u(PXPXQ(}T2ZLxY{{tIqutN6gKC?+4VcFi!Uh4GA^Z{+ zpZm3$Qt;a=NL)puw(5j^k@jShEL!XtOa7(WolGnF)$nDAmjqjucsl(FZB7(%pED&P zhs$>{V!gIzH=d>1z2eZxnLz)n(~Vjl2R$0Gsb{{%jNEr?cQ!3@H{srTkK52_q2h2J zx8ZS#1&YF^lLuueOO2?DOuK@|k#XQravHjgRit7g#6%-UWWqNjq%*6YuOfrp=)O+n6Z{IXp)h`CgmEPzGJ$DE>~#x#Z@YTLor!TfkRZ(APz%ouEj~m!r!9JlmgrvO-b|}xckrYvCPx6 z%_4GE4&%fImBws1c7e*bjM+_F?7$+QkZQKg>40A$EmnbF)fQD+6?h=y9r&z^zM@t@ z0ajIal=L1EwV6dSc*~NzaaFd;=1ai+*`_ zHyo)p8@2kZ(8(tJgh$5qNjD*CZ02dp>h@J_iKkV!8#6dtPP_`80KAd@0KGB|+jUI&KFK+f`&Yct(Q?_E^>03>Vdp`rQIPv6`|@I50egm?wTklo;5R z8~f#)-}u1ViU(U_ z&sg%`qus}}l3$I4yf7~zGn`nd?bX_xDAInVDG51Xo;4*Qhb?2?gJEkxLUO@5C6(%`?g$sk+Ye1j=K39?k*lXrnfb{o}LP(!kdgfTX3?3A;Kg{$LAs z1Vv781)g=NR&!}Ry&La4O)*bC+7d4!^)AV2DnA>fD*c|e+9VGn5?NxZ(rc1Zm3~M2 zSCt|2-vx=T2CDl}Ia7z^CKSG*YEM+8gXi7qkix#29m;kd$d_m@W~1iUCmg}{j5V?Zqh8uD_FIeDoq|umx^VfuxF9Zpn03<$DKbUXCtuX%K%Ixi z(Ihl9-(|4gtu3RpU}eRg4*HVyO+1eJ@p^g&oB!E4piCO75@)%MFCzAC zxU1MFM7gR-zpG*FldQ_z!Q!E3CYDg>jTpW-T1MFT_{nRb|0<-tZtsadba5}C=1)fxlT8aW+VQj zgcHlbtKPGyUiOqStL-IiFw?5-cm~B}gdC*WdWRZ?uZx>kgcz{E6=Db`EF5+Y-t?Y@ zEm=b9fvepKMXBqiB>JvSy8@j!6a?6hT(+m{6)b=d7?Dm-R9Q6MV^#Q3Z4snZgDz5~7eCC9{2K7f!jZ=ESaJc@9d7Z}wCfpH8}kydP^oLUQq2ru&7&1-9*9es#C)_w!PwRB&ugBtnkq zX}-o84kg7-YgKYaN2_*a)-Ac#(CNdY(Tl9h-y3Yik7=WrY(Pi|i5u~6N!o}X)&5mQ z{T>h`u3`Y{&POO&7FYB6TUCc^m#Bn25=uo#;cyLiHFJduWexMN*LI<=X5j2wyNIA&2%$SxI^wWLg)-8WpG- zrPwKwqy`M*^V*$C3*&n1H3S-%MZE#yDu@NHNQwu&PMaA8q}P~|NDY$lw%z7)wHdon z5ViFz*g)D+78UIxH*9xp#dy6}yMJl%!X2&5pBQrWJW7%xKdoMClpx|YT9B;s|jN$9q9ENM|i{i7Hcc&H=oOMy?$peQNJk- zFx2nR{FRW;QPJjJ(VLPfiN5R21{&L>B5uBPvDC~~P$xZBWiYu3BiqsLSXyK^X2`t& zXW0m6A*d3zuHK@}jsn#iO-ZDN%6!Rn$i@{D@FHOE5Ec#2+D1LZ&D~ltV!x{0zqE*L zL8f*7BF*mL57);Kf8va6Cj?r;Hq?i-IZ{CTfFN-dmt1QespjdJ4oVu zx_&vveo8TMqe7EV-S#`Ca+Ozee9M$X-=P`ct2tCy!VegdTl)sd3h(j_6^?{rwcEAX zP)5p4rX=EU7_1ZIQiF}80SU=PpOzYfr!ZqQtfrGMR!dK};zxzm$E+ZaYl}6lJzeYH zI4wB}&0z^U>}R!EQIzDgDTz2d1_(w5)>tLPkEf7HQLhafa<5J0Zo z#oL~{DG&4aX^SAG;M+H3CPSR!@@e8U;STCmNgmWMYO|#v^Yf-8{6s6RD#Q-<)f+8s>~Azecga2mrG;P z4&l}_PQHaLOC&4IIRxiu$KZWayNhYTyAF9Z9*;5LLJ{CqN%7xrX!E2X_fMuI`i|UY z&mF=gz9I)Dab{|FZN;eFxUDw}8fCg_Z}FWY;*ZYN2K)s+(yFBTc-yo&QZU8#7YB zl2W-=;SfYwX@Tedf2f|*a!KrHeZTLeNd3-YPfq|!{k0u9Sg~| zw81e=HL=8O<#LJ0RuxAN8!nb2;@b1eCD=(UO_GzGu8b5T`vq+YrbYI?%p~f7;c9JB ze3*KQJGS99h&y4kK}Y$1cfX@*j?^lYn_Qr#p8uIP-wN`7VoE{|eXmfVkvy{HD(m|Xp14O0lJA%WNw9L32v!m66zQWZo~K_*_f<0 zYYIZQnUat;W+#K(YwvRZA>^Zpf>|Y0U93%^Y_rqYZB3RMIAg4gbqAFq)YKV68Vvkn z+M-Dd{Qk^@*S9P(J^a3*2nM=_qwmJZ0_jB7_w%-9h@a8!Zd!<~%p~y}Llaj;2K+|QOxW(6(dJD-vuR4A?`ZA{-2_D@ z`cQoC4sz3@)3e5VjPN_PMUWQZ+cKfMpvZ7516TGTf^|Yh}lE(=3IdMbveMuXtf35wiiq8K}L87aB+4H6GhQn~1 z*{>X}V~VzMhiojcoZN(M(V`oAvmQ|7+}?;oao~)jlajN*?5s@k9v5h{qX6*lrWneB z_H9!Va=5b^T+wj<^DE&3!##g>hx=_%ze)*1TU-kj9D$|<&j~1H3BjA>LH{ekpnpIc z{p3MUJ&z4~PfcWrgoE0@rD)ncf<)J$-usp0O8*?Rx~hmi;<0@}yNfBs*YIPl}h@u+Eva_gA8|B=g5OKWXv)Fo>vpK`} zaqS+bh3{s*>{D=@1;<%%ob@=)GRF#pAJgVWf$+Zx5?67m$)2UnWmG__Z4l{ z6#TwqN_tJJ1SaaX5H8&;7Av?ZK& zL~lZBAoG?L#{zub6g@ep&4QvQ_XrYK(ekeKZrd$YVYVt2?EOYZWWsZncTT&rY302x z%$>)|B%Q($_PHWLa=&zX{l!(?Fhr0L{=X+T19;hE0YEkU?oX|tB6>mfRC z?d*DW=ef>Mc{Sn^#qH%a__)w1Z^Z|FyN&<6fj@5IkDK`;BR<+o=n9Q$vm+uf?WN?- zlB!(j44rH*a}l2_(6;aEJNT>VP`nO9oc>zgbS^hh;StA7Uf*%rD{*YI(keKUC6u3~ zx62=a;LE>8|M?C4$7wH{wl5cuOIqwa`XBgTdug7^aM!Ax_TcCQ7J|IfEPT~Yd!S*@ z_|MB9#ry3=QnNZwTA6ZSB4SeiqPEmed}U#h*vp@!4*Wj; zbF%y?@!u^_z4BF(^)jRb)DhUfgLT;ZYj5PQeUZQJ6o0MvT86#uNVEO;D@pIaLB0J~ z`p>uMKmSJm`FHxyH8;&Y3~m#7W1)JXNFO`fs@H3cru8`I0DJDQ8-r>IDC-T7|I|?3 z6_NAZUF5tKoi1;iN2b}x)nG`$i(iW<#(}81OTK&5?%e!eORV<@u@1mYdmCL{1bZh5 z_O5vZyJf6|ngOmjR8?xuqS`nrUa~ol*zWuOYsv6lA;aaqP1zIulWk1!l#NXACs~;! zgK>)F{ay5*v-F?G=s!>3KXXl)8^=&2KiHxOZc&Zu?*$O)zgaMFYC?XIh!Yr)kbcpA z(`W^8ft{zq)!lhI7@c%?TMr{nh53Jq!c(J-hCR#sqtjIR(DUGrI(D;#TihaRtMW%_ z78O(JcLV6wgstQDEfY0{;B~-}TGGIBp;77xQN*^Q#13frllUpcIe&%ruPV;@tRQg} z*Q?Q;r%1+A^brg8$F;kdya*y4iwHKpDHUuqUz@bi{G-}zD3bnRQxbBZy^X=@^=dWy z2+DRfnz&jWTd#CvNKDDzp(osa$%CVg{So!*bX}inO5O9wxspeuY5r?;nBk z`+~^3*Ow6L3E7Chsm-T?`mdRiNR7IA#E#38dypquxc0{6HVotc((YSY7&oE(6WyNV zonV{vt#1gfgml9%YxASP_1A(#S993=beyx;5Vj`m=~~O>!IzIsGVKkCO&Eazt54Go z(s5jHmSxTu2jVOYK$4wYs?CrB!6H)Oh!B_Z7+R)i~Zi~3}(2T!)9>$Ac{+rn9i z_#|!hF{{c!ZJDGUtZOrf`qXDRbBj3&sYybr=snu3C`z*5l!R20KEbn{58Q_)QZI6F z3dxLO7c}Zbw$r@85WZZy(`g~hc*pQPdmEZEYm(~_s0ru0FK9ETfcjEX5~)Eoo+~#T zvsMi$6Ao%5H(}&nr`@x($ZgC_d7NdP5N9E*5{`7fMw=Z4tM3*huHHns z>-b~Q0<(OUvs9&{oDQwra6&HOeg}>B7^-(^%O)*UyE2tlv{3z7H{x)VKN|!=zNW1Y$wP-^5m~EaTkTf* zmCxfpo+8W4+^=f?sxlV7B1q^PGvn!G>sW0XM~i_0Y!-GO|OVRq2!`SZWAYSl`G}0kQM_wtS!WpW2t>V^4t8rqJlZ@#iG6nl<5c< z^%aM&xvye^^1F&M%&sF>fWSTp{oAz}SI{3aB_Yl6t*k)3j+Cu~sf0t^#6TsNEY`p?iyN zjyR3&E^QdR-`4J0TJSbxroo$-)#Oz|(J3tn9lTltbU>x#1cl_`n7D~osEfd!pGnqSO1#~(73E4TM+h2md^kOiWGn+P-s z)qs|3v!Xz=#FRweq1iIpf?F@|6%EQ6Mh$QGW(?OI+8sKC(Wr)X7R9gP`BiO08Z5&f)|O#P0NZy_W!<1}+1OoiSb^8P zzR{61l*4kpzLg1gB<7DVl7W*qfJ^50ZQA^LM(&LPi61Z|(OzD`Iwpss(9@o}yh%{% zdAfr*kn3Nchd>)$8yLMlFOm>8e{&YyuWR=^E!2E|*XKC-!as>>5NDC7engu!1=SCm zlK2(8$O7sUoL+VijRp#bVN6(OA0LZYR6m0#-iG=5qITEP0vF04DpF7kyk*@>g-&P; z2H_To)?aG#qoDN#QxbhgYb}@l!CflJNF_XHl$K?pE_~0X!uC3^=nr^GLMVuwNQ4$^ z^P(VhttpAVBea7M66IbojKuE4BVvD>suN>;ouJKOB|j|fON?b!yYp#B+)ja|DA|gE zCKlAZwGHlR1i}I^_n{#e1Y0D)w`nt|0KC(bMBf442?#yML2Mv+7E?ft#MKIXB|54}}~baa)IHRT=Cs{~?Tn7>UIAls1P7#>Y)b^c~~kRaw;r=eWZ$wvvIMS@rE=Q{~2x3q{aWv3{ELf-$Zd&-UrJDT@yne zU4u4o*HDWbT?@` zV+N;{h%>&9+%v-Aakktq<}ac8b!~POs6Jv!qVG_pjqkMQYRIttjdmx~!nP<=D%=r9 zX+TBki`u`gp!AofB>Ilhs6U`+EH#1%oHrGgyuVlr5*>X#pvRc6xTRkU zoI}A-rBN#T?9GuSmxr*O4fQz|Pzr0z3RPp4Y4fgV%wkg#eb<=X!R;8Qj-R#Yt`o~9 zzslc#x5E4HQRUEkmGxkcwjk1u{j9w0qE>OU;?@?MyePjXkxYUxMj;IQDi!uK+H5MYpE4!Uci8Xo=LQ9S;eDmW|E>rbsR#2BVMTaGTRdqM;jUmF zh$;l}l})D+t5uDd(dJjth^8rtxJJnB#QDR6%td>}L<JN3xC^b$6Uki<;5^}g`&p52V+ouTkIy4zr!E@!5<62 zIJPMlijQRNqz=Y+Bp0K7c?tgGL8I(Q{<8Ludz8BFNWM={>?-Nr=XgiU3q&@WqK{bI zf2`f(l;Tz@50LO_TX3P=7W^oFoZ0w8Z8p3y{>gyE|1cz>ZVUdtpwx9+(0CAReWGYL zEvx|i>Y~^b7WX|o7LGG0KdUv z?`D*GKU8*(T1Tm7CQ<=moB5tZRGy>j6shKIuwsj%llm>j>c(56tFE5ev-VKNhKn8W z&(|M0 zxTF&^zhW~NNe&SYeSKPG$xiQ$_E1^q&#h-^&&shX?7_(B25XYlqPz(!Y9msCosh$bH?{dQ+eo37A32UwDkv*;vW80tM*Nay!=?Icb)q9G zH9p!{yTCI7o-n9y#qHtl;B=(u0XNW5m9(RAZ^ji!XBp|xl9k70Rv?{?N>vWkAkjh& z)P|w7%hIc<2@+lX$DUiu^`{zBcD3{j=`?NNsfFwlTQD~7)$VUvY&K+S0bjVyqrpY# zL-SgqSNR@oeiTFbE>jZHGX0Dxi6UNP!LkT`lb*KyrTXV;s@ z79r9Yo2K|H7WXwf$WP^2Xc6VH#pa+T=V9QV+8dpH{4bLX^?)EKiv_0k5BS%`L{okf z4FM|V`{+-7_hxS}c$|vqZ+Vvk0HN>0-P0LCX|-uRUw%?7W%q`1{pGw6N`sXe^qni$ zpl_LykT&7}uo8BQ_LzFy`;V5IO`1Tckc@ZTFJ*R7sRd5_8Y8a(=WFuh(NL-YWl)5b zsBK7t!9J+1dTGJF3%A9JzF~0{+z?B7U$nD@LSo49L(y>8!D$`QH%c%N*&$(lNLW7K zqs_5mQTCgXi0gySAn~}6GZou40}_&pd1fk8t=xSUnP;q*A1HWmu7OejtgLjY z2jT##%cJ)y>tID2j&Sjg{bWxKtZ)*njE!y2p3#=t(k_o|Y*5j8WIX=H3 zw z5>ZgEh!P|Jd2KPJMgAaeT|z~A^i!nvb3e%;@cK#r^!k|%bu%lv8B&iR?NF-6=d_tt z)Z>p#N%UPk?xm^L*Nzj8jVe)mIIdtFX*6(n1NRAIQ)o>}$p}_0c;dSI9cl$1HA2^!rG9LZ=$NK87?S$a0jL@f}lGWnAf7 zrX>2N8E^ahH(ewChmBoc+a0;=H1fC@J$6uFx%{3XytQu-4rH8whY%z|N~45iyEYHX zV7Xl7#HggRcYT0Uy;MTjP0Y^LP=}6 z_XgNz@5s8`scfq*iiM$5{;gq=PK9(JNK%wKa7>$5MF+-AN%UO@4hA}~_W){;ij?tL zT1eGqs{VFqfr_hCY7MgMkuZ5=X@5nOSSxDUVoIwO2c=f*#o}G$pHnxp=%&A`3iIuE zv+Hg^kkBaAqoU2Uq8=qv5`9+>D^QPp2dtB|#7+kdFKIZw1650Qx+#Dc1hrBr+R}i{Z;L5r^PfQ zF}*Wu3%0$qL%C}khBZhwl(2qCn>hv6515kZJFIfl?FX!MevRu^IDwNBG1T)nXF&f# zyTfS#4Gq%$oX#Q`hW+(D!_bVp#!nrUXntOsGX>4hnUaX3X%G{YVO35eG$0|l_(jtU zE#d+@L&Zhdis_b|)pZ}*I9YD=ROm`|FWy6!hT^mg-6&6MLc2Pvq|#Zq-x9a{AmMmA z;)0gfEb%u}w+!`W#YX~3%rM=ApYrrbs^UiNUsY6bogmQ_gI?|#@(OJ5Y!lbNiI}+K z<{|xvNZaaTwg-1>cQx4_C<4Yc=(ti)aD?~@a}*eWrAe7>cWSeuNc|2|5_0GbGjQGZ z*^I}?;Nco7CtMB!C9#FF`k3)Kuic~c_@JgTI10iVsQv2id@W5%WqVSa6$PKiS-9Gk z$#fryZz|$cA#q9zX$~>LCpTf?zf!x$X@&1yX5=L?(WxuJSr~w%A@Zy?Lka|!O-aZh z(lR9>=SWtBE1Dzc8K|ZzvlTnv)q>q9;Lzt>26c{htUT}2mQq^fxek_p=+Sgf0GWXtSum{7amYq3qR@8Zm4qN#(i}}OaT}vzG z<(bAbHe^w0E<6yIB@M(6YBQn8_m@pc$btAihNxS>$JDH@9vvG!VbwS=8qb)HKV(q8 zpxv3YpsdN*1z1eUO=uQdmK5vxGi_EBX#T{ML>w9e11D`H4?qJFl8ZhqS682Ipe(wm zMcypcV@sR(LspR0xAn`XU4uJ>JZ`drn~;JeBsHzjW<*hsWu_$h4$M}%w*qGYa4f1& zniOU5ncb8B-Gd2ZmSv zG_c4`kB+0&v4;%IbK0Fr4-Bv0>GBuBO$eBTRG??H8Bu_lF(uJ=VD_A~r*UH_75TyKSXuJuHKM>;@H^UaNvj2SW=ej)utgHw zE$fE}f{2|cc9vuZ60(i{yEeCqBK(#h(RDT0^U%&iD2Gw1;MC@j>tN%dfO)ZiGT?q5 zI}P_;{ZB#>-)OV0D8Uy^Nl0t(m!>4-x`#M*S-?X)Q3 zB$$Pdj=m)NhWIAjP^Zgj)ou{x2DMRKJ~tf1I+^6GskGZ^f2 z0ZDXJM<4Ob;QMThA833nT;83HoQ?)4@fi(NvPEkpB9vQ?pJI& z<4l*T8pJ6>e5%mbLQpC2|`oNp3CE2#mq^il0zrA5lSen7$!WGqQoep~z36qa*@4VNz(YvwLxs1bgukTCg(Bg$cy*|=WSf}Rwd+T>(ZSLSNS z%CmKMbT453PNr3!ZCu(S%q7(;VJ<`36lAGMUD~A0l%h-PO-b|}yurhbnoG!4N+?XC z_$2CM2I>Lr4y6TZZ6+V*CK&p96GRhV_U@7Mw}(~4Rn+=bA{ z4D!Y#(67^GQ33ijrX*4WZR~w+@{g#x;_4J6oRv{#Hs*nLYs477SG#vxAQk!c9 z_>Y^CkVExjtUJBVS_Vau#q^S`Y+lI{9L?gu?u>;e17uSU>(* zTTE&7BUCve;*GPPOWhQvKz1`5>Sjbi5+Vs-*JfK$kiRn}(RT$oE=`CP_lKztloXjI zMYh}G1OBMO4i_y{Iw^95y|<)Ug|=q973w&*EtZrU`PxRjYWY=dm>?@nUR!7eN)x}R&q!$!A_Onj zW?oU63#KIct~57EOCt)ZxVYX4r70Q?>Qtzqf_Ymr+|O%wH!a*ZNgKiUm*b`$F)@z_49P%HN>dVj2Vf(GeJj4nNsc>p zX~Xc`t=+Y>gJYA}6Z3gX2o^0FPgu9zsm+gq)*YrK`i|CVFIq2rs7mRgQz;ec&ORSE z7q_h;rjKhIe6hjPLOpTU#nzdKB2}Fhl?bak&6!%`Vr<~`o;>Twd2NBF)sgRlzb0b; z!cCJB3ADwA&62XTjk&rP+Sm8aX-4#W>9ycY?clFx%M9D=8 zFLqF@!?aP(w}3>v4_p_S*oJ|AwRYFi0=+3S z?s*>7nkO9RVrAAk>&ZMi_LZkR>i=B3+bJd89?Z-Leo#i5b)e&o&rjed`N#iM`^P;J ze-M!PV?z>Z?&lwxQd%+5s#Kh2u7agshdaj5%OO9%{(2qzR?x9uHe{!~U_>l^D{~F` z-fxvIKj&O-B8e{MmXx2s=wz8}7JSJh{wI0(zKlOE@JEh6Z1It-Y1FZlV|s)3PtG@{ z*P2pXPTG+xHJxcsMOHIRyjZw0}>3qv<~x+1Wzp^mImRi+vXbMUFR3$w3H2ET8Hk4)d-z; zL%L2xdCC_ta9Pyln4G40WY_rPY5s8eqa{9)b%FYrINZzMhqw5*rnG;{gZ1?RiGm=} zb+Ycg!QAvn6*j7-R+bjZg0i`EuTBAza)2_+@$DV zug!xZ`q!G0cvXPjs~Ejr8`B+uA3KUwI293Xy4F#3R&H)8k5g&(j;H|>eW<$E?cw^v~1$ogC; zQeCh8yNXl?O-V>?US&$+*5>7AK&cV?KLS$0jZPy@^sT>{#OsHr4Yh-J!{q)b&p5}!X}q%>AE6}5;) zDNUce?LA{hibI-adW1=k|YhpfMz5q1UX>p(13V zLCfb*5weJW-gZtFh0cQ4jO8LMd>Yc?2P`QVbM^(XVB2g@Jpr5f6o?cza&mF-#Ik{u_R^OR$25JoN7{1MS<{8*Ae7o<& z17QS&fj}4u;lV(tfbe`ER6_V55ULFBsv#^4gqrPs!K=dk4L5CG>jAeYYyjM^uo3W}!Y06D3Y)i;n0}!#y#l7+ z&vOBmDEh2In*NSL8aQ@}6rfFCRY>MBg=Ah-NajCONaoKeB=h$blKFT#x}t;37bqn2 zE`?;iLm`-lXKJRQTM4>5@LB|aO^u>IQAoOfP)NET7$|<9hIb)Qn%q|?B=@{Ra^I(r z+#gX$?ynjsIw~;%4INnf5a(`MAh7AL2n06$d?2vt?+XMp{m_n!j2%_5;{;3K!8wjL z1=C%CwBs^`w4+BMso$m$)C*p_X}xBBM!j*;o>L9Amtl#74tz3&3KUZ9P6NfB$5h}k z4lCp_>aYYOD$+d>g>=s+A^fuiF+73iy{43zCh1A)Gph^0rMk}Y77LbiaELbiZg z6|x09q>wG(MTKm;#|;$S7x2VZca$~cS;nI^Ful-<{a8E{_4XPl^?nry;X zRp7DzY@igX86|ouq}k0?fvnaVD228J0yXUq1h&nCfxx!;p$fEQ!e~*8mWS;N1A*0D z6?m@q7|24Q!E=nkbujoiOVo+M6HgIHgBK~J!CeN*arUY}8xI&L+uu}yW;L8DTF|pW zWkw*7*VTbQv$m^1OCC^2zQ0q*=J=jMHo9?RL?s$sXv4+EhI-hL=l*)JVV^?U@Q6a% z@T!5rsdB88#+3myrUe2ytyO{6ZBa;r_A4aQg9^#?h=HQA^)yk5FX%wa>_DK(gbFaI z;MHN(E#7Z4R@aW+0P>?Jpl$=nhsMn{`x~+UF=}WO;|?pNadqQlD~*dNB@P{X(9$Or($XUeXY&$NPWbqiFs3f=P*lJ1oXNw-%aN!>@-hDuKfkV;1t zvQi7LZEcL3(XUL<-<3*)LBH36`+$7Twk=)qPVQLbR%1PGTz+{v=9C|1@+gx-OkQU4 z9+MB49B1+olM_saAs~0gG8xC@A||t$tYC5_la)-aVsZ@<+x1$yQr)&|6L32(`1bN!vZ*Q)2!j+bsVp7ZMBSdAFV?L_lY z`qdjAldL|6B|a>-zo#?P?K>C1$zIh;m2UUXxk*j|-^^X^HMprx+>Rv{bY^;OKa>~R z-I-00D^qcPFL1e6AGc#zJWu1#6H8+?O@F!bTX=K1*PL>{W|x*3@M^7C5+^)o zr~UQ5*Jwr4ne-*$mO`gjo$1uy({A>i>0sG0OgnixVdG2QVoEhG7Zf237O9DuZ>WeN zS`mZY;D=Q<<7k0|X4l2a1#e_5o0ST?W}tb_#NnLXK;;f`_F&U;=MJV`tr))|w;~G% zG~szBRuInHSHKJSQXG60Y*~;?Kz8nws{14rr#N>34>)%t#o}Jyk_?ZsCc^V2)>Lx2 zq`p#9s)x=mpf=|hka2zi8Rr*}aee_A=NFK1egPTh7m#s&0U75Pkg=m7D3%q)u%g&i6tjwARZ)y8icLi^sVEi=h(SfMrzqwW z#hRiRQxsc@VoFgg84yDT#EyZ&V0#*o?JptfJSRpC?j4!KyAGkT;dVDKgPB*4pM@uQ ze}-4mP%v;xRd=-0?iNG6nuML!)9kzs?a9erqnk~p(tXprGg$J$MmM)s;m5`<8W_~z zSMYuP{_Sw^8z}8HTiwwc`YkN9A_Z*h@v5{wd=J$&-j(#d$|cf}^8QV@h@qbI7L%h) zjv?9ZZ}U61`(6!BmU02}el9_vQgX8MCW=iRQrBx(CMS2&r)z(|M0=HMM{}{H?|h7n zpQf^Yr4*WUMKodipTze>iN-RLLn(XhSM=jEZs|#7BpGMcakvRjf%x4xAJ@)zaRBE% zp$Lgw4Pi(X7ecB;lAMgF!<=ug5+OOdJx7v?M0devlkh5IcVT zZQnTue()L>Wj3X|GtoF}xLV9>%j8lC&g5C5-YVo$bVMyjMax`~ym}137T(p&G6uWEp}`)v(AAad zgCz)Ll6NKV+~u{S2+ulBscUUY#S=E1iE|`M?U>LC+KBcS=}kTP6hex{3toJ7^xRd? zyy;cv3TfRmf5B-@TX#HM=~bCeB6<~En;Ff`EyQ9rB9A!q*V?&UCTHOlYTsKMxeBew zjr1fPv@!j+HKGf_vYalsmnXE%H;H*dbMe&+pcyY8T z^-bLc@#0Um34u=j}gWc#WCW`tr12Z#k*FW*S}`kIQ(i zzBI*(vP$coAN7JcFJtE1Gwi9}p_NRxU-v(5E`xJ~#s!QP%80F>wQq~D z=A9cCt~nYTN(|QUKA^19hP{8k)L1imSVQmTYvvDWWvIjm1eLtXaI{`r*Ym)HlF@qJ zN42MK{kYLyZ zBNm#jA1@CAs%7Sgs@fOt~|p>nF+#5@`z~U zzR?=I@fcmYHI?qg+(8C!#UR!mSGtFBejM>@$^N!%G>scnE9T;V&~xMYX1j&AjC*W& g(8Y_#G0fkE`&$`c=NR_my=n%)SwDGyp>z7b0iu*oZ2$lO diff --git a/doc/LectureNotes/_build/.doctrees/exercisesweek34.doctree b/doc/LectureNotes/_build/.doctrees/exercisesweek34.doctree new file mode 100644 index 0000000000000000000000000000000000000000..f3a14d212ab73ef0228389bbde1e34c4d57a9af4 GIT binary patch literal 37724 zcmeHQdyFK69c?ky;aFFr`h(CfQ5CkQ$LO>yehY&)E z6!?8rU61bV*_qv)i|iZ`B5`d;)Tr7{v(~ZP?sIr~fa(WM(=WvU zcA>UxYmY=9zUy7ZqVe##$4(n{r)e2x7il~UDf&7eOIRph@@xwyordmR8a?4Z>&tW~R1b$rWTv8?5V zqotZx3RVNdC>GzF>)Lys>)PYd?sL{E4fD;2;)lA7_V$Q98jZG1P&??_`@)JD<;)=H zIF&HKY|-rOa%w(iBHB}Jnmz_hzeilN#ea>}!nS&-nAPQuRcCcZIk~#gZX#GW8eeTS zkNa(tW_hgLv3#p@(dvfLp1@op2wgiDjS4n6GxH?g`!g-OVY$BL&YZJMu!Xxc^BDWr zI&GO9m!2^>GpDU;5O%CZ4>-@HXU1u-xfNj1bX=?Z*x_633452l!`>BE?7f@;!9+JR zCaQU@D1`3^!uM^2FkyHPu)ArIp*dmS1$vH0<6tFosbjWn9f@lSg$JK`qWHwQM;GRf z9h*6dzX}D0N8UbUkE(7*(e3+aa_sz~U07rwNByAVxl1~{gK21DOzU4r*wOlT=HYMlT+F3^2t@L3?%J1P<9tf$BvjY}lErqET% z!dP(;8y^QONrV9=U|vjsF?ZZ}!U_WDaU*OS7tN02g}(9lT3~ywVYwF_NIv-rz97$5 zno1f)TWJll{^=lGaN-j&7rdI_f``u87W|S9#8;5iu!%7o-|*^&X|(7eu&TC89t-#- zC>c*%#)|0%#+nxzfoIqz3?=g^4n;$QC#hopHHP?Co8z1ZD9WF!3cm}gjH36~dd(;i&*)pDV9fn=7M$>r4Z+gZB-i1*&t4`Ai9Ls-B z`j;y!E2Yr!;KR-Y@MwUz)tPaEnU?7(hU@xOD zzD4PyS*W+{5Fjew)R4W;UC!|6T90w{?w1flAnM zy3r}IuGu!Wd8r$tO{Vst zn0k#vanpV#g~lox+3{F6rsv=5)Oj_gj&w)Pzl@H_t~ek_%Y;J2+7wGGXfdYKpu}Rc z3(X772_L@IBD>r)-K7xZVmj9e{9J3T>jWL3NEXAXGeeCfE%ZRiU1ot zWM^$m4-7+nXhRwfylstW7`jX!(NN%RXGFSURWhQNW!eXy8tI?NFaqUlQ zG+{vL|8RQ6D+=v;#y4P%aE7U$V>!Pu7(J^6F+fKVsz$#0wD(>CxKr}|! zJ}gYrhf+E%%kE^^EWB8Df?|`lDhz*(>b4QNAO{;>xVIumIaHX6WXD89WUxk!SE-)rOir|1`$T^ScO5$K9M(Htrw+;&3Mn0` z)FlOXNRf6zw#_2Ec!cP%y@yp&?!tZ8YHB)`YdmBjJUMMVj5+n+_~3`nxJJA=Wr*D= zuFRice1ICPm}!5G%7Z{<+BkE|SquEI#-l~pjPz};47mZFhH_5nPddsD6ZQ}|gy zVbL+8@F+JD8pfhzV-$D86Il>E!5`Ed#V;os=Z1u(9_wxFCaGRezy6aiD3CGvzeUq@ z=85xcBL|T(ZG|&!n01OBVvR)$8nwJSPho~>h-OYKeF}w2$AWn;87J^BwrpTweYVeP z8op(=nwC!%oyCQbBZN2{ZNk-R*sk$hvze1SOtea>10!qIY_*Y6!FRC^MR#onO>K_1 z2M5G&r6x9Od6ZFSOi)V0MU`d2Rqp?8qOIgO_MmYbe>u_kxfEjk<>GlE7Xn`<7q@M3 z4z+=88AKKxGMoc}FVEDWd=JdR^*wn=5vzBdjGp_8l^~!HtHzfgpu$gy_eFXyX|AD} z6GCC|wTo0N*)Qct8oy7oJ<<5RO-(wj4Ju;Ne%TWOqolr; z#EkyuV48?+Udu+`AFPp3U!t9u8%=$EFAm$*JolD>XANPc(zZHKPc70*&@>(%nX}`= z2ciVcfD{$!n~TV;LNo*J6B<$E!$yiP_H*GBS*r+Vb;#ecy%k#h_?TEEY1OPYV%bP{ zz-}*t(KZ9yr=@&OMY@9(4_+vW{8xT|-i)^ZtsLq?sGLzx0?Ktlq!=@DsR z>-)CN5ZZW#0Bk#Bxv?K>Dn-<{Y6c$t^|KQz9H|5TVPFUEmJ9c}CUWR%rbyeR>9aretdt%7Ua_*Sc zI0FypS-cm4y8aZEpzo5Pd!sAzk7T1e25dxWIiiuAI~kfj|1NmyTpuDK6rLe;HTDYF zVVIomJCaH!1_4JZSQzlQY4793>8vwE=Sx*nv`xS?ki@JJ)Yomp;W zMx6(TGZ7k+!-BZ!F=^WQAygf&U1u3Z(`i;L$Y|!zVGx4%U8PbDkV1ktEUnr`se}_qWbzlO+z)lGR z+ZSVLZ!e1J_`-npjTS?)(Psy2MA@yf(H{=jh%BjSBxh7w(&*!4Kx6xuG@Z1Nz+OD2 zj9E4^bi(c_Gna_ekR{8pIzFIk2T@=qWzs8wt<%XK0vKXVg!2ak52*!J2eIU)Q>C+u zYO7+ehM^{@iDFFg*JFAJKF_jJ8FlVYSgEazS;J~5;5I!P%|6|;P5NtN**2){jAKml zff8+1mI1lH=en9+3n(dYunVb{T_HJqaHQ^w9et#HIXqvD%nEX`;n8`i)HAH}*?}y! zu16>bOpVPK2c>rAGYB*BJKb?$knAnHz0MK_{sXMXErfp%@ffb*F&! zw>y`Gq7!&9yYt8p>7!5`Ce#ed^=B9{SfMsmi)c<9ODoP5A`Qvnqf!G^C-;Uv}xCV1*Ng1oy zm7annZ0@#Ts7c)x7|h~Pux!6Q4?1Ce_AC5gVPlRpP_e)BKGu$qJ$Bea+;d>)#GsHJ z83Jr&hAvBIGA7sZt5cK43Oj>1xoSKM0lzRiJ@?#Xyn3TEZ)|af;|(|uZy?95f^%uH zp~NeB#v*hb0BXMFrY^)^Wd9ZdrV>S`2&tYUwpx#oH;7)a7BldHhRLBcEcU;*0? zyedMFRL}Kry8=7V*QwDg#{o*0?qFyp`ZOk=rIVAF7BA;9kg3U1?i?1%Dx-^Ss>qcy z?SIPLt{RSpxCnQGfd~`goyoOHZ2UAsL^tycv$@$>&RJ~_F*;v6h&Q`g;%i!Q^KHR0 z_ouWRPM_^Y-eKab{cf6I!Rq`$@+>>6v6up~p{i`0rrL$a&Iu+IC@^igC*~tOpOGGU0d9LHU&kv7~BTNrz!-)&};&)NQFD>d{9Ugu+_A`W% z#iG7dGsdKx&|b%brPU$Vho5MT_O}onF1r zGFzG}BtHf!!)~ty>|d(+n{G{3NkUm7{%h`4b&1R&DdqUf!@AthG^P#Oids4*kon zn~KI4v8jBLmyKfo89rY(x}M${x47+F$j;FUnfr(}TwsM0zg3Hpf!W1mPrc*9?41j9 zt>Wu+$hSP)Hi94k0$zw+Vt*YMcLh`3^IyYyfeC`aP@iv| zg6}`~9D$!EXPlMZ+PaHo97noc+&wh+URHRtm7k(&IrcQq#{6Dmj*5&`v|G6S&SQ7Z zADwI6jo){)P;zvk<+kz99WCu`iM{2*lv$kDN+D;g`m|ogN~VCu+?@;aF;V6j8f%$o zNW>tjtgQy@NV?!yx}d-cR<$IOEulL&{Wip<^xK!aQ~t4yiruVj0?|ZQE-;ws1?h7ltk| zbiDqTwK&6U_aGgB;LAa{S`JzCrbVIcwX12Sit&|>ix7GrRD z$dGcZX{V^@btoFHmij7KcL`XKDh3Mw8GK)cj#m> z)k%_dbqk6en)70ZX4vy6Hgq28;Kq^II*Z}6wfj?7XrdK$u2c^h6PXljoyL^6fh{VX z3@3VMm2@*~Qt8d^kxIqAydC?+QiG#r{ z2r~LR#bFAcTHHT?kky+MJH;+<*cbveit5M-Y zLBhwQ97V6!FT2aR$DTNuf41T^YdAKKe=e=b47vNQGk{B(Rl^P%X9_aoVdE!e*VT5q&qk49Z@ z1$XDds~+XsF5SiWHsHn|(>xU(`Gwo1HP;BKwtqiQTNotAlV_iwpG4LJj*-jFpHdD< z&XX}QbaH}@Wryw5P}4X)FfGlKPBE|nlAVW(w@S2DDZS7MEv^hGeGn;)1ejyPF~j0I zURVnyKvx@%3G_ANrOLqTf0^R-ur+0-U$^K)QB)gG9DjRT^ zHq5?N`;uTcv8xy^e@G5(*C=0H3R-WA8}{jPp^YP1-WTaz&{hIx_B)C4#4u5w6SFa} zD*<5tQ@@;W|7&s%a=B&Sur~>pRk6Pl{tJfp(H1SS}3%1XYue~nmr-r8(^8DS1)<2`2XC+P)X*{^a6z7s1x7rLT zZ3e*ueOXsB;=^oozZ?taIk>V^KN~AU2=Ps?T0(^E14mQYIXxZ< zH(KcSZhq9IM@kSSBXWGTZt22qB1zm77ry!`GV;n>fsC3ycx&JbRrHG~y^-b8X4LuE zo0Vatj7M);6f>yaUz2|AhZ(9yHH(gmG|u9DktbCaNkbHoR9U1Te-X>FqD-p9u`+e9 z9kx&6PuWCw3Mj@Fax>Rug^ILCdw{IaS(z2OfDcOGcp7$X>ChVmRdLlm(-zpgb(m&i zn^|#o$L1R<7TJ7T%-v`D*bNGaBI)yyH*VS~qV}(_Cdx2GpFKs(Y51>_-f&vPj226lMg-fCwt^LxEzo&4|P(h6@RVW*54YKsBw^Si9mRp-kYM&e6$0v5ctE9u+$} zZW>ECTj**vl`g)22*_)yawy<1FJBlylV#N5N8YweMr}Mp0=VfBY3DAc2A@da%OPj*bd75X#C464w_I^cY;FOJcd-PkquGq}2d z4!ioeQYt95YjrI@_?7{Y(&m5QCZOWqrg>KFX71C6Db@hVLN#~Easd8j3UKe`0Cq5+ z5IdMdq@mor{1-LtVzg|Hk(GidDh$5B(4do&UB^_PID@=!Zb($H0LPXmS6SA4abbD^ zY4bdN9bPTIH9;&!3HR|HzA@?EjWm`?J%5~kZva<(hD}6j$qm-rVIK7*D6CtaaSkYm z&BmWd5Qn0aM!l{FmKs;}tb)n@L29y-%gMdkPz4_)$|M`Hb^5M7TZpNBS52}&hE4Jf z8sQM%dL+`DlN zFut$CBpBH-rCgjuii|q;`O7>Mhk^gysS*~)NWQW1#kpxC&M5_0c-r~mYN0ScWgIc) zO2>=><)E_C{AzK2iryU2(o*@E7#w~vkb5?5tdxAL1`c9E+1#PfCUq{!G$zHU1CvdnB*^7xrlxn&m7qEPu!ylI{Ao?X1 z=#7(*)~xuEfFRsD)S25f!L?Csp)r^zjYUF;q8;a@jXbZqV!G@zo1vbNp&kT`p1x9C zqbB`|b&)~D{t~DWB1;0#-+>r8uKZO(h!}@Rz032%@0LSlI0LUot&oTZFP#ZhERy&d znf%x!a&mcF-IXsC*w-Ekg(+MNQIj|9(UnbfK$Qa8_|OC7kY_z1jv|s`QCknkXc<;G zsZ7TY)h~aD()ENm+8eT3A0TXhr(dsi=s#iY{nrNtX6tym!7)f z&r&R_>xzof75J&xW+%OvXnZx*w7;&n>B~}Ah<>wlg?Oh~(v0*8RV)5_^r7fRt}A-- zg*Jx1RWZPIJ*Z0yQi+iK_*k+~(JNUFz;6tTauMPjn`CW`23u&nF;zyG)Clz$KPfth z*AtEZ85#k0=!}VnECe@X9?cwm^(RHv70G{MhLka{{Vj^7Z!#^)CzwU4l0~_%R>J`> zTp}N9RPNKoSIWb7khW!C(o_*~=Ix<4j)hOo9h=1WEFX_=W5w}^0Q_WjYQ~w|#r-3;OBZ>Ti(SB6S8i)O^`Z*if>Vg!7r-f@eh`Pq)W*z9F*w2k^% z(bT|pIgc-}&_eI+^7QV!wm&WQ=QY$03HW|-#j=)*3rBh9`RD+?qBh?!SIY>7ATzxi z9iRgMT7gATrAPt5wc^&Y!2qOgjk;-(pK7yr@qHWBe|irw8s+!tcI_$JUf&TNsD?Ph z;s)iod@aTQ4C=+(iwJ~uqjx4>mf1SEY{NTh(gEaA!m*0; z$5+RvgcXz#2BjO_kb08>(RdLyCpy64u5bO$z9LrB%?S4TgNcL2u`hx^AvRlSJ{k42h1?=%Bm>Z>42N|*YFw1 zvg9~WlojBzDXW7^_h`;#NAuP)=FNII~Mfy7wu5t3j}Mf)%qY>~Sxya1s4<7(wPK8en*U^vjW+BhuEK1z!_ zrBP=el~hc0O-vmbVoR8=9pzWR;d6#dGd1ht3~Kr_-RN3sK@h1$8Mi&-TVDaFI)?J% z&9g^AmIIx7HPsE3sBYdPx*_ES+EkbEIX1@%>bx_r%8xhG#SLZ!Qc%NrfJeuo8^Btv zvJ@{lkTTnx!P%P$h8H$#tPX`(z`;bVTKC0hZ^vAbkC45C{+b&=o7v8i-NYX-?Cz#V zyuJ^9D@5vrkWOo(<=faZe=Osc5!}noXn6w;UjuW@7ndx|9jTx)S}%um5a3t)FKtVCm6V?a{Dbnx08h(S;e=+5zg z4eJoF0TTu6gAZV+y5GIr%|}v;|04ZD+VpsIUAii=w5VsLTt}i^x6i3_J@i}p zE6lz83y6AptA^Q;b^YmftnnIorFj^IN_0pehKnvknzP1N@LKE%MmvrQ?-sw3y}#b8 z_Y9yt`VXVS2`bPXBt0nB_;YlZiSOSK9yB#|(L5R@%>y#yx`#bf$baV>Ra9*}kH73M z+MjEbXeaFm{q-s20yI8Je?3Dro?;NksU=;Az%XKM7ws=DvSy2{$s()17}{^gBy+fT z*`KptW7BeBjhfiQfVwFq&Omwi`D(gH!^uXN?{|}mIAV&ZI literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/.doctrees/intro.doctree b/doc/LectureNotes/_build/.doctrees/intro.doctree index 6d9402c89b7ea0cf4bbb0caf91f39bf6874876b6..3c0fe5a3cf63288dddd1a98956a3be49c3159dd1 100644 GIT binary patch delta 271 zcmZ2EgZc3+X4VGQsS#&4vi>?a*>AV6K7VROYEg1#acXgSYHGHz$&|J!em!h(c52pS zE(V6!DH>U}+zbp^_LBn+is-6@1C{1fD&!`Xq!wi+=IAKo<(H%uDR6Oz4`ya3rvhK%+BjMGq&mwXn2w;6ii{g3mz*?}S`QS$s^j3~ f!zZ=a0)Wz4-jhF{%wP?ad2TY-X6M7lm>6R>a~*rb%rcXkfnlKf7(||%fg5~)|DarYzc_q5JnR%rpsl`*WR!shT eN{h`LD4k_AS@(1Xqxs}%r|lSxHeWs6#|QwdN--$_ diff --git a/doc/LectureNotes/_build/.doctrees/week34.doctree b/doc/LectureNotes/_build/.doctrees/week34.doctree new file mode 100644 index 0000000000000000000000000000000000000000..97615a79298bc467ec60c59f9b1ae080e9ff5f1d GIT binary patch literal 778590 zcmeFa3z#I=RVJusOG;a|+LFtXWm_S+B~_IwtNNu^OO{mqD0NHSt#)?_wcQqSRAy9V zW@l!k^3l~L8(V;34VyE7X^hQl7+^p!wudn=e87MKvtAFc9T?BRF3aMT8I~Qq>@F_@ z_U`O|&V8MXjLeLvh*C>l^C4Gf#JxA}IrrRi&-*@q<@a22$)%Uj|Mg$nX;rGt^YebW z+ziV>=R*JLa#-p$f@b%^cVC$K#tYwYVY+`!+h6a5y>=sJXdVRxN3z|XO?*?UW zwH-FRgJJ23W-0L0=e+4kx7+IMTUc1Db}PM=`BKyZS5~(5 zx3zp=yL+MX>fVaqFZkVVySmcrVzp>>F0Yn5Sc(2srMlng;358fWz%n{e_dVfwcL*t zzjU@8ta6)e3SAfaR}jK3^shPJsPF5vd|KtJTkW6|w9f?>di|@q{u+&Np;G8?Gi0!| z@Fd=M78;dW(Ch@wh2_8pSv1!cj`82%XyCV-^c#z_a5N}&d+p$Oh&eClm8rHink)Et zz1j>e9DCiJm8r_*l}js^_f{&eQgjE}d6H?T95(hXA6Y(nXrK2Kmet#{cb|8p*=>jA zUa4CRn_f5cx|P5yfkr#QoOd)_t9C$O9nWu;y%TL8tw3$nZZ+r-0jTxgzEz`(r2|lR zVa+OUz(Q4S0@2-!|K5!MZoz-IVxi0;RCZKuuiVkA+*v79rYkd**<&|Ij9k;dsskJX zR*LC9BjtKE7aUiB9kI9^D&1mc&Yd-rf_PJ zmvXwYy|T@&1ZZcfa+3C2W%_t!_Bi8c8-H|7BIt%V^cd2Je`&wbkAII3pOGIPy*p?r z-+`rkt655(Fs8b_|2tn@3*>P2RR&m;%7T%skQV!dT;haL6@JFsq&$?}MK(4Tq&eC%8`SSLQVT5SXA z>o|%HMn=81$f$fj0RCtazP-Dr`SypIC%rakLKEWzCA%`??*&=`7M zapd#wmLVfRKYtFKx?r`vwpSsEg1Ih0`3TPpTf3j~h}~R3*sflLKVKE4&h{VcplElh{R`^UBpusi%l*&U+T5MIwM&IKSCh zhZx}bD`Bt8qAK*1pk4LrB)YDU2>)GDLa;q16T%b-0iqX)UxEYdGn;YL zYYR|+R13iQC+Xn4&%pUVCxz4Y&~T1F9)3b@kq@hHK!|?TO0QlIx}zBJ=sk3sWwrP1 zy4!CQckOO9n5FI7z3c8)qqw)#FlbED)S)y<>&jH^GMGEkVR)ax@PAE;VcX*?8egN) zal1HdMk2)S;)JaoQWACmUf!4*FAr+G)ZU=GPCC&0w;R~=Qbbh>w#DCalZOqJ6P-0{kGxQ^JoMchxys@I$h{{9_&YoCYd|~oVs{o zsFh73%)k9=10uw#w^ltz+8;D8;^0a^!auxoe!Cj>I$-ZDsEtG4V)pIVJIwssK`CsP zgR))lS1<&Y2y1U6`k1Ockeb3CG8A@S3JSBG?oQZ=IGpKJF+}EM#UG_ZzwWQ#{3Vu8 zE9IZ_t9SwnCaj%x?;db4%hgCab9;K=>^6d(Ji8$`v;)6Pl+)?0tr=0l9O?}pBBJ}; zN&r(s;IRZsLbsw+Ck4JD=IRLJDMc7hrY4L>3`^F_<+w&{9}v;?E=wO4h9XKcKXt+F9;U z`SKt;oE{^;{<2Y$ckhEO6fCa^8xcfPLPS?sO_`+;A;Y{5i&%Gl$i~_USG()HAJD2D zv;vqN^*|>!x)lt0q8-A<)@XpT=RC6Pg5Ub}wP2;~Ge2GNVMX(TdO&tHW}x7@U1hAD z_fBB1fFpPCzSLF9mgR}Ba477;_g~?FfTe#^>YaPgQ1AP7*GVK#LWAw}UfU@NmGAkg zpz>EH6fso($^-;d{(Y&be92Jx_gaFTY>2R(5>fd+!^`0T(@r(~L-AS2x8Uw%+?s>k z$Zxdj+H3?SAGSA)ewAeix4v0M0dhZ?8o7rJa{pBdqJb0uNcFIMhmKmM z%DGPc>pS|$j~6rSYbg;7$r$D|uOyH$Zq)0nEJ4cQ6ts`}D`r7QH?Yf9`=1DusoEdn z&m`7M(#FY8YiZ*@a)&h1#!r#pycr`aNggZ!8}li--w+_M^;XnZ(v;nj{-s*dFVY|3 z#v9LZ$XL2;7WaSGT_;fl9mclPpV>}h4&xcCfA5Q`-5qaw*Y0^X`ar9mF+`lAhebg@ zMJEvd<(H+!)YlnIU81`t!IbTk2vf(%-2^H(>L4OGs%uRp9n+3Dr;NbOAb^2@jK!U> z4r_+-1Slf~I|;DotgxK(s21b3HxnYJYB#4w#w zO0dmT2$e99+gDz7)!V?!bsz41dbt+RUo_}KQ|Z7SgD{d#5WqHH57#YTxAGGX$$J%& zcc(`3I}MT-QXtuON`&Oj!JrE>2xw9BWv@v=JtT3ErwUk~gThxQ0wqPTgMYBb8IySaHB*cIi}G!^ERU;YIg)x=-es0$DaEqbPKlP~ zoTmL&9j+!OXj+62NNt1eRiGDX>ZH8m4 z=74j-23GuB)o0ff*j3N~5jcKxu|Dq>XGJZ}yHhXD*PF$8I>qAHPKg$0>39ffunAO? zSUv~rx4~H~3-4Ay{#S?^(dj-${!?=4hiKq8X>U=?j&7j^fa?PmM4*5T%kpBbtGfBK4(R}5{jd zcXRIW#(ogzZfQi;+DXDp>n`1fwEi>MA|1SttA3t&n{L?EG($kw&i#ZAS|KTQ!I`lVHT(JPr!? z>9Nf)9W$mO%E^mmr}M`a@8e;h#j>5;VvYY2St9HT(_SB9dQ$_txM(?#V|jxu48A&G ziV=oqm`W0cckKa=vi6oZ*xoF9f^`}p$fCv?+ZT*r;&L*CJ|E7^BcwvS?ur23^Y=l{(_KVm8#^=@Frq;?3pZRI2&NY~- z-5nWe^j@v~Cce?X&!n8BUH=R2taz;J$k;%_OzWnRG$JrJrqM*qT#j`%XyJc5oTyh* zwZ~E-gbzlYgJTId39z?LByIt8g2oep-=_A8f#3XeC@@bj(laTsaJ?Z5AlU4`BuKa+ zZjTDM1HF<_5BB_TY?y*8sE zopjRRM3&v|u-(Q8Q13Z(&ad~Bl+P9;Xlpn(_E8v7RHch(a~!#WvvJ70UG&8VuB_Sl4#_d&EddG_=Y~G{@4*ZeDIJKCRZ` z*HW*=a}(C$WNF^pkJz3vSsz#7%%DV64|ZS}j$L8}GeoQh;Wf}QIt$Qd1Hh2URx(;17%kJ`Q!jVAwX%{Lpytoh(8> z&{^0e$_zL84Oqi63?|T+Xi~=T7HiLkWtq+SbX<}S1i6V!4HTu|)Dt03!Iak0iIziY z#9e^3(7+xI+MH5+LOU2(+Mt=JO{U0#l$hWCYQmln?7q6-0_XoiWN!_sdaoIz8^c7S zkL`}_N8jKJUyOWV^x2|Rz(8Jt~5eBbP~-=k*KHmeh+u6O}LwB{5Q6&E*JS zS26VB9$5U=yenApcnkha!U&nBz6tT@+B9`DcbRCKniewO-EoQ3bI1my?DP4G%05@M z6dqaKFm`yJK5?_D`0j!)jA{X**-e0sG^X5=WDc0Z#C8_5lbd2v4@4$4X;a=oF2`=T zZ_;dq73KqO#S991D~=fBd`+}tM7IS!No>`~(LFBW`VF(QClC!C%+Buf?%I2AaqoS* z@4k1>efOHxjO6Xso{FGC7}u}lfeaPvUk=gc^;GTaBV&w4Cn+~Dt3URw4m)N7%#mIb znY8;_^bJ0MpDgIlTxl-)9$oJa&j)M7Kor@?k?SD(9QF?(AZ7gVJedZ}pH0 zQAVa$xu;Zbl?Tod6cn{WIpS~|QX;S4hHSkC899Ks%fT033&#+&9*ExTXnhVO1@&Kw z>PX2Tx4(+2V0$6OaZvRfE6zq)#=ne^i=et!E2>lTGPF1N&@JuYb1cv&BLk1eG|}O= zJ32J8VmiERpC^td`i0?53$y-6D!n>rv?|C8=|~swRT}qZ{`R+Cop=y@6I zuIQGC5@weC3XX${cigY}tJOJgRsEiC^6$8{zb@KEEcWm_VE2{?b_ZAYrtx3I@{VcY z)^|)JOkp}Qh0%yq3k(M`3+AKeCRX^Lx)nB{CRt(8?M-@mNjQt>mS}}{AFKMs$9}D z!=j3R$F2Ddab3i6$G-=f!w8xOSNip%zlv2J@z$;Ihz9uV`N(G{t?rxj){dW#SlJJ` zl{KIxSy|Doo?>N1w?r#@=VIA!c#om{=L$q^1gPWZd`;mymv1|W;vH1oKzY}-(-@g;R5k$=crLJcZAqHi)vkSs5I0r( zTl|?M-jGH5L(f=?^j~q8N~B88I3wkLc+jYtoVRPA_e7~1u0XB1-rBW?xmNwET=n_GeLoEt@#b40Q10iu*#-`~Mfv zj&44r=;k}=j|f2-n&{n#J{MdgHkV;R_2A@ryaPHGC0np$C%PCZZsD$Oafl8dC% zp3HV*8lv~S&e&ab6njw1s{YMUj#To|7WsXNziRRLl7l-{`=QiJ_T#ZjmLhQGy8Cw* zckSG%mvNt`Zmz)Y8Ghj~GyD#G>St0<{slMrpVplxF*?#iM7NLIZez5+*WP0=g!1smJ z@O{j|_xTj?iEhc@+j~#(?t3kK2Q7Tj=K{XyE5P^V)bM@6!T0wm;1k^v!nd=y_ihW{ zAqQXNxr8tB3h-TjRa$=k(+GJ$ zccq5-lMdqP6cCGU4q_);#zi%f#Z)jVlE{Y6({DWj5lrjETpNir+kvu3q%|u6$f%%D zI-PDxmA!=SP0DdZVTM90^;Or^3Se4W1;>B6Ovf$#%VkEOwWZXw_Op)G-0C}`>0d=6 z5X2p!cZf$3Yq=BdlPDM0oz$hS&E*wlvkkwn$gZsrgj2P5o?O&8cPr1L@bh-J6=0u!3xI0ARHlA)i5y)stHSxmolr8r@P3U2Ilc zj{rDY%(3-fY8ym1w%W^MKkF#UpogSox@{H5i%GW;gA=(%dGTqgJSy2S zbYW6c28fvR8R_bP2C4J&7|KI( zs+G=2K!qXDXZsCs@)deo`odu&XB6Ys1oWA-()zr<*_oAzs%e~B><9oKTqu79282t|9)JKwu`XRFOX(1=JC?hH>NRa`@N_cFsXCyRJu39I&Tf)Zip z#N5Z+!%85M9P#U)vvR}>+$Bxg&zUohtX2)=n_sV^v3``|f!&*r{n(6!m_7PtM)j)3*}bYkLDF{ZZ~ znPen}WUl3JxK;d={?Mewr&C#U`?m?Zk%dHb`vu!=j0Nc6fl3=UUW6^m4>^Xscs{JX z=>z*g7$-Qsyy>09!M@g1*J}_3MUO+Xiunv?*R=Mh1i@78MGc5a(-o&2;Gd! z7mdK-YlMn9?^KIY*@O1r9_;9+TrO*^wLtrD;r@H?+leH;`}W><_q{u3v>-pYCBJTK zCTA9VtA@AB-`}b)fTaGP#3ZG*JoF`X^*e3($qOR)|7RU7{)44O*c7DfUi)@>GgUJU z45hpCY?SaWM*llI`eaHM-I7rH={?W^&Me~A^g3>pRwAr7z4$07`!LD3FFyZWPc(S|5yJ*35&L_|^?zIlvOAz?yyeIBF8Q?J57^Hb5w9rE)`Fa%>_Ov$^=svRr zBTnO3ryJgbqB)&tN9@{r|L&c8cHgsS_X4atTtS$nZg^ne7^7M=D6feK3l#MaJOp!4 zjXS@J8pYVD8~6n|eBiDM^Dz%)Z|l0KD%VjV*BWjT3Q-=Zg0j;3vQ48e;BzK_p6^5- z+?G#oDH_f0d}eFD^5XL!Rm&yx&aDP$sMV*2r8 zB=WSw6@LZy+rw5-8scblxDQ-J=C~HzxGRmiFv_(W^&%hD^gp}U;1(x1@6SF7v*+^| z%27}rqha}vrrf)8X746JsN9;r?Y8FOgj7Rvw&->!VK*W+(ajOtc+w`nhY-(6rBk*x z*R6jIomWIKGE{uvVMI@LI${z%^=4!J9=Pa{!}s6qgnaKS=KSh^&es>>Nh2Lq= zGO9Tykc3%Xd!mhM{Ll8PbRmeXofLU$@lyMF0%NN7bA;do#ZN3SI4w3@25z%KJ92O|0=x@ciROBN$Ip{f3iuo>7?c;f}iS!}V1MhAWFykjW z^^i=a2539NvBl+Gh?Ig;hyFqN3jB}VJ9&oiUqVSks3eLI(J=Hal>P^sZ-ppi(p_2R5^{96{;K04bN@vcR)H z`ysIy@+VcPVh@2gdSZE zTafQ?_Cs(OGVtz0?Iq{P@ev5Jfzn|t#e^+XYpXtow}|}!gF(WeZ#0p72X*U`@3>gD z+pkF`O#c<`A=K4KZ{sdbnC`RhOUatJkOm12>!DYIkV*FmPp{uhP)fo{c1}&$7EMUqd_L|85?sWWi70Mm@ zc2UQzf~YXuJ=*rW2t7Cj-U^mY`CD3Pqt#US)>aj@SA9~C@28; zSgOUxKgvaUuoc7AA|Gt-_&utI9eH4iaRpIfL&$OMfjMMCawkoa(>o>G)mhFdd%3l` zpw{X`_%n&ul1==wXYHx^9o$8nns*=smg=rj_$sdu^AC{-$ctw)0h2bjbL}VXfm8N} zQ`6oRgOF*?nIiY@j53kHL0>}Q7rxOMjE0yJq$#@kh^}@%2d3$x!5m>o{2QjFx@x)F zYnG^s63I~XAZ&LYFelWOe#0WV_VbD;el9f}XN8_Gq(;v*4n3bwfgaH<5qj=?l#+vhBCioZyeg|2BrK%J zBg7&kaeIe?uP7A!V`>z<%Aw#(DNrD~B|^cwk0Edg7wVuyBn#i@UA=kVSfb>%WDgYYQ>D=k=!W8zCEghvm<%TYuA9kL|qR8dQ`ia(PW7%3;O`-oLeO59~)@`=1)XcoSElobkEHa-_l z=yPG6Wl3*(_s*U7DZ|BiTemSy7YkKCO)n7WbDGACk8`~AM)6@cI6SQD&Ttu~P-RXY zRiR8LgEhKqju<>`(xc&;iZOK0p~v9f#}1iO)ImV)+O8YC%BcR*a5D5(3~fIH?WA5V>{8y%{C)WSR&RiaxWR6T?ulL(_BMVPKGgmickv_-cGAv(H4`t^Q)ZOups z`$NP^(2uBqkaZY`uumz3{c>uAz0o1;7gHcibW4P=J5KdlE!>8Nya80BrEBRq+W=K1 z!Y1sr9c0i28#9)Sk~Vh;_)iJ}pG%E^H#r3SUJ3+=Zix_Z^vPzYhsdIHRjOQNjQh;F zp#39{iKqz@XO|+9%00HS;0_9^mIgV5tyc9O-H!=6vS`Y8P>0O_Lm~4oQX}&%4w-+R z0-2(lL#7iWMoAxn!MUSkd^h5Y;som2Nk;jiJbbU^gif<&MmC5%b=bS^Rm|sKLw{2F z(maQ~SLx2_j>g-L2R(8FHUl^X>HcUs@JMQ~1ZPPnNmB>X4h7O}sX@BUfpm)nX)-?+ z-QHxoMVCleci(a#Y#M2Z;U7SQ1$t652k-+5;QLbpyu$%}kDd8sfJL`N0KX9_q==0p zg{3*T<`&p-O&p*n6`&_l1A2!8^jHc&MYlwNzIw6O4H1nATdKS7bMT=9rmBEhOAXAl z1E!n;7||^eFdL5sO%#-*8uCUKh>LUe_J{TEbI75yDS!|tzK)H+d`Nf_C%nhixOLpF zUG^5s6p~4;>)vtA2rN-t#2L6{dR{Hl`%*8{tXrn{q*x}=%`MaTbApsx!Gs(;9^Z|Q z7s18d@iM*}eNCK7q#s6|2j%O>Qq$a=qdB)OZaMqc zMH6R5G{1RgkKf_C7|tw^#85GJ2({CZ^{&e4Q0{Fv49Uk77tU@3m2{o#qSSs>ZPQ=D zpCks3+dgFt9KXa}#QA*3>@15S)lR9`ApsbKN&c_8zkF; zYNDX_ZIj0}cu3OJLf=o)8U*zpB2ycG)Kih}9S%c0?B^5R+|13{6Xr z-c9`9K6+%C)3`XQAH0S#y4xYjOA(qMt#%LhR_Lx%{Oc%|4)*eP&uf-BXv z%3w}~&f_(aJT9|a1lO!`9el&a3_+M>ds)rN(nsyD3Bgmf|2x(C(<0sB*8eZql;o=#pa2D6FQK zTmC7~$JMVM)B*>S@oZwf%Glu(ee*~(J0|B*EzNVr%Jv9Aqbb)V>0OULB1v!B?Er(G zBzI7WZdd9_Ogfin=*1rph#p)<7sxlZwK8oSEP>E$lkt8NW zf;rpGRL>Z9f)9mjxkEOIxUlaD-u5Oj4gCBd#UAQS%8bE;n;OrDM|=Cp>Jx^RvJ7Df zr}i+d*HrCoLSh=Vo5g-*AJ*ogK*(17B$04HG6?=GxvbQbhCz{1MG`68-;jVwrT&*r9M7GSg8 zE%N)D9RUmWq~2$f-8-sK$bOStym=JL5(Il49lTx7WD@z%2`joCPuPtZh3NLU?Kb9w z{YFJ-yYDBJo8sn%3wg0DMS|`Lt}zRZ9MyX$x3sefYTJE(0&07Jscmy}G9;I23`cjF zh5_-N-*Vy`!l)#^dmQm~Ejf%n?I<>$h);BTHeoj+KG7`^@qP5nspZFa({&1lj~V+k zvvnKQ29ncWC#FRt`IM{dT;OJd{!vAf`xX`^?OqCoSkT{((xgty9h7rU=090OtXna( zqv5R>9i;vJn<4FN3nNMUK1bT$YiX3ylSTPsQ$o!&Fh_tBfktVeV)P4k^zj>qc7W)X zXb1e!nIp&d?7VyD!e0E(KoKF*TNc}0^81q!^ax`6>h6i9^_08B79^R(zeX;j+$8(lCixXh#@Hmsb6#qPyu3#S3eC)u?_L4+FIliB z?_SX@(eC{NI=`rz#2zX`u^0b?fHC6Q{!7Xow?y)OigRhilE`iMpxbtTX6X;x?rE|lQlXQrSvH5bE&dR&Y~V`1DFsr# zV39I;Q;KehHsvqkNZq^hzMbB2zsnxFe^9FLv9njMo)t4vOKj_*mt)YBxATTM~2P;^xhXlv3F2HqCa+VI)g&wIy2EsoFvOnPeTJKLZ@P~B#AobQF?GElCrbdIj7ODaa__@hE!}MDd09QZF6GZ zjgA;3E-6+2IOd8QR^Y)S!zQAKI$96$)^KNxNODkKAUhP@xLG>8)~39NSXO)QCUIg2 zV$=OC8T2r(Y`bdahW*6af*cUM|DXb?GaQ!Tko^}Jog@oP^uT+!_t z5_TgCmgts!wI_)Nr-NVNaB5pBzEue7U_xv=sh+m$M6m| zARg~EkP3hs)xBdX=`=Y~y1%2VO%DNb`^~KH(l8rYg9aHuxP&5efGgg)=YuAmqgV$A{_xs1DBQQC`K>g??)U|EPNekQuVdY{C@*(~}uk zMn^L<)m59=$XI~I7nSvZ5&S>&Yae-7DCSHbLCs_1q(Mw%9rZE@J-Dfo$LYdIjJZRS%Z1@*8+*e@WI2<{|Y5`;L}Hl#;Nf>6~t zNPF8LVkuXt=rU&Jm9S_S7GZw&w!wl~XugV5pC?`1TE`bZm!9pRumxR_%ky+KqgVYB z?(4wSqD+1TJ2kq!2bbKbxj3X7vH{i?l^QAD5g0h2_@ts}M8HFn(Fmq=1x;du72T-3 zbUP@R+Ft5DP#;C3+h~vedgmBY?;u%K=p9OBU6Brd8jOKJd*Y37Jy_QD+?qC^N=9~H2I8| z1)@i?3N!+m>|R|okePU)dUJE9jdwEc2OP@XUZM;`J zrKPFMPuMl*>ylEGrUx4z-Gm1v$UU_t9{U;!kSs}O3YN_&SD!4)$7)*C-)GZ(Uq&;s z%UJs$NyAgMZ;c4Wlg+X#Zh^j8cb=q+lTk!;d){^%W66JPKLks^XPD-CH)kL-D^+DE`^hC|-3a{?{o`EV?B^@wYr_^SR9x zf5x)03NNp#(4>fRpd6MiFB7(r3Qpl|I2U2w=)f!h`4-iHtG;&><+DQWda_EFc7fDw zoh>)ECG(fbY~1qwj#|FoO1*p)w|t*Xv3#OiqUHN}d16Egc`(|!c#Df2mTNoCy2T<6 z>_hY3A!Evd9cwj!y#-{k?zhX)vLS3;6(v-NhpGFw$hZ`lnqBZ$?ED>x{HY?57gLi+ z&5_8Tq#zN|EfI-4AxT7u7G+(0%+wxhW3f;t3M;E>kFAK>W2jdeUwbU~aUDv(tWf$7 zsZm;YDE+$>C>7ljq4Z&iQf#Bx!eq1obe%ht+GuuwU3ku@occw;3%b{dP3Y3U$ ziBPg#7W3g!@5bdB^TvVohyv@O)L^w8SPvuv>wB)Ca5elNU*|3QCxY*foo`8Pb2qdr>jk#y3P4E^NSiVx&D3l2CWGklh|a2H&+=A`b+{BYB!j z=C=ng1~tLH9mqGhpvB(u=^daOKAYtQh&fBBEQmeoP3wQ49O@n_Q&Gq4+B5X8$9i>$ zGx!Hf>MDc{HtDZw7%k0xb%P02^2yYjsp~e=$z+>}_mJr3_Rx6C0{Pd421?%GA^-Yz zBd-;$Puh9R?Mvsd9&&*@cXJXK_^WNI8jdr=v<4qq+Npg5p=YZ09{ibfuuydE${FkU zKF?j8IMAT}NbH7!SkzU) zWJw`g5<*$w6L_u0RQ_dqh#4ac3)@jMijH&zpIAsab75ag-4)3GHcgV&U-&9> zM^^u;2UptO!b1hKZ1%TZF|N|_p~wiMmHYlSCD~l4?5Nycxr4$V3jzJo!bRmH-iI|Dj3kd=uwuB=b!VdsLq8x9bAtFjWL+_aD@^q zl!&_gj%6S*j^12n-yp^+Mx#zYW@PP1b% z6K6L6#ndZgPSRl@Wsg6IU>5>`E7(!$o`?}KcGdxdQUO5-d;?PKh8K6NA$kS-f-#0$ zeDGZi9x}EBV<8@dRKarrVjq^lT!Qm-vjl}<=ztKKo^q;UJk*1xI(%#e1`-*6>Z+dt z?hx5fmjr3NYL%{u`+*=bmCHey(a%vP8fGmD70mkJ^2gFy3Nk)(s#HB&?G{-*SK5Va{vi1CdnE#ru3h39hg5QNje}XpGHcdohJ&97xh#j8b>r=V)hkg;gaA=(>Hx_z zErLK1BIuk#uW%}%_c$!!ZizaQEI1eov}C?qn4OgiG&}2#nPRGaE4XXQj^H~#7dar} zLQ-Pd9Lj}Ph?(a!U}}iOsYZo0bKzV9to+8z0@)qOSgDwI(jHl6wU_XnmT`+YZmxfp z882o&HN^CD7%U4cr?%ky63}hQ@LAw}Aj8=z;L0>imiccVI^V^tVqV64YcTcW)$;T) zYQ*3}C-DwG$JpDhOKPE1nR2)23IU_U4;d?8agwb$dRXYcf`l~zwfY?c==eiSu>7}W zl6S;ZG7gjlhDW4CvKgsMfRTNyERb|BhWIC-zc}B|g0X$etn+L}b`@}E870f?AIu^? zi}`08Cj$i0TFr)AS$oYpt-U6~K(BiG0+`g`!!g|UiNxHN>^4+gdK3?hEA17X_GH_G z7@e$pL$-%~_6B#S3R5ZwYf0wZ|^(4jyhfftuIGm8EPrKxNY|F7absd;^d0VcC0yZ*zw_ z-a*z6*%8*-expIg4ik%3(1m@JLV&{RCR2m{b}HdIHD(ZCUx6=S)h5Phw?v^{T|ucJ z(l$xs8`cgX`2toO12$lraC+dLW!~PYa47c0rn+N0R3g+Ccbvww!A+Yvte%*EP6MWf zNC*gvnQ2MDlfz(HV0kzrRSQTm50wRuw`JR0j?HEqCr{7nC^IgnJGHCuxZ+=6Kd`fMN*K(%+TMuT;UPE=E?FSA3D~1ep)!tBZ6s9u=~yHo zaE{8;TGJs(-Gyb`Ocu6R+3E)4toJzXufeTtogjcklb#ZvQ`O>wZfR!Ta2@PZB zF!_*qU)&+%y7gaeh==0XL{fhAd8A61^XVsvE;zJ!WZ~FRX2C3Ile~*7i>THV@#+R# zI()cQ;G`*g(AoJO;Ful*_U$En(j_>&}_q5VxZp5afp zOXUTl+riNeb#8i0Cc*D51=S+{^Nv>OJmwlexRPz3H#^(#TNKZs&PXJdldCao*Kt$J zsu!;CsXsgGO`m}#pxjPSc+lEJBEqYL8p3d&aUQ)Z{!#Kvjd}WyJbcYyy|b`;*ZueI z++D<3tcuOiI7Kl|(XZmF3N;Q_pwrIc4er8oTEIn9^)B)tX4JA!ruB`ZKogsBbQ>(d z-$lyncH+WACb|)w8u^C~d~0(0=KRz&d`!Al)&>QKiGg;D4mL|FYf2u!&Edq)X9At_3eV9^#JRH61ljZF{Y-)kDbzkMpFzElhZ zY#rSxY6|}EuHQNP0Lo8rbVa)d54Mj0DqJ~szre1}O zAwbNAKVRz1uLSM02-4V?M^NyBU<&i;01+#cwkMYJ-UxAuOW_{LiXgMMvJd38UG}j6 z^dk|VC-bFqjxQOEjpj?Hg)w9f@}=42Bx4Ymri(*WnqEN4a$=+?pVXVQPb!w2C36{t zJIP#Xchc8~$#F7w5u@+b5G3I)qFW;F^7BujwC9GmzaEr3Az5D`h=Dt>6?LWFYj*1! z#8Y&bQayxD-R?EP29BusEEsC}igtBz4Y7`Adm#mFqsIxq0V}KT9V;C`8d=?4!57g7atHbduCgNu0D_k!YmX&c zkI;?aAa^E2hoSSt?H!RP?t+;~i$?~T!&!7H=|1zg#(u$R94GRg zp&<4A#|E|jVuV@;_r{$G@nbg3%s~4Wpa1Acj0(7te`Kom>Bw|P^BA?c-YhrQ(7MTe z^qk{G1}91GdlcPbN%os&XAdKSbf2=}D#C;zoP=S>(V*RsWn!2*h|=^CGn`AS;+PTc zDKcfHkIv*~uyj7)up510@&6dXr8rsMcX8)z$mwEF!}0oyJrnHI&qtPPH0c{{|Gkc+ z&05p(8Vw9E?<&*AL_MBUQ+&S|+L68M*540NOE?wBsRmFo+9fO(gC!<;Bo9a+&d2& zdL7fkl2Yx;*Rd#mIsQzNq$yYFn|{*T_AcQrE?4RFb(dXo$tmNhN1iKhd<3!COT>ow z3642rH0V@{u1hqcqLixIC1gsiS?ht|ErY3N14SP@IA~EX6Cnn4_%h33e$(*J1_cpO zuTEZ?%$$r(u--5gCjnDI+owv4W;RHBXhSDf`OCUMP0L(K3GWx(qQc(;iKuL=I+vHM z`|Es6H$t))!-1!(KPqe>0a)cdU|_`Z&mk5A6)E{L3eF9N{eS}TP4P@vBC5>-z63D< z_p*^;0yYnF^HovmxJfl9SgxQsVE$!D9w92{m@6~qyyc)&X@>Q1ZDX!*%qD=Ti63er zy&^yg8Ppe@;{`+>t*!>H|4|>Z3FR>}9_+d(N9Exj>Z;nRmu#0k1@XpxzvEnkOZiZS zIH>a)5zK5mZ32-)9v8(C(+1{=Y-B2+$v#$UI1GOUwEMH~7XeEiqh*2SP%c|n0F={! zSzvEOVQK=Tyhh9d)&A@}RzQ@;Xj!1yW%yfic5elX{w6xd%+(2T zqZ|$&7?en&TKegqLvH7hE|haz$P-ZUayH{}CM4VQ@Tft7KNJrWK;zaN9VfhsBf>j- zMq>|B?2Zo5;cDWM15amDoj+jWWEPeD^TK2Jf)X)y~O?P_LFf#A0^f5mmKC0FYnbpaR_F5gC%SOxh@;fTm>% zOc30z7&uK&UHeXg7tO>J7pGC_Z2(OXq7Q$ia~76N6-bKaB!j5dg*e@j25?{EvGfJ= z@Z+M8gq1Cj=+hod%J*;|2sn`Y921(l^BQakJ62Anfel%o8@W{67c zL4iW4Si~c2HGd3&_#{5k1gag19HS-Y82~H-lA8wS zOOSkkNHrO9)_WnB#ew693W3iR5P^o~RDK)sD#e`xNh?*DeMszuJXM19Rwrpoi?(x~ zk}n}XLIOmAF(7)51W5X5t6e<@s-ai>v$p~=0?E?dxr7E<_#SGjx0%u_usn4ro!PG= zP*`7EMPP6km$}9Y#%8uztprHTtZ<1wWpRR>sS4WD#;63><$sJWad!e?L-mjDNy91xZzb^1H&yB8mP=PJpIl6(N$s(X3QQ(jzN_FqcCW zgtDsIvNakS1ZjC(w20JWs0jhe@t#9NYZT{OpYV->R>PrLm%=sD(4g92CnF0F*M+jp zT15m_*VtV#y?A8CYT@u3U_Q_j3k|sw>c5DbEITLf8*Q5bAP^7MgQVA45b1 zMeksTc4_WlSAs(;fR$d(40kz^#%uIVsF5KhZ0d|91rsThoV!GKUa`*XwL=qRq$=sV zif>9^H|nLaE;tnnmL`rSa<1OgizGHp9;1OAznDOUi)pi;y*8PsMZ%K9XsKb53S2IP zD*!r_&w!a=%4Hi#nDQAg3rxpy-P8iMoCnSV#vMypMv0mIwwbP=Gs(-5BACL0|oYh@*1YvFk^ z|5gz|YP0Dnho4K~RWS#gdKiY%R$w`#R8YD;5l0f;Tnl!;7Q&EO!F?I)xF4{>CD-5r zM`0*F8(}j0ZrL@n(ee;0Hk^06kF#sS;~&J8L;eeW{WqD z2bwUd-!+uc)y@SY!6m)Yt0EDHYuh2k7XIdHsUS^={AIxT4x}@oDnJj7;kYLYC7@a- z%kSABXwiv`>H+`;D8ON>B3${9wBN%~2E<^#zV?T~)=tJ(eG4Qo0fTXcJk!cN3_NtI zYZWXX9pdf^5}c-N{8aVfA&{UTDv1J~x{?IO*nvvS1+K&leFiIv^dfWRr0X_hSGF^D z$Lzs8$V~8te8$73n!bLr13ae9677iW6fYpkYsAzri6a+t$YTQzLY2#iS)kgtd43=u z$udfo*&j07GZRY^faEk_7MLE+$RGtYnTN^($5K9wT7Z++h?$_u=iHD`8swf}z9Elp87P)sJ7LnY7Ku}aw zLNQ<^-=GpL;`jM}Xe}rpJfP{7dmJuD=N@D#64P*jH^hMw3=|0PH6TJJxAi2vPa-6C&9;X7BMhzMsrN9A~7>TreIu`A z+kUHBuHy=e6FTEET6ht~M<7)~J|lGNf=|kw9x*r2MHK~8g_$iyHiHrpcYGD*xx$%c zTWV!6Cjwv8C0IDCdj?HUY%GVkKiv-7+JNW0R8j94urpk!#SMh3Z2ZoAHC&j10jORt zASjcF$d&n4XF3v2p&(E>eI$%}k5;>fdn+CV4#2$+TN02B;hX59p(6uW5LW^u;uHuo zL{4-sdI#&u-hYZr!QJ322_rh9*FgFxx?vR9pNyo7zC;Js92MdLNzpVU6EBu%9?7w| ze;pD!zEQ9usp@G+jVd=ojetIbh4QgWpk}Cg*675M3J4>p;p!5SkD0XL&u;}(nOm26 zZU+#$6&=C|1^tAT23v)i8e*yin9X3cqe%ztHo_6;UqoBf+Xhq+oykoBaa%;03sUs$u1{uC;BeSrDuePvmgI6(N$`njOF)|~T!`S(B zH6N!w#M{+UF*tD?bB|m4JII(BBU&! zm^7QghfYQwe2y9XdW2JsPMJzy2l^9EJT}ahbC)lj6DY%voDd;u+!-{;M-B<{DZc$j zDN%5jwjHtE+(q;klB9UjvQ)BG58z&0{$V16Nix8mXG+-8(9?GWE2w<{~lm&_hM#%C> zR4V~OreU(o|3D7voq!{k;j+N9Kf9_az{z8@EYK`w)_MgnISiHwmdyIMgeCi6SztMs zhb2IOlh1frAev2KmT=NRnqD!pn{53R@*Sm}t0XVit$!ou>sLEStcWDt8IdOGeSSle-ram$`s@f=MjY}2`(tzM5mGIgq%(f8*0^?|j?Su|-NcM@7Jk^17Vif3@|japMej=A}~;7 zd*&QjQpudnp_1(Y*HyywOsms^nYvRUGp4gV9$KdL%t}mx)8#ZEw#$bj`beBQ5Yy(S zk%TQ~Ii^v}muvq$d{Cy+ZU6ch5;g1~_MK$}Bf3u2UWkA?TJ=KBmD*=88&vGL;Mjyg z16u}e9JN;k{G;2%fIlAbLf2l(hZ%@@<~3p#sJ1EIBxYrv#Ed#fDzA^_XNFitymjO< zq5(epyLl4K?TR~1;I9JygdJkglbDna=E2Ryl=B%c3q<#A_Gl4MY<`SQ&wP-er{YgJ zL(!vNJCYvN0`vR(JZX9^&ABTuoUY%dv5I|gD3`-tOg*Oov%vJ$B%+5ShD6O{=Hv8( zQMDt)*+?REw01=Qn-@`0fMOpIp-OVBk%6C!xy4vzfIxtwc1R3CG%u&>7Z4c2f*XQ> z5q3dB9fD0aP}%^Iv`Ec{AgP5*t}8$mKE!LN(IAoU<4r^5;}7I889@iR3C(WA+!KRbsmK|up;Gi(KL$rK!^%`VGW5&`pO_4 z4p+Wus>7vn934*MEm9}}U+b+$#P?BgV|ByBWY#E1%0?`1ZBK_yx7|bSLiZuH^KmDL-u9}$I+;+lrbp3T`aC-4*AZW6 zc3}S%41qflUco?sy5E_ax_{JB_fwWk$8QRvZqY3fb^pkLI)VVXVB|jUAQi0~I;YkU zI0&L+BUmYDciD2NMVhvcBLo-kRVmc5%f>ab&XHZMTwP^KN3N3^U?qT&3}&t-B}XB# zmrl3}h!!MHHA8Xg#OlnJj9&>;b@cj5m@NpiotiNJg(J)`1!0PAjxfiIWdAtGt({2b zy7h1BlAXR7u8wsRa3i9ysJVRDP2f!MN!$cC9x`=YJsT&A&m22y^y1MCEUDDKkI*<( z`w0F_I>4x2?~i=RItjjqyErF7|9LX%nvg#j?n_ilMJ2_luy(fl>zh!?1Hk(jk@KjF zYvY{}lx&K8a9$jaINtn1%7ffGX0wuNvy{(FoWzvdh(Ps^Yr-9;;Ka1KfAeUutKH8q zNX*RLRv4ua|JWe8XoWbrQq%uDt>kJI<@X@bKolyY`ZIIyCe^&70>*q#sV;czO%LR; zq9(+k(t{I&yl8W{Ga6Cf?6&b(G(vDaGCna1ZcjZ%{u)kwrIUn}XgW@+B?N-Q_0wjR zyVs$kChDFeT^)h%)K6cAiv5U05O}0$h8)CA#lO)P_qmZ3z#cGP{eW63tbIoDoZpCS zgz*9fN!>s0$m!Q~=SfV9CS8TV@Qupa2>I!wLh8VKfFhEstk`?%BTb(P72bKnf zpo|6QBJSjL125#Fk}P4XzL3D_k}#5%N4f*OX>ho@Uph;K)DFT`5alBI(u_6@mbogt zEzrp=iO~(|n5Z%GA#VibD(UJnzorxeu#x?h4hz)^%4XL>z)orFfR-5UMQg5ag+-CYi zyDP@;LE22BTawMx|I9MDA?LP%rUV10sDzOI} zqFwErwXbIMCPeoT6bg}IN37xgEk~2InL(peiGl1}MC-(| z6_4iA1=0rY;Ka;8`ly3nZ7MP*+2XAU;1PmDqUJbT|0=S6;<}P+EOLxsC!SbB=!=IK z@5_<##xLwUu&_Vn=)p{e9lIa4Sbupfh=cs)I>VS{oh&kmW%xf+Y)jG2ZOidrBNay+ zjU>+x2SVAh>P4etu)b3lflxQvHUoM{gW5#mpq%=UM8Vks19I5HPOd)3vf2Vp7W>1L z4>5ZpSz}8ywfQMleCF_Hk_< z4NHk4w%6q5QDP_V0ffHfxZ(h`;D8bpBG(VXflig=+SFX;tJ6bbvQfpzsNzPQokg8* zB}VH8v$Kx7ee9J4ieR&^ta*bm>?87%SA@8`m#>R1kRkEm7nYp+!|(8s1CftRJ_LT& zk-3?n%KH)SgW9)W#aqWTFpBJWZbS0B7@aoJ5VzUkH;aG~+-7Wb@OG{X6EAii3Z}G$^1S+>{1QBzF?%%G*z7XX;}w zUmz`ueW@gfs&5j8u}J(K`b0UzFGLFfe!3Y7`RA~8a3h+cW7s#3S!}@{$2J4$_4IZ$ zds&E|H;Rnch2!2g8`7~m%xxZvu5oPyy2dpFEFaAcU8qQra8nDMrs$L!@gltQU>9ucA zBCt7T#-4VQ2;ylMu<(@p}dD)ztA^plU zw?lc(?HVh83hEFKV`iFccc9hlh9z3@k+`=pUrSC~@6Ho1m&T~UQCV=Z8Y~V2(%-aK z^zDXwpPo_Kb`(k(B3P?MwPs<%fytWPE+`(1S0gkl9b{0jXDW>Xk~|kGEVpO3Mk-WYZo}MwPR}G zk*WzT%6(O11ODC;smQ282Cp1$M#GI_ctp5w6)jkY?TqRJYNg!!uCz_fH+#7mXcb3m|%4hXjJ0m)$-f!_rKAG!j3@o}dKKD^BD zHcc{O6J{O?J9a&?@8bAFOq)eZDw`!-0Y?slrG~`F0f6_BtXoKcl6|Z!kQ~ZoFAH#T z8ZZk?4-$cGma_;@vW=4kf&+r<5MSC1dlP`3u_0Gs(15|B71>8YnzAMri zH0Re*AGU=^GoNd@(ys;K0C)!Z_9}=TRbUa5htrzQbHBijNp8IS&(qdpE$ULC?QAy9{vSsh6=D2 z_#Ov{5Xh{-|7ZzG*!#uIjNuwL=923WSJ)~k3ASbc3B$VHnq(%gS1TM&vC2_$G({M! zogA_w$%+WtPSzgu^Scp=VzA&C@Z_AFacgWWB9j|Bik{pS^ZJ9xyhiKsst(^CQ^70+ zt;rf^f8ch7L5cFwiQ0UoIGt%=Yz%QySWKIPc@VZ(<9x=8p@H%62wCj(R_!tY<<@;8 z3*-lKU;_fkT!xE*C%Hzy?Z!>m%|>pTf`g6CyKXlK>bY$)4|K!yleC_QfpK=IH|2_f z7@LMef=a@jd{hhkx}(~4>)+`Y1ye2>FM6>X7D@j{AsKbuyzuVdiLlz3vwGO^KIGtc z4)P=p{?|XH4>(`N33A-(!{)dc@z$OZwMS?%rfLu2PZH<-@7JyK{sHddocH}t;pj9M zLXmb7G;y(czzI@hhf&3p0=(#PsUZ}F?xrPCs}nHnk~FkRz_z;405co;MGPv|f%l($gjxJcDkMg)xVeW6>A_Sy>b?8=;cG%FF3^Xt|Mu|gQoucC3=qkr0+K0%< z_u)ej`xOXvIK3{_GxDnY#k zQz1zcaG^4jpfoKmFapQ(3e{GCrIE_>C=R8x1Hm(n=Q|led3leFwfI~DeiPGX@1`*1 zXlQf&W9B=Vc0NCzc1|C|F}VoIkdvy-=|S#pt}&*_1ybF>1-kaYB;&WevQ7nxm%6j+ zOf29SgkTw=F?0oO1}OCb6_Zm}5!UR{v4y%eDB_C{GF*1AjMtDF%1E1Sc5$lOr0Wgk zf~Vt}(_SeYfrca}=amcn~=Gd_AbwMfS{6K4=viAg>WYNMDQ){$Pp4tJpbF>KYBp{&pm6A7QY+ zjSMzk;nbyLzu*YdOokQCZHjyZ*yc&hwBrP8Gn!Gm(B`QvqD3*uE&D`jsD;FEuidHX z_KJXO(?i6}I}?%Jd{Y`j?*t;dRuP$I0s`E~ix?orlbk?4+_a>7XJIX3VtI@f1I^%) zyjrp{v-t-fIu0=S9An>?=L*FcRH8e3^64j-j8hU_xBd-uP83nv-!D>Dc#+w?usXlq znED5;!9zA?XU9!qW4<%i#*Bm^ef>N9!%)<^*6N(NyQ6p+W@Dsk$A1;1|vt2?!Jus4yX)2O!oXCROcZ6K*9^;j(QlfxJ|R7>vA9_ZRm+I zG$hias%Z)CejV-!N6CEQ2#f(hyd025(x{!C<#`~Lk8Jxq4xb2=dqZOl7z;0(MV+2a^c0tqYuUa1A?*gx!P5 zQsq=doOaEP(8=Z?^zn%bxa6;2D)oR{$hOeiHsGs4kcU8}`ix$54%tu!p>t#`R_LMi zw6%?xBGku+cZRdx9$Tg*f>2HuL}X?K;(*I}U@*U0~ONP$^@7r@fZv?bkFnAVN-!lFr6ceA+GHwKy*rG}AK9;ZZZ2O)gW21xV8?ZZTpq?;Jh^V|zSDF-A;3*I~23_e9R3m4Gq#p|ik= zB>IsepP2L5W?8=fbz37pzBN}3Jy2fDxJ*5AUuEzymTVbn5Wiw+{PWIZPX~AloQT>(F zvud+MZgizU>d+XGKd6*#sDDt+5WL5>1U(>K%YpbF+>U|zqqxqc+Vsn0rRzc&Ucwf2 z(E!RHvL3e2swy?4*PHtStkTLFAK7K%g8LBW(+mm?7;lt6h{J}YG93ZUTs+7zLO_t3 zDQbp>REZJRgm48n3HCbVSs`Nv)MO32ZtB#^y~`-To)+g~cWtgeDCkvh95gG(DTy&{ zPUOAZ0+#&8P7SA!qSh6pW_4dOIur=GXj907xB5%Ej~v5&X}=qTp~*b>ubiS2$AkOt zKa47^;8X!!9|BINqoSC-kQbJiA*^M@pxG=26CmZnQ}wYm8z0Hsa)k>g<8~t|K2MBT z{bLP9Tq(o>BY-D^6~dsq3LBgXoK##1jNqcr!>(%WUSd;IwYwwSYP?cORrbCVhikV) z{oCU_ASlJ#EUpK|azL(#Vznz=B6r5RL^$LG_FdJ3ofkt+66a({;v~8C6e8SlaV$*m zST`~P<3Qolu3l&Ba>pp)30gWto=}K9jz38pBY*Xa)-m!pcM*<}4?$FTN=M^zy@B9- z7u6Ujpq}y!K~X0*a2p9?b2@bXqL5<)B@EY4pR);j+nNFl)Ldw6w5vW>qw~-CsL%~# zIgZqoHtM`l+@bdlRJP-oUozGf9_zI>s5U4Z2f+$#{S>Uj8-==os8I(GcgjHw&^^g> z(C*<*<=K<7voN2-eynNV(?cyB(G#^p!kBF+~(tcv=0NPomGTtOTnnWJ$l18!!f z3JA{;IQ%`@hx0s9`W)X`OgNz9zDh-%4a~ao;DNjDnk$@yn_~`hs8{_tl4YRIA~d91 zqR0*^<_D8H;!AWa2qj1AX=7ZoRTWxb_y(303{Iptn8(_PQz)PDu!-Ll*~D=oM@*Yy zDl!ohPuVf`427_8-)4zZOn38R#Le7U?k+uP7zX}j8A6{Pqd-MuO~x};vex~L;|yl8 z$XZv~W{2iv4D-%d>c%g&K;@xacB%k0rvc-T9AlAla0{3Y=D{}vNcoJH1)|4uWMTra zT!+m9pJc3-u>109%;-+M7^Q`a%|pj(R15Ea_J}$Ha1}k0*wrbj&wd;x(-1VEjKg^i z%@TfBm^A(M17tM_>II5xN2MMvhh2q|$_JYsi;gcB>(_#?h+~;<5u=D% zGc37c3pAYW#HXTCajCzao$2hIt0LBLwpDFW0bM1dP$>P8+#nKUlH;IQ#-EAsz&KVF z)8Y z>}R^1=v91x$#Kv z>R}hMv4@{5TjRyqYPZORqP)hM`it~bv~Gb9km!wUxp3WJaZAQ_NZXR(K_h>eYL!Xx zyX2Rw=JFSo#f~Qqy3s{+`}2g|eqykjSaP?dcag19Gw3-m*A}ic;sWH%Cz&TR)7&HP zA9#IiCqbek8USC{+#E6MrP@=ki}W6kHj)M}bJ#YY*3rxio0#)9bpl=Zo|`X){)qn@ zKiey?7xoz9RX-fH>6cTrJ93+FT->FB$jGphIeVp2dypw#gNQTT6KS zLlB;NIaPamWMY$d$`x*>m>Kg9J-+DCC)^?AjEw8e{z%jd}Mriy#Y0$bX2CHMqu0_(WP9%_hj z5nP=QiCSi_YAG`W?)AbW} zhyigjHxZK*J0&?cxwq59^)G&_TUVQ`RV7Atty|SkC*(-P$V9haP1ubXndl~#$1$?2$zqQGtk=ptun4Msfj&4@ z`(IPT`dSC8`?{m4{;lTpL(QggQF&>KQsn9g*O6^J+NCq(G;$7@Bu=C4nsw?KK(;hg z7W2lT=PL?5|A;@6jwHJN?XnlGYV{@V;+#VLFKRnaIc(mE8&-7zCfrekbEBahjv;<3 z7Z$`+x*N7jCkU~Ea|oO97FceXM4EPY_a`=unTKO>qfFRW3^D<2!j9Z#KE}v(=b6Q(j|`x)@9NE#!%}A+ z#qap>JT~vb%$(X|gXK-HHeC&Oa}ONY?it{FFtQ6q%Py)c=^IU1(%~t|>Z8{$}H4#x~3#^@mM(}$S@WV~?>|_Tgr}pftK=omj6K5)bXx{|YHh_PFDk|_0*g^Gh zgMQ9?CzS642T;31362E{yio1S6ud_O#SGKgK>A?=It^2-wB5a71gZoF8QBE%x)Pz+ zI87j?%~C!KAtssEh^b)`eEU!?bRmGsX}~Nn9n6DD1xWdfmj$9?sx2d?p0Z=y9GxC} zCGSLYF6dwGP%y8jYTpsrVdMEK*|=Zjn7!G0irL51j5Q6CGkY<*V86-F)3=>D)PzR} zskL1HC~R%Gb9H@veI9I1z2$Qi@mRR*y}JOb0a`97Wpwb6y)HVQLF(`ehu`6MSgubd zMzuP8XUyR{ao+MQ-tLy?6A6z!(q=`sf0?iw9kQa^3kkc?AuGCxRd(9!N9b(TN9W+F zN@pdP!&QeKoq^^^bSL0|-Hi@8+!&-v2_x7dlobDWCh}Sw_jto!l##dZHQfM%A1(;&O921@NVDP@eYKD5Out* zL_g<|%7e`|c$RGXzR(O&mfrM!UluN?qJtZL*1F!}#R zcFrJIe$5~rEKgRCVSLaC5}z3AGF7|k&4WgP(TAVn$oCtLoEqDbL*M9D+yKc3=-r!I z5dq;Q2Z%v-9Gfz8HD@{6?A;VGxUI9f{xS2Nyv+pC-8OUG`rDOp+88a#Wat~6-@G(d zZT0dN1h3rsG%ajpAC5F<4NPL8e8ZEftgO}!ZJ4(qc zPbtFmn4?smwjs@_E2zUQJdBED5YcFi2XkJdSMOGfJLD0?u1 zrcif=!W#WL)%)Rq6kpjLAnDX8mY&KvVYTQ>2&F=a7z(wj1XRWth6#JM5Sta0^&0sN zCVxhKOR@SvW_=}tXGmNBvWV}S*9eJEjC;k8rwl(P3E3e`GTYa?#W}4$za{=C&zOQd#wmgO8ZGH*^_q3#&1JXMMSrgw%eFp_x>}>Y|WzRKqp73 zXurHj!lG+3HH&VEsQLTO zoT6}T$6M)?<_K;r)=Yz;q6!`^F(`X4KK}!!Aa-~ss=^7gta+0TonBq4qOC zL?N$BQF&e9e9;-ABnml!1(FUMis*_2(pAO9N!pAvNYGdEYz;l@Uv5>+YuI#BQuBb6 z;^3DAe6u6qe{WX=1pFTFBI3xweo?0Z5;r(WPT~Tgzhe`W?DnTm>2iICOm8VeJ!vG_7HvEsMO zCD^V>55RYow81xd&{d&u7z%?YUY6RZu#O4%U21RY^$%`@f`iHI2KmU5g?x%t`bx@u z@dhKkNAM*##+%*7_TJUv?8IJp;U!Wz@RiP-`j6Tju|2+I5ei;TVj$W#U3s9bL@}X$V0xD<(=jSWk zMtz2YD^W#bT~($cYZ~3Xt#i}L2x>7KjWUez1v+`mBi`kwk1o!6D{bX-2FF4L8YC`} zFv^BBVJGllqr#piTN)fOEsoaVAsT)ORcolA1@7e}o?;z2#G0(#fjWg9E(}6dGCJza zSCz<3nUTBL*k&S)P?mjw56&U9r_@CN>2r{ZDdrS9Aqsfl$73kNQ9RNtF2j7tvqQlS ziZ6Zi#FGolC!SO?G&!MSh3CtG7{iG8-)>i^SGYDaCHq-0>?h$uN^W$k<3@)q(Lrpu zWx`?$XUfE;;HYM7vl}Wd07GzIRf5$$aHw(5FqoehKV*2ONQ@Csp5BJN@HJ9pQnk4L z8d69E(ffCpawl751+g8nTWT^N7u}NZ@%}r{oZ1-jg>WJ_>r&w0l66Z9m{F#P@v4i7 z5flk=Q2Q_9b)75uYld^_UzTY4=JizVgNb*i5}w$E)qxF*jUJSUK0JTPvEJ%9)(0$l z!LjBC6rIU1y1A$A2^M6&Z1~w^su0uuMmz1vR3W-0qKa=m1G|Y$@ONjte9HK>woy9N zl|@q-e~eKRM;dJJd^$%D-8p5sM{zDXfoZ!8SncD4@K?(T7%Cck=oVz~Ine#nDQU_I z5jUW{`ON@V#ne$nu}i)os4*vwAmO=P?SfP( zdP|!qiQGzli}eiH+7aZ%9FaWzVT%F;9Ffg=2aYd4&KHBjb%`k2IHJ6gXK3hA|8nbP zUQg97xh0*2V!NS23$vuX8Ap3xu?qm&<6Cfi8RhU&;KjjvalQ{4Hdshb!vcl>U{N@k zhDEnTG`zYx zz+OOZtfEa&z*PMAg!VV81+cgyLZpLcxvTUP4Yk|fZvE{e3|5Q`HX64|th~Wcyu^eg z`nw#_n~Aa=?s8yDz{fu)0<%r*X`s#gbfB0g7-@e>NUSi^+K6AWBv4(!%j)K%e#gKy zS&|iqyH$fSDXfm7PDD3H>f^sg0q`PNg`)9P%$V!ee}f9XD`J^ZI<~0fex&w_phVq7 z7w&MCV$nnd)KF-m3;jzHnt0n0Reyqn;FI9*sL|?`RFy5rDqN@45Oa50NbYTm3y3XO*M6G(OEhsacEMV5LCA(iyLtq<$(*%2S^`(VNVy|MXo#>v5NIB7VG_*nY36eRux%QtSYjq zNY-XzShn>*ZeT!Z**~1bu;4(6g9L$~1hEk$vEjt|;l%O=a*zo5Va0G1Btj6`27*{n z@*T3sV%@HL`@Xto4QbHRT_g_=&-u=GzW4b)8}jx-q|O5R0j>$@DbB+1$KyPN{^!s4 z1>#V;!r&|d3=oD3o`VpCD)nIj%5!9}Ou)*O1a`(pv5u7>SLn+Kz0cnNzVdIbYxqHKI z04((<=ZQjqyAUuQBq$COGodUUFtaEZG0YLgR3T`-_nzMsq)ZI6ogV(m&3c( z8JJof$QKo41dg2lp=un!5)1<%J0mv`l-7d-^B_d~Iv?nVYlw06O|VDW9h?vWIXjj* z3P1`mT>y8Kq8V0i4Z&i8?ijQjgAsK-fg_d>4(p>C9NB>6#E0y}XCb2O4W>XqpWX&1 zW7PU~@4+7WGna^Q%rJR6|E=+-P?ooMjf+b!Ud@j}BjlsQHVnV&Go0_eyq#k6dHLC? zF`qc0AATW}U;w6>2ic{u!97rix1Le#xXc&gG$S^&ylQS4h+wK(ynxmo-B%u(`bx5%|~yVmD8X^4h%;!>sJXu#DKi9;(Ot%<6H- zdS1oBV&lH~JEul=Q;h$;2g3^dwv0D6*W15)YR)&su|Iwd$Bzx_+GeM|e^Z!$@%*M% zgjeyf&&Onw*vOA=Uq90CI+B<7R|kikJeqD6}34h-r{|~?{lQi>hfFG< z#-KAM$W3V{toOl=X==0g!71=QL{9sve@IM&vvMdM?dN0Pk1fs7ZJ4-z3IA`b16Y?{ ze6Pe(}cO z2~2iuXtKNAddJOQC*k{QM2@|7K;*+~7eBVoj|ZlN_pSHU_X8`>`+$4~s>TzuHGTQe z^4tP*(6LP@Y<bTH7|#O(CwO0&~b6skE;A`32TftxixEauJH!{YumSlpW!pU6&R zeEL=Ym6-7<4dFLgFqI*T0a(upjEI|>jx|aT>rWAdIn5y$0CK>c z5XIOX^4LBNN4yw8$v&9+`Nq^&8x6`0qG)I6#Lr3)JCfuV3$$K>wZD*Hf7ii)4`uy) zr;UXoVf=Kg%DD-d4B(N4Qr;rl{(J9L#VoA<2n8(zyHU=|uqJMB4bt@4C`h7%oHYBa zib37veW?c!X^>yY@|{8`MBD|w6uFCk@Fd#Bw(b=TU>p3Oms!L?dwnaw+E5-1FAqB^ zyaV0*vrFB)S|tDaVK;?i@I~^k++dHyW_F|Ir{;Nr6^7#og>_(;e5l>rdN;<#c#lqQ zJ@Oda6CZ*e#HR5u{oQ9qk8VPoVk5p;`#X>5k}`LSB!-*k2TndbY^BrBVPubg;-v)1jqHYt2W<{zk*E1VEf9ZcLpnLyPvuTJBW?(UftYh{Hgo1q}VhbqPP3Z z<{S6pjGe*(DGsUk?wLtzmUX# zEa{)e55Zi2hUDs~@<=|9pMgO-PndFf02wvltRK`o+swe`9Gq;1$X{PpM^Qzyu?E%0 z;V_M#-TQG`XgoRh<8;6sf|M9P>0`_k6-?p~?tcLGaKOH3{!3TJ#D98A8Euq{Z^zM{Sn^ks zsTlZUWa^N6xRKQ%bsh?m14HSdR7RBYFo?v9P5^_pi4FyWl6kI#`3H{9+=9*PM3Vb2>eH2-J$d&o%(Dah4 z0LsNgcKnxOR1XIAn>Y)88svRQ-mur+0~<>L&PH2nRB1;-sSZJd2cjFCc7`KQkk|oP zxc8=%s2zvH)KtM7fTvg+Id;13TZg866D+oJcM8Ygvv;Pk0OLb_!%gE1c@-g#^ne5B3@1KNgO;mNHzf* zRHo2C9s1Z-E>zHOWIRLJJ#5v|!4L`W%u%?SwT9+sQ4bt)_9hoO#$Dg+cY!4RL$o(y zSyfNDN=~#tKxg-vOXGFWwgY0mmhMl? z_aLZT!rM^#h&w^8V#l$vfzzRP41Mk6ybbKkq*eIQ7><#*HLw zdolCNfrCOc*HAv|r{H0VE_@yg0uM4aKv=^S;!hUIC_&$lO3Bd%Hkc>Q|71*ugY&T< zS)stJZ=j@IpzT7|9a{(S2`TZzg=5YZ%FxGvpk@fvO5Eq748k+^Vi3HzKa~PStwFjt zPbb1q4hU)RLnDwO02Bnk93<<7GR{FjSiF89__?z}FvMun{Pll227gqe@qyV_th0oqs`(Qda{>3a5b)9yk$Z1drN;|iO&60sG{=;`D(0;c$H>35B#j z9ZZR4c=4$X;l$Ftnk2^n`Sty>+=Pn6Ub_hu>3#^BRH`gA&L6{3_d>xZ5rsnG$=A;x zr@+6gqT=63YZCUxuEu5nG#iB@*$|rqn-cPRL2MLgMShR~JVPnKp#TD?SCD%1e6Z1H zeilIAxEUlC**o5mz6q&o;TwA`6lr8&j6=HV!DkMnVFoc!1w>}{?G6w9AqRKgf9cI| zF8e-TZt(0pzpoi^Nc0sXlZbLW2g(+z2{1d5AD^-v$Q+`VSCGeuP<2tz+aJOC-f%w9 z-#>z=U+v=l5C!U_U!waD&RDEVAGir)S;IRq#xg%JmT}EWJ=N+h0X<_v^i& z$L|`yzDLCnkFN$q;AlA#^3XT@DaL&(_5o1;@xF6^7d?O7U^qZ*v*q#HmLS1btEce}>Opw=<9jHtv^wdw*Vr^0C49%f4&; zkNcS(3^>Oac5wz_gtO;5nh=nIKLHiJ4y3^NP4xDA#h?Arx5C0%-`e+gd0-cxL;W20 zaPf}c_u5-(;59~ip+BtXqpBGEC3R8&po+BF|F_zoKrz! z2M_~;P3&hW0BblT3Pa^B!9>-LDtwIh;y(c0~DY+v=j&4R@H^& z3$Z{JkcHbsx#Zzh$dmvUY>F<0f7JC{ZfTH-w93lQw^ZbKz0zY%%A9MLEfH`b21k}gqz{LmVld~lx z6j?uZqY`MhdnuZ)zgfQcl~->T<9A@x#{Uj~A)#ZxK|T*J24?v%&)>je=zts;MTBV@o>P$N#Di?1*2h|IDuwQ)q;^GI*0T_Tl z2Ld<-`9u)OdU5fE%Kq~3?&FKk2QNMkHs4!-#hNllu^ZsBd2s>PLudfk$TRig;+r3y z5AHJ*#jq^;%`q%bQ!IJr zUl2UQ@FaT$mb@?$0B^v`$3SGDa+|UQH`|L}ym%X`#QU{@Ui<=d zYjDsA`kQ=LfVYz$`sBsq$&Uby`{VF`jMiOzVGk}9YO$19m22Lb)JefJxr`~ji|)iV5y*XNJi zM?vKKBl80Se-x3);3fpRLna?^l-lQG2&6Z|YWJ0wN#XgU%47T~@#N97GXanhs4);) zAhP=s10W$GE1{-BsvQ(vv;REszdc3wDJeS0`2(khdC&l_cn-kq?3W@WZpOFXJy5BO z&$=P4hF`~o;%6?sc)XHvq(Dpe_n}w7fnWRc(LN79?8{Vd1&}*@;v&5B${~IZEkQWS z;Np0o_CpTzr|z-VvDw|Ln^%J5tpKS)poEvlVEMU+1j{MSGF;#gvrhSGPqbm~wh^Mq zUEHyQ6Hq-SE0<8(XY?_Az9*xnUY2k}rthm&k3S@yo|8|{sgGag_jh_X0+`SC|3kor z%c5v}!bnCw$@c$RghqE@`+w^{jmW{T?)w3@{}uRoEGHR{qgw_BKS75G*_J`THR8iQ zUpA=ymCuz823@`VA#qLu+;hFJu^w&?JzoPy=fU93Lz5#zf80U*k^KU}G zfx>i5FO3pBcu6|iGrSNsZ-U`<6|lbgN&{#Ge)JAV4aeMXf2<9axIgFz{>?}3bNmD1 z(NkcM&xsGnN4{AZev=+{D^AfGmTvM>DuTzVht>MK37#TYXA z`6=DtWAN!EgejqOdcmY(=fSwUf;z@b?%oH`ynnPu$dF1O^Z)Pi7c^uf`>SKh@w=Bg zIL7NZ4T;GC-}+yojUNucmjk}zX*{ziy1K~_0o7tdL|4Z{{LG_Z(}i=q*ATzU$H2oE zF7`KWGEmqDv4NtiXE3;_*o9^GVL$A)sV_9w=3 zjJThTL_+v=4CKT^0{LqUi~A4+)s=YvT{mpf6tEH%@^6)e>)va^F+p{i&-ZUf`dP!P>PK zeCem>$*1Ss$4_3CeDBpq1Y>{4tN}6ZBmu7?jJksa{M(=V?$9D^?)!Oo3HU1?63-}( zC6Q(6oOs653{T($b51_PlRQlm4E-qz$1w=}LIi_}(fAMcSB(1$9)R9x*a_pFaX*2Al?Q(wADJ`?`IG>bgNJ_j-d2OffO? zlK2ZLn?Z6#-$myCLx2a0P4n{&4;Nr=0uvG(Gg)omIF(;Y669!~LIw6>l;i-UR50kn zmr$Atlz0NAyoGMh{E?%a2w+P_aVFsyQ5Fa|vjs6~`xt+)??Xko<-+u>{O`R7^FrAY z_ESCcF0E~h?NP7n;APZ60`7khQIN_6&Nd8=MVYq(P=wqryga5}zkaEU;|+9$6#jcy zM0~;__)_?nUdGedEKfZ>F^N<1VfdApe3*WK4G5^&SwOycI#g7xfOS5YXEg1VieIcmnL^@G^Qox&y=c(XZaZtZ*%M@ey|2mdnK; zdXG--0o#kmH|DbkAm03)6C-=IjuP9%_ta6`6FQ2&Bhf}X-#-KZTr$;18p|0vX`q7$ zg`y?K8uj(RP{IQDL;Yj$>_ZQa(a)C^oKJ56(Q5fyCJnfLC++B`qfHW0 zykpVi+OUA{XhZ_a2m5{=UOPhm0q_ufiYicSnw@Bxm%wzLe;|a-8&QGgpaOwW98`_H zF&`DO36+xhMx6jQ1F9oOHn?Plo^K!nUxKLx3eJAhAXVnF@^CnI`;y)OrEkCij*82z z!KDt0i~`x(z-f$P&5>aqh68!r0<{Kp9fS}d72^?doFU)Y$Dn@geGVwqp0%_zRAI+I zXg@|(TtNLma|v7!@C2zokdhAtH$C1TOh#jlj^=}MGSxgNf9D_P1s}!}px)pJfUr=g z^Ww)h!n_0TLEuF3`@f{wff@qJ2>7eenIISwl=?v`<6#t$ev*42u*Z1DgLMgXdKc~V zm$>Q;M72M2>Fs0mzofwU%=AmUh&Y6w_O+MqkC^EJpJG!!`S|+b-%PCfiw`z~w?nZi zHtfT_Q)4|LYKAb0>011qH&JXL_vqxr$d6UsccQ%-3bC;}#|-J$Zv*Wx{NyT>5g&pw z#9;I={ha_;Oj3O(qC*j^4tFA0UBsKcYy6$tK<(OHro9>3cbLc<`rrwmzgpmo;nh6_ zPWgnuxsUxsLgLFtai}znLKB1VP$&?rtXcYmq5Q%7jTXii5C1p7t2>E@|0qJpJBWv0 z{wom&_=o#`9$q~BwlN1h zg`cpn5*VrNbCy0IpwLS6LJ-Ir9GgZskIc!-1&C|s{$@aR6j8}x`^nQlB-EoUJ)iC% zqFrO0|0Ea>-Zg#<{S!-Oa=dnz#U6BIp8Rhkq`xPv5pGOto`K$)M1OJ5@KbKM2k0q+ zj89KU0Lz`|Ng{&h9q38w|B68Sbl=Y(GI|n&KsJKFF-~4fPeQQ05qk3bzOtpMQ$$Gi z3KQ~D4Lp>C9Ky8b8akR1Pypm;LJvS6Gog+rq)C8;R+Du=QjnGAB_b3jBFIyQ62c>I z+5QO4-f;#BA$YFl=qN_aSp`WW*53JJG_FA1LlNVkj6wm@jjw1zUyA{%@w!hQrJE4@ zji?)XC3L2tUQm+tFai(ENWL{5Rl10@!aw(I0fm_G`)x8HAg+2u3|`Vm-wqco6rK+^ zB1&{|xCfnCmpVJ%6_+V>4=8p0-M2&EemD|eLH*IA1wKNP7#Oc>bz*Kuij*-xL3x1* z`=!ABpS#UOLe1%_xl|q)s}I0`=$=o&?A3yOY})r^w#EB0+h-xhM#sD!vdakWKKuQv zfFgHdzkemdtvj&a!3WWP`FHmHJUsjT4(`*71CLGcKZ_=KOh>Net|3%jA9p?Mh>Pz9 z3nUz@2mkZ?J`0Q2iw~apmfh&t;k4@n6^cI`Gr!5f_b*8(mkE^9K(3O&NE;I7Q;#BS zHGT)d@m=E|!#~#{EQB8Wi88HWgkt}z|4Qtr?6>lYpAahKwn4lI1cQSS7ahMc7mjat z_D{U8A-o4{%t2BB7G%=aVe{G+u#hbd9#kI;*=FoVFOj|Pj`Fi6Dv zP2m#=Y4@cSWPuK375E@X7ZyaYKxm7qfXo4L=U}l8Qo`)Vw-^33%C*}^IK{QrB7upu^qOAlT-i7y@_QaF#xw~wdtBM&*1AHE8AW3#%t&4~$~ ziW&-mcOZMvO1S^#h~V-Not>D=ftX%CCX2`s(&b|qg2+*DFz8O_#Pt*6g%B78kL`x% zAJRwK{L}LfDRhIt{lDRG@yBq>Z^P^xetdq=`GrXOc!OVz7Ruj?0wC~pwEoJYYTt-3 z{|>^*&;IKX=lH38KMyaQ%s;$Ar0zMi!QqHGfsT)p&nO(uf-|`PlTUIyoD%=|47G_x z{IWQN{w8_wb4S6yAAgGe?iegv5iD;Z2Zal8APC1>?Y0NN!@YOt?;sX*KZ`S;p0U^fZVo9m4^!M~^8ur}@O^>wzCSvO>ZMZc86|$Xw}1;39vs&#hfwj{>0~ zlo=lF?}viLxeYP0$g8Y{FQCXkKl6ic1`Q*TefxoKL2NP{=ikiD^U#J0DJT`wPi49P zV0d;9iChZgrgI=$k^zPg@MmjKnAL{keFMl=?HpMfzIY6Z7@5EP1PG!VZMc{8@9*+K z7;>!rRkY79!pH9#fBI4%$7pg&?f#`-i*{-_0H5vtP^5MrFIQ|1S2j5@nPYk4tzgZ= z35A!@jC{)V%eR?JC`DW~9SX7DTcTHNs&{(*m0)-$!ru@Qu{+1W`1c+X82`q%*KUbs zdhu;GuoDY-Ogb-bC?uuCFfgQ)!20fz&F0r$yOi>kJxRPw2A^IgaK8E!k-*@PqJMH2 z->1_NKK%MqaCU@v{4e6rBQ*4>&)ezAfnzpyt<@(4+d(=x-m(uV?Oe<|;2{C$1oD#TCe-=}!q|Y-z{wn6 z+CZ`kvU~*DY62kx7?FiG7??qtAfSY@2)Hy*NE^6ieNN>IFE$)O0>LGCFWg*j_^Z*r zFh29xHy`~$gyDgpa=bdHXyUJZ=AF>C|A&eu5>LPWFDgw8h90fGPZPfc>whPj_{R|9 z_wFr5>uVUK{+81GOS3c2%WXQUgsxa1BDSCGV~QU(5E(oq3$RAh9Xzz=>NcU?kjjvMX(bm zG~u}9^c<>%BatS^&kFiWP-y`9tx?S-l$lY{XLeB5`{l%~&(2#mU};k@Mec^`AJTKTxUZtdBXuxh+szgDWsKz zW?ei?H=+NcQf15xyrW$?Q3<;$5LPTc*&UBn^`o{v`U4;%Idwo^2h$ddV zswIr5erZIH-;6m2&0&Rt6-o6Y97-22Iu>?9R`JSq(Ka>yD>UhMjsGIra5vRSfA8Ol zcG*1%=Qm#);e;^EvpwBJbr`Q9o1f|!qSf$;=C@!e?nE^IUbKdHAe!Itx1;6!yZe4# z7tti1foJflKY=&#s;6-Oj~WEP`80`U3|anhgerqSCTa`j{52FD&{Zh62U$fU zBmSHJa70Lq&p+|W`a9^KTOIiL(*F_BfqRnmzx3M3`q6@baRG9lf&l#yEXRi^w&dfp zJ$^Q0FX6g2bU*0~!<5ZJ;=R!aPZntPc@_vQUDSz_#_UHN+tnts1484I0Waw)SxCrlJ`5u8IA(*p#Er{ zPtqLYPXXt7o?-X1YM$Ux6Q~72L#9780}4;#FoXRpNdPfP`G7?gDRA0< zlL2tUSkMS5_&hAI4{lgtoT1QKQbdyFSU&h9-~jv~(XbhwWoQUKf}bPQ@uV8K|@UmQR=03pc+pCkaiP=Z5f11kzc$I;XXz)c3H(YSE{3M@DR z6?}%E0fg`~i6_Yp@efOb0-M z0?Z0<6JoK@%A-$mBw#y5pay957$^qcU zo#CM`*d@U%*(Aa%e;r^AXuwV%C>#l>!lO?D)&tV;=#x-r1VY7J;w4Rsx;#7K#qkq1G)fh$RtBTHc5ZnJjEt?AE`K0 zodggI8o(FOo)6X%@Sle5(-fs*1eS;h>@7s8kSf0k zQL6iOFnS26^l*?1B@mXN&ggi8e6%17`N89G>I`x=q9`=vkwwAmwg!=CAWT8k8gxz$ zWH5+si+cz{cLuO<-^dIFO2Cp26`@=ORgij$d?7ZJFGa`e!2h7z>IUi>gKQS1iS*CU zp{$ibyxFt!21KhbU_3C`qr~#-2}&iu4{JX=ijH?bft2jAXnF6_^$w!vU(WZgDe5Hf zM5~KSqd;MHVSs<^5Z&m_z%XW>ep6orOF}aeH}+w8C$fc{XoE)h1{VAhOy0y7Zw-@B zOgoPGdltvcS7T~u3cBrU$3mxZX)%up(~+@CPKoP7L~ha4ky@EzWKUmqrPr`J!+{?&eq7?2No+T@*!GoXKy=YCD+SSiGd zKRAn{&qob?JXbz}i2aWS+E_QgIS8ocC+8p8=*jbMy6z|EKY6(Fj;?ml?jsPFuKO?S z$UExDJL<^0O-H+_35-+;hnW|IrB$ z#f~6kLC9-@lofdU!|VV-j0Q+J5TOBQ2i46CEM!3fftc761zv0*g*ZBO0Ot_irm0Cg;<7fxCxbWf^6gZ8mx2xf9MoMwxOS7FOb|&FMvFdy#N*9*%!NecJ3$X z3o6s0Z|xH$MB@bsGZ>y9T(SitojQFvx@4_Pp&26eVYxeyC=Z_5N5uYw1Pn7c0RfrC z(ZNx4;_~n)pq}SP!er-@M=yRE`wRdR{tJ_SfBy-<#zc~nk6|DD*O&G|EKC69jD-aV zcU5>9VU3?c)^G5x@!wzS=2~v<2MAcd8R1Ge2A`EPcfx_#Oz+si30NEn3K!qKN3#$f z(RBn6e;fVwZ6FcKHCF-S3vrZ=&GGpDD~JA_uo=Q(`@fHe{Ldb8$jLiyy4dXQ+{Fn% zoH~aY;#=&j?$^-nxxtQ)jq=8?oEp_3`}aGr%K{P-u2Aewv`!bXo6z6)y$2ir#>RIP zk8>49NDTR}TB=xOL}~aO{_FA5Wn4U=w(?ViLpQgG6{N zf&I>rGSWGvhd?PN&_AG3p8q$AbP-f5&%XlEj70j#zHHvLx`^lZdGEb_)tfLyHHfe1P>Y`_Iue!7(Dx7Wt6vrpNu3b0jDFj`ofGm$$JKNf z?2RaN`yfpFgE)(1TwkwynbCJ4`Iu(zi$n(&)3cYR4-I`wuKz+% z5l1(b+4vc}j3ybepl3eKLDAGtZVdPsslLTaKnxd`DAy%c6jJP{(BYK_K;26nT`T*B ztmv1%^7+vF84ko3f4}rH_QYm*>gkE8oDzD%uY^L+@pQjGjvRXg4H>}6XUDto&p#vx ze)y`55d*~4ZBBvXAZ?+jX28$Kk(>T-Bbw30;U0AME4P`Nui0F+w2-#F6VEq)ax=%~ zc(eAW=6k{}epvp{1`J5@lQ*Lru`%8BjT56fMz|xX!aXf)!0l*Je4+2ZyvFS@vv|~Ub=r+m6smD z7yh3ITbi>N4?TP{e)q(LUM+FQNZCO|y32|A?!OGe)K|_6bBJ8?lN=)nX#wZTmS8(m z#+rS0BBbjqIOBkaMQQdyS3wcL$0$?j2WEIo;jlTLoO$4VpfTz_7lXSudC7xB<2!iC zeeXT$rtj?FolcOLC{W)YKX#vB|3=x___K43{ei*x6nYc0&(+|RncRoG)E5%3z{L;Rk z2M>U|_y_O%kwO#%ctOAu1oc3y6UUuV&mhVQLTez5CdpC|^~9c`_$-Jrf~YYFdm>5p z?86Ts?1h6i5Oc$vqYOW_$M}&$M~@)0HjRRgAV%q&d6pz0;t4*B zqJ3g-#UG)JT}cS>p&={@!eu@rm}e02lSCn2D82^gAd(9LozBlG6bMC85b^eDCP6qX z+S6#!jwSwU4|Rkq{rZAx?n0g~{@|-%I$F*a>zO_Pr53rQ&Q6i!J(WH*ezx?`YbXqQ zIa>!+Q*BjWdSOv98v>VY?M(Cbr5)hMV>+aVaPj0UXz_f$M)e_ri)X5BJr9gokNq|Q zB|Nlz67mEXDcn`H$L|yPQ~wVhFaRK^;QayZ^$f=K42==!v-fnaSX0gV;A`&Jf(Q0w z;T*JqKkQ&QCX21RlEwOc$l|Z9BC;q)JIwgXR}OwwpxF?xVLXm*85sP8+P6j`xb@f| z;K#*Va2j1x<}VCte+8-0*DnkPU9C`tIv_8>pCA|!oqq7|>@VTBf9G)N-vJ_ZevT^O z;ZNRy40h-ORpyE=NT|^jUHBQ){ATeS1(yc;1qeG^&J^$YP+2Q4F5aHQDR}>9*zV$; zkq-6t6mVg__~gaKJNg`Cro_@5zqt6B1RROC?8JOxzx} z>)}F8lp#|)gwR!>=$V~(KG^89VZzufA)mbUTyr&bv(^%hrcEjOd6IrkJVT$b?L%^f zi_aa6>+!|e8q|OU3|=>1T>QkJ%<5z49$Y{>4eDIrE(A{pac#u8;A-|aj>g{ozyF)h zd`54l0yck9_szTO;=eSAn}y?!lK z>2>SQAjji_+)OXz`!m(7EC(dLD;uqi)Nsa1zK*MTi7k8lI^MQC=e@p{==DY=9`&8> zrV5>|1i$zR{8jDMa*HZET*8yqqBYCZ+I8BLNnYztTFj72BzE*}-soEW3OCMqIlHE| z7MZlW5$X-9tS^-g-!L1w&APr4Fss{SHeD}K9(K%1UGJ_;qDIcG?0R7+18=GnR$^f> zO$oKWl1ip)6KrP{EqO$_jG64G2T93Z_p7UfHD7FtyP{dFs@@Qzk|UGp8d{&u*Xly1 zR!?WfYaydK=(m53O~y5O|GPAH9_Dw!YxV?ZoGUNY^03<`^nxd)q@nB7>HW7-##kOQ z!+e`4G+RWrZdE+7r?hJ;bz?2qWvMeub?sg%lc}XLS!rBcDY>aPl+A8NA!Q)jIUrzi{B)NdeY;nDZNskdV|WC z+-|Go(jvDZ2g6NH?pEhcPH=h>tI9(i8x8DHeTpqirPQ#!@upHsFx`A^nQOIXeMjO7 zi^4daE7@4KJLaoRoURnhWv){$Cuw@Qv8|H6bZ{J}dg{DUY>cfECv8XDOgWqCb$V;~ zrdA+Q4y`&_kf(a>_KFxxuxj4ul-sLpTiJ9AGjZZ1LDG_HrD?~aN1~cZm&{G2Au+fm z*H)WJbKA`HGUHwble5giO>jATyPCAD_FT;BleV$lvh6m(q%?EER_2tF64JeD*CxdY zhZ&VbBC*xz#Aq-OYUu$yPuufu+0Ny~)L0ZIV|-xeN89dvr>uq9cAFr{-HsC%ZMu}E z3W6YEOqLm?#}l_smE2KfTH#Dx;CgDImy!F@w5Rs^rkroJ-PL4W&Gr~7-=mxCLNtxM z*K7}}RbgDK^tPtQ5Try=c|okq6Pj4w4#g!QPaG*P&c)%z?1gQZtEMnz1q?$ zIW(7eBHwI`VMph8i8Q@4MX%}(XKsyN3_9kx;N-WmnM*Ee(^V_ew-YNz5XxlUF7MA_m69M(_dv29o1F#w!K-P1D}$4-hSZRVz{F-0YEGiOv8Stild zYx!xpJ6jW}fzs_N>FmT6Ct96NP*u7OA}E{Mk(AcvvfV9a3x-pj>sTg>X@iZGH=D9N zb{(v6e*8Lzc{<*OP3iM^=p5xR`Eb)o#Tr5RQkl=0ulLlQI0yRmoJk zRomfX7h4bP3f0}&g*h)}g}gDvHnRn56?N8C389c`@@=BWZGe`So6f4(nbfQKvc@Hi zJU(2xeV*uy8L~)Y_&h7N*40j$+2t3NUV~)^EruJW2X#Jcs7`rd?yB8g#t@2FZ@!Td z3?^kLdr_bp+v%iKAGSv(jWwpr(l(V$vL(5yO09MqPn&QUg5GK6tnNr;b6V~AKpSyQijpShYtndG$c$vFClSCD1fki*2XOQU$~`2ov?#Kr)oN^6 z1e?gO+Oo-2Ez2USJI011KjnPh%|YD$s9qmcoAo+n%;%kbo#Nww@|rB;GU^k#{zJ?hd{ zwXCwT!zViBl!eciV}~AyB)3fN9JZIsaG)oo4GhCF%0S#IgG4UbpAYm!WymQMy%OY< z)2?r{gq+Is1Z5>ucdLc5dP}UsxdJ{;6^u=l$I^9Y?QuD#U+ydwEs;;Vm^Q{qt1w+v(&_Qq>*Zyd zlgk?wD;n6aqSw2bom8!6N99a8kyv@7WvV=bonTI8O`}b85^}{cMq^c_T2)i5$XOb* zF)`JuWH(N~%e8xU8Q1c)<+8_CCo?X)v%1}`Zs*dqM0TbVgOov#yJojb%SL_S$ym|_ zUQ@^(UuGwO>zSog$|uR4$JJNsYNNku z>-fAVx+9w{_cUt6jJJu6i*s$-?HIc%jZcUiW+~WCFpIUNJWHrbqrYi$`qcJ>e3kPI z1!k4%Bvri5IEH|!l8|pw@*vH0RHGpC)mpL1$SbWX7T9SoUnmqRn+=(t%3G>8E!3Qe zoM5;?eWZ$#r&hZ_yCsS%Nt<#S&#ftXvTf4NaNS3x)Bt;;x-zglgNNRy@~U4l7=FqV&gIDJSINP@3!5^GUBPRE-Q(aI_U& zX}I_>)tb#(sZ>_dr`rmd&!)XfYG9K!rDnKl({@-Xk!mH>a-P`RIIsUoR--djv6VN7Q`Hz%o94ls$1RpV6>B~DQ~IR-M$Xz0=fP=Z&!yV z!Dr{oW=Yes*aG;@M3N?rYRVlX=5{K(b;;~*nauYMiYW}8)?l8>G%{sznxhrXS_z(* zp(`%Y?_$GZs$Y`}E8>m1I;|9NVN;rx+G5%x`4NU$nTlyTm1Ykwx?6y0U1PF6 z*0a0Kr8vwC-G$W7l}Edwsg;w-(X6=;xIS#|)uybiGBUZ*sp@W9tht-Uw4egMQSW)i}Z>ptz9&7d& za!wp=`h9o9s9c3zC-Mp=ZU&h_8RHm{t;h|gr>2@YC&|~7{W=iOYL2d?aV5_%wo_oY zEF=w;G>z%h$<07&TP9OCcx;wSRFiUp>ze6K*BgM!od$t7k)9fr#Z=Sfk&4f!L`Uo; z(<6>L-<6N)DG$ zw*=Sid@N z*OtIIcGGbur6!S(ECvr)!{_NeL%ht(|RWtvKH!PwQ*$Z2?!6-9_kOc*Q~ znO3(cDUnEK`HAFIsJ^(uvNdhBbX(K>tTQI_+oD+IL-)m zZlaXSz2T})v`W3XwX{{knWWP?Lv$pqqjH6wEf*RqY{X%5+SGy8Vt~@3?*`Rm4(l3!A{9J`j}JyXiJhIms@~=4zs8O^BT}QL}Ud=lagh z*0LN+3DgM!n=>wM4GMjxr8wG!`soh~B`G@is4b1q?y-F5F^+Y>8o~pHF$03Vzdm{>hlkN7oj=ZUfwqm7{ zrQ}+gs>Nnukl^`-wX0frwpna9wr;kLSn zWfgl~xxrCJ*#wz5V~jH;E5mAKSIK1LoV9_miCj_~3x(uxmP>Cz_SJ>9U7hTt)`kFP zG!lziZ35O5#&XHkXa=%2)50CBsjao@R4Y(kDYd|NTD4(iC~{1gS)-*ki;GI#)v)SV zT~2#tQ*#nk5Rq|3BBc>g69snCSohX*Ixo%?g5Ith(p(gV%&LhKtlUdV)0I5+s@sCo z8)cJqeNC7-GrJvadeb_-E(n9Z%ar;#x>X+*6Qv#P46;nd#gmgp+bnb|bgz`l>{b$y z?Dr+HHE&iu9_G@fj2(s7XR0D|S)tglH?EaPa$B~cV!+3%+p4E3i%Mm+Drx-`*-RDM z&2k?bDq!5|h zutRs#t^y(NyLD_NIKs4^TGg`k#73fqbgz|>M`m$Y8H?j?PNMX_otg@rd@D_k%np|m z(mUDMj+U6Kw9QOAZ@DRu+*4edTG`^FHkgPUAup)%jOv#An^an3xrHn`J0`7wGV5?g zA1s6m1NhN!-4QH-)k^iMheHp`0T5-YZF{vGzTK_{d57AKB*$JScZ;e?w>yQY4xD+F zZD?~so@G0fP#0H?(YRst2I6c>w539K1u`njRi;d>Wz?1S4$F(qwo@g%B{Qxm!zRO* z@p76fJ7X3^x|-3c*8yEXPq!y#ZId400zFAv z%)%TR^x%} z$djx(5a_Nks}(2gJd0yR4qG#JcVJqLRc_SjPqapolB(%+s>Y5RIG|v4NN?-I7T7L| z4u-Xxf{+?)+3`eGq?%}r*K&1=QLAy!oNSsFQLgEl=n=h^%vWYZU7)%JPj#L7Xv{^^1Jr1%G+&0bVw{AWw3;`&exRL&L*sS zA8?h=dIgcr5s6e~<<62;uhJqbX+^0tgo3)7l)H1!c6z4E&a0*MB1g3qCPTYI4TM6x zULR5!Dx0fIs{%vn4sNo>LcqkNKo%=qo>9s?M#pW5MmCb(mK?ZR za^(&fnHu#tqch4)u*zaspV9ewd6=eCGEJ34hMKBn8QadWx>c1mtq%0h-qshR+;UT& z%_=g7w{xQkF=}Y6Rbw)_spOP)vlX8jN@IK5O!cxmW?txSww7JB6W-LTOr1uyHEs_? zz20l>5(_FZHHN(5RdzU5PU(U@6fsVltcLEqT^cHOc36ib7`Zt!&ezxGCZSo7R zO@GI3GZnX)-W82Ro7M_OzND9%y;fD=)RG65IxNxf*jlPLtfgvFdDWTHw6s-*mIvkv})P2hJxD1O1kY#MVA^*xMI)A zd2}kh8jo9T&Jpp>ln~14Rd?BH8w5Mc0Za27zPiei$(k@0S8~E~cAiX(Ca`*j0JNr> zRtM}dMRwbRhTb!^2IK+h(&7}eC}v*nYQ|#Ss7Myxh6=fLfwY=aX*iq@n;W$}OA@d- zi{q)h(81)BbA-YYG*H~_)U48=OqE7%Z`a0G%fbNR8&<+BX`MBOwW3gyYL&6mso?XO zS~Vw1qQ){-mtk_vEUl>tW?Zf>=l!Xw*Ct}RyjqJ(cga@ER+1@*eJL|(OY*$i0oi?k zSAi1m@J5&3(evCmS0lRp;#{!RR7D|qMP|u@kxbKNu*QsZ&EvfkJr&1Ro)*;nEGJeO zcbV$vSNuZcwwr>dRhE3G($)!_%O_JMX11!(sd9p-cv69Hujj?;isGF0swH*unL;~T z)D*ij@AgV~k?2g_awD5grp#F~vA|oc9>7=5fMU=a4z{DwI5Ej57ngf9SQqq8m9qpe%?jGcoXwbAdsgchcu(iH zqRU}WPso8JD2W6`jWL<74KzkgYzBCa>VqA+pd>678<$t ztF*0JK54eg*`!l8&F)BZhATozB$x+%7ye8WWlwV zZIF5+>A7|yQR@=!s=mw*iVD|Bq#4IFw;Ne;H6V3hwiATDjcG=U98~nox?Cd?30WFV zk~^(07AA6Y*59UZY?PX;2SDR3uy>W&ga!M2x6x@v9-5;(-Koh9b(5}=raD#&W=0e$ zE6b>8gfktiQ(0pI+D1Q7aMqoPgX}_TztPVK!hA#s32WHSb*JRARsnHmKFX)Tfsh(| zHBhEn81iW}l3Tfu=2ekRjdJ70WL?N~x!k;N+x!e{=Hq&`3Jw(lb`gPL`RbNRy5LGy}j$u_YYCFm++W1}ioPj*T~2mQ}!Be3T& z?r2g>gB?t*W>_{yI`Uc;GUMf}IIS6NWtg1}!JX9F4Lz!8&kAj-gM=}&F7>GP#$aowMNyZc`~Y>1gJzV`a^rJ#Uq( zTX4=McRatgo!o+?!QNr@#9r0v3)Eok6sFvCO0CNcTpwVI+N9rFC)Kgb4l=khOs%Iy zx|UBQKsPrUy6xGMO0m9{!Mh;Wl3xARrSG+ZFaPHx3pNvGw=MOx3K z+Opo8x%QNGXi&o)%jvT@ng+*8&hi+1Rb6JvV-2$=qj@vaS34CD^)0?ToaVS4T~dkW zWCT|H7NKnkPuiAi_^hZ8M@F&H0@?*m|DL`DM4-wXJCx}W_!qS*X0B<7PjGvjR+}yh z-fEO-c$=`%BZoye8Q0penXY^{rNHNc57N_p_^vRVuZy+C1g$3LS7`Zu?~y{8X0R zvD&4X>$>%BVaI_PhsF5G##$-Hj;0HekgAUA$llHswM0VdTEo6g4Y)yiQrpz$jGbgf za-?lQ!N{)gv^ZsC&9IZ1Iv8Mca=y9A%?O6|hTLev2$R|(?W(0kslE_gIu9ElkxSRL zS*_k!uREzyZlh>k#sfpsTFnjWUNyUEjQMQcs8u&+rY`5LX3I|KO3fCkO>ExU;g$N# zZFAMOp5e>Woy34KAnOTRu(WJc88U%+EWR!`$E>1_&_1qqi#s)$HC%^wOY92UnoJFB zAhI!LY%~`RbEs9%mnAD6(0^U` z$NDR#(gV#_V$-GocE7FIDJYpV&zfv`yPJUOdxtC9{}y_zY58Fti4riENil(Fef&F(Oo6TDFsCub@>Zi1XC z6pD#Lw`#C?WG5hXm!$GZCY$YOuG~&r99e`Ud}*ETk9VnURW2v{gYmYZbcfP3Wtr(j zM=wlj)lx#}maAH;NUGbk-lgPPgS6a?J6evjMR#nfsyLwf=@IhH&WkNII~3EZ+N9+Q zRZG{^s+&=J(-a5`lEq4lq&wZL&;wjA52+5`#>O>dUny8-iJlfr8St8?t&F!)6)=h9 zb*ixF5v7IIBuH?Zo1+$Nrle$U*jz`W#|$r3y0sk0T_s7}<+1v>INVB7rJ(8s@Fc^$ zY$7uey4hAHlV-qxgU`s!m`|7MQckAHZHHNj37+VV6&gzs1F1pxI}VdC*Lo!f%u_~r z?9FjGJ;lrw@UY$-yeQk|PFsN1F-{SBNfX84e3_IQ3nty~+1VwwNUFmwj^%`E5lj2} zMWxbNXU%#)KTPKpsjk)RoM5WGQMb7eC7v?mj0`X|)ASrm1N7)pR#!F_*fky2EOEL~ z==DXbXiu}fT(huph$Uu$Y%SGvh^6VawYi;6n^g`>{~)wkjsBuOYH=eTlbFm{wdR({ z_IjI9s)Fa1s$Rra=|Q5q>*v7lB$Wx6P0uV%Y*x+Vbs3m5GD9uZ9-i9R8?boV8PRAZ zx~pWFFMty}tv1tClgcegWg~8~jS_1UrKB|)XDT2m3Y|`^(Ww}%l|eVb_%j4~MwvCK zwIi-LIjh3wz_DDc*gJ2?vThE7GCJF20p_w0ZA{nZQ+H8IXI55yn+H>Ry@L0fE;3K! z9OwYUQCDeI5~`5NI+fOZJ_GwBlb(0m$)+*H3$sBP$15y^MmYnpQYR}-5E$!?T-RGQ za}!-mXBL(1s+I@8pKFwt^@0E0sa;Cv!7x5w$O_1BP0(Ch3$K|GYl2(A`f1I>Sm4DF zn}esc?3SMwT2r^v%uo)~ZgAyQ%1Q59__)__r%M7Dzn&S0Vr^M#yI_BWI2?7WrH9oV zHWRJY+y+NPe$e%DDMIZTV`ov~P0wa@W<;5SBuwNLKAo>-PQy(w;OZ%Yfutj4>0(Z? zg;?h1>0{Rb$b`T1=PCPROM*ZOW!msR)HeFTI)u_8Y6h&H#x0T+=^ zek{%nP<0#j2!zW`YwHg238wEVb5=B1P{dhmBGy@MSg0@LJVgx)sX0~Lh;@6K%a-*8 z3R7yOxJ_qnjKIOcZHBp2uC=OzLYb*0Ds{WPl=EfLX!O?d#wLZfUUxQKvs5W})LES^ zVG9Q{bJo(S!x-&?W&jWC&hm1lw#AmLDDdDXtTt53bJ<-syW6Z4nRE-yzM98sbudsN zUAQ!MyKRZC74V&HHni~=lzv?^B)%+;xqM>hW(r`$w(4cODlTigdT&G2S*ux}iyeL3 z5Nh4^s?Vueci{}n>Y(3gZ3HkiW~2i6M%8kC-Kxxa@N=b0Js!m`pjeBkv%&ecTxYU) zZBbSiB`rC&1>BX4MWyYqvQ8|L*|OQAoTbp%VJ2lsdY+dJdm_v?-JLFWi$-SU+hi*A z!cv>+eC&{eBt0Luq#C!ZD&VcB-N&}>d zhMicXxZH|u^m_@Vol=#tGRRNa?z&j$=c^DnE zJy;v2v>S{K$pA7r$ZEn+$vBxjlQi89gboO1HMi@SiPoaXtOpWENurRenUg6PsO?-` zpKY4UL2+AV*V#>$hOkOG%Wc4EHfqke-DtYd3oXzw%VS5DmNms{=GL2RhG`d&%gJdp zu>}o*a=MzXWF4=dL)?{47=a_Vy@g`6GTKgGO}O?N zB;0(>8>Vb^20;?u%mX^37;|vAx95CcUdfExpUI0PHckl@t&l4evSO(y&cTMMlv{;d zwzuw1(*;EYdmb)LD%q~vql*1jy_p}Z$`I!ScC48=gg}>hbvM;lU8bk>YQ?-aUsrmq zoLnjhOw(9&iY6}#i(xIF^MJ*IB^>mHENtYQXVp3al^PMnOtmm8k4hC~O#pw&I$W#j zx$BZOUyaGG&W)Jj#>@|znQn0`I@L5pex`5;kP>RTE!p#hH_K58krME(Otwyq4c(ttl1Lvm{A(0V^#k7w>daAs>KeWG_c-a^N#9w7>(=VQGw_n;()c1_~% z^>Y=Otgd0w`^*#DV)P4N@t~%RJ6#g%NBp=r)21ySHFETv)^*(C4;I&e z*d~VPPI+DU;~HfzhiNQP)1ynbNbvy0d_7l+q_|}mwgQv?6yq+!H`Cy`i{H1yn%UL) zZeNZy-U&PBnGYRL*Ye-r^*j@D``mY!Z)p$?qa*#q|6geaC?i3~aK~HyARxkAQL*vR z(l6!qH3y$o>H6~30!gZuLAF~iIA-;3HW-=hBpP1M$8uO~otO>iRT2H{*uZ%TO3Nl{ z`J>QZ+PZb2le)FC-z2on3`fqu-LLdtFVP>=oAM|IG``(keVf=Gj>)vW5~l>ZYU0oW)HZylnVhe_N~hy88^JeR&*j zluJfm{eTvq3WEAX42gfZNlE*tAyBxwe#9f#_?XSOndCsbuktnQN)7`ez1hoO!{3Xu zM=_B;2+f_8`M>k_ttjT9@hs_B9|Jq2zZCw1jz<~q+{H`voS|Xg`%LajOM8?^|9O9; zqb$EMYOhL+iv?Gqsq=a+8D*LulyJX3aKl@C^`?)}w;dLjwGNwnd&gX36k!^0Z&yHn z8hy3?{P_O8H=-ejDO<#;58-U1aIOeZ(YEPwr+^qt3VwLdPJmZD~B zN%nc_|NOk}4BY}EEk=BQUgFrEspIRoQ?GW(MoetVhsKbGqj@({$GDMdH4M{4=@kVWlk8&zsxbd-%(ojm3OUcow;?@rLg2#(jXIXXdA8 zmYt3Q)rUQojxVA(!|Min*vC1C-5Tfc|FDJf%67?B&HUS*o(|vhZx`$+fnTy^514kG zLCf%q9U&Fy%sb`njW>EkyGF zWkwc;JS>dHG8p+drT*)B&Xtx~21`~13HFYnKAU|_5!(LweC}wbSJ;!9gBlLAe{5PG z|5Yu^LaghL*vT$>{^Cbv@#kE|rBxNjZTizPKiZ#091kkQNL7-G{18uvbpG==7tqXK zfZk{u~W^)M{Au!_Ov?or3qTYb~0~W&ly|B zo;09)OOwFTR)_IF;!)T4`mGsoxJe_Yn{|icee0H+cJJ+CJInBHGbkQk)yAarN*qr+ zGUJY~A2NEXS2q_CRo!st4+(p$C3wHD&w)#lz9=aBNk-kFIa)t*nm+~sNskG*=Ot9& zdV{35e8XEbC(lptP8wAU?IuCzb={1c5<;ONhla}j&-c7s0 zdbc|RRifon#gsIO&jzj+w8o8wX5>W>8wCup{tU3UY8Dld7bJA7*60dQ#UimV0 zo(gOL)SREMc#+eue)@`MnO3DW<7&5PocrCw75!6Sxx+xju#fp~d)TYtV6cKa?l?RJ z<~0gcb*_JR%D}t#Eq}z9**(g(OSSn$oQ^jQu}aruf6M>)qi?itMH(w#yhnAZdR8`> zbRjnQZ+1KXAjkG2eK<~@;Am>e5pR-w3b^1*{A~C}Y93*?iSBM0LxrE6Gi$_r9lNSD zz2(!MSpHFF#bQ4U>7S=g(cFV>-!CDauhw{Rmq9uI=$-cN#;g_IZKAH}omcLdoEc#w z567C2H|SZncY37i1($lLJi|S=Wp~A1ZmEu$I_?`e?QaC$-y(q_B|&{P#*%hxVC7tCdriJx{~nZ)?x>K59)Y1iIc(_g`8uWc^<>>StoZX{H0e<|B6Ff9NvD~P%1f{U+V@J zlF(S!%Icjjne5#VcPUJ8q)vwC6=|`8&0?}}+SSDGW)YIDA0rHLm-Y}FRYohnzw^04 z;6?~Ig7EzgO+Pfy$Pxsmc5z!Zc6V7x-V0w_qPqJe!wf?r8S&2?SV}X8(SB z6fL{nZrY#|fBpO4hta$$c=h?*JIGRvYvtkq^)K+x=TsK@ok)gHWb8SP*qtCJ8JLMg zzu+mU@5jeGzCGG~_|Z$-M53)qTMn3uv02mBT7_?MNT!NfVCim_KPsB4RjF&*&tAuRt$Q(w-~>N#}|SN#sc#=ghY2 zD=PDR_$WC>h!2|GXP>WL!GRH2C8a?v_fJkcNt^XrF|Kjo+pUHAR`l1~bWnGlJGRHl z;0pVcXMwoU)9v`$9q!D>s)zg~Gs@?X@bSZ%TW3T>1Ad?1u9!lKesk)06GgXzzLIb( zH?sUy7~gDATALzB4-zd=8zX!n|>6`M{6xP)KPTaQDfN zMYhb-%l_>$hvWOh+lMKY-tWh97HxMo7PU#mRk4uo`ClI-yq=?bI4PEAzhZipn*dk) zRUPh{Ry@c0powmwK~Cy##|<;jfw+(-^V7id{)(o=dS6CZJReA#ls~SWv9%X$x}b*6m5k5Q z5RN-{sUywQvRV*@zkdI#5}gw=s=-U|$(?HcJp8T^{EZiu0}=hV&zZ<*XTi%USm>pb4w(-sG@Mv8jAHfAJqJCIygxZNbH$Acwgt> z@p3e>kvu*hIAQbmAt&5aZ+vs%&zixNQ}_Ps`@;J~#h8aH#S&d%ntH@q=#Z$HVVHl4 z4E#9>;y-)m-f}s{Jl6WjX+9rEvwPC&CFL(=*$di}y`P@p!r_kOi@|_NixbUpxkOSh zZ~0s3o>0a&*pgW@HOR;)*s^d`Cqz`xLS_Cpcl%h|sOf=XCDjQz20lMu&^cac&bx*` zU9Vzxc#$eme0HYw40oyKy^IjED_z$r_jqZp-2wggv^Ed(cD)h?mBB1*v1;?)e+pre zpCT2V>Uk2zU8N`*f`!|!hVaZwUq1dywXJfTaf6kQ90dO2#m*;8epFhc=|Tpr$A+36 zM4mpp%TQd4xh64QWMDap-2VMbdlUT;g}Xc98}Cys8#CrRhbLS8p-TG~58bcj-SggG z`p>VeqbR9I6-iH;SR_3OC+EWx{4@u;gWldG3QU^jJotHG5c|8j76CV{Li6- zc;=7hy|4}>^#XrtJYJ`N9?<{Jc0%PtNes_^Z+mJqs$RN&@A+ggr=Qqo%ndt?losyl z!O~d%?*E-f1KXdBBct|<9U0D-`(*)hOotzz@s?C>*&#Pb2bI$ATRol%Dws;hg(bRm zq&|}765sF1POl_AUoO`xwTKJ-%SbA-Zx|{qj{|uX*M(bx{b^eGGFVN%b+Y_#LIS*10<-s1Mmdn-s-b;EnS) z9kX$KQ|Tabf!kS_mAvSKv1+wKa1&YnU;;6Zd|CT1)7z9oPVZVpvE_wR*6l0b?ob5V zP2kEEGh_WB{_f*N@REP`w9Y}ZuXtmK2Q3Nt9rHP(?CwYZlkc@lqHabOtCXo{aN!T@ zBzkcqinYn=`Qtji`@BIgK4rYe*5y|mAJ?T(>{*7&|2o_BGRQz{yhWN$L2n6 zEh4J!V;9Z){K=~E?>7E@Ry=`5|LpkQBJ*9QsD+x-@WfB1=2I~A-%c$J6aTyf>_Gp>SfRSt40<$$DRm zdFkOEKJGp*lj{*?@;kobpRexx2}N3zJ}?YgzozF*PU*6wd_DE+gfAOQJ1 z-sc4a*W2gjJ3lX$3+lqYV*?}Nfl<5oeEF%eC1lA+_>T@z;gY1cY78N?V02{ewHwL5 zt*qmb+l1nRIPz&)JJ)E6G)CyrR*viLOgzrEP zL;NG|%v&v}`H`kXwdK8ZKaKPJcVd@6i_)M!4`va%c@Dp$s>3OyQI5MakJkyI*(tT( zcpEV>wja|6=Vh-nWhpp6nT8%3maLDdJ~W_fB~Xo?j!G?mkClh~9fo0k3j=5HD~J#? zk@;L(u6tUN?yzNw%S~g=xIz1(Dy`fI`y-Xb{k=y_agc?ob>0<5yh3j8-HUqIgtH05 zJfqAPLeBCyU5T5mF1M&%^7?l{d`fL1eydN1dwY4bhjPl2#lGIyUEE%Z8!An@@b+uF z1x~=f$2Z?@{fkR}NB7DWGeZev?ON`8iRZp|92hCRr_t_*6+(~Bn;M7-uYHxp-H(uA z2*wCO@-0$De78Fe9+v*?XrG=Q9Fw<75}AKfMBKGpq7b^+pH9jD`AKgfoZy!uF$o`;WuM_B0m?Ug<(5tu4@qM zzI(pUZvGxC(|Jt7Yp^THrC#!Gp>O?`uSfmkc{@%M5d`gD$IG7VaOmrN09pJewgK6w zmz&}UdU-`GgUaAm?}kX&J?m{LHZzHIs!-`>$3KPf`c59@9bQ z?jEjlQ$82r#H}^=W36sXxN20+=ThQx|FT2o7#9syJ$=EY*#0<@Z--y%%M>D=MXQ<+ zL{duxlE@6z_Z^XZE}nX8b^0V1-s`+5t|fAWZ5Z3NzvQFRxC-;sDv+a06JAFA>-Jm| zv)<04Q|lt8(J-a#{%LPJ=$RidJqEfeMv~plJ(54#$Cui;+LW}s+k=XFu!APkTvVQo zo9QCw>p>0jCyFLz{C#^&A729zKF*0uh!m}e-Oj?0@tO-M8MjtM8yuV~7tE%2`$In4 zwq-%2&Ptg4pkFrKHc0Xm8CUh9c%yR*v1iVXUN2ie|K(bXf;?uEq@xI;kBBg*)-;an zW$3TJe(EsLHKB1H=B$#F_Krz*;9~f6Ozu7n9XR%6K12k?{k_Yqa`<>* zJ6!(tE#?pRQlg-5+0OkpV@Rc@iwmA#Jg?XLW%bDYoS%g0$uP`OV#f%&D=z^f!2i=A zdGzT5V&Cvu4mTL2YVDSsI&yo?-ve=^#zRFAkR;$KJ^eHItm%vS<<^#d&#wZtGpk-z zlMs3p_RX~2`#&Xyulr##mPWW-_d?9X>0R`jjFN<4KD=AD7KB!`AI@EV!w~vbCZbXN zr?w~80CvEjxgHXu=d#y_qg0)ITColALij(v%43llUw5B*GPUm~Er6-g{J~aP39x13 zC>e$k&TI~BDrg=qKm+i0=cJFS|73pseE{3tT@{t%36d}_2E=tY)bf^@vWeJW-jkDo zp&t8bnLS)}*=nZqm;J_*Nb;LzKs3h;^<4~GorfoO<{uF=fgRDh%Gs;OR|RX^`2lmJ zYMMvFlI<5M*#2@tCVgq`>fZ0mb)XVm(GI@_$)nwlBI63ZXa-gKdY-iJ*GZs($Qzs> z2?cb6YZHNRt;vyEm51c-RIgQJqTI$^v)RU1-K~}aFW1R2mh3IJ%Dp+%w5=o_g(1Z= zKkPt|G0L%NFTb5_30{pc%?fvIF)(|AJbQnS#I-xc?BN4FDM@zr`OFaVKaGXM2;0_H zT=NnAf?WBdr}^c2hVt12STlDg4nDp3I@^*clGg4=$EJFcT$nV#Q))3%ehTLtbL5^a zruxSG^?6xwnUlmuGR9G!zK;Mgds1*VN(V|8>W@n{hmc#tdWWYf#&B+<#Quv$ZhTjn zMj@PAi^PZQr|brlNOSeI6@-8=M#J0tl$q1prTNv`>-lt^eaI3j?3=5EDtP+(^5Yjg zqz9InLj8jEP0bfsbqMK1o}u{p#pB@7P$_l4;rj9MmL$(#m@O|(B_C78GNTUJ>jABW zTX71WVRbU4bJkx|NqgKmhY2tJHIO3ajqLj6%F>lBQR@uhd{r}+y>A|dD>{x7HDKhFtfBZQ(6$3@s6 zfAX*iJ9~Hn3*JTI|9Rc}{*O|mAHm5GiWwh=hJd-v9v*n^X`$|H!xYhO?GvB8VYf8G zlK+|u1kZy~o(o%1M_TVw3gVKOf_eG-VUNk2+1R)&3CtM5lGilh1Y-B14EvNfIahV9&e z==`o?jeJY@m0*uD9cNjSSFj#+;x`<*;zSRuu2DbXuO@>igpTMVO4`FK<^4 z)?GI-v%cU!YoNQ_;EYt4&P?r3dQqA;^Ob*oQKVaOP4>HA8)zW@=@g>%)A=!vf8H!!rF zyoG{I3X9U&Yw&8d_;AWYkV$xj-g^La`nNlnWSGA_Pu&LRtNMLDuR2oUhIsjFDEK_` zDRoEjR8C;rLziFGZ=j&ebo^~6kF18rj=qBR6v2kestbA=GVSjN^`P+NwB4OQ}psphiFE$-k@hg;iRc)7e@+siS&Z%vzFo-9_aDKI(nMs zc5pW&Mx=*{@fP{=#V1O>BFBPgi_WdL-0{=6OQnE6z&~s1aA5D)yA&f$;D?Ja=rTMH z&;n{vybr&JuM}NAygVH~h&YyZ(_gP2Lf4^EFOa7$mwwCe`mXj!oj2m>wW{E%+s^NtrQu?_9NDFBB{hVecaIQG6tnSfDCA zQs5C%X|&UK^g-nu>V0!x7iA3|@0FzFBusj>YEJ2sF*RKPiC~Ryz5GzU!zThyU|(#y z{)34+>h_I=yrt{ChvM(C!8`h-+wLwpB7ek;rR>sJhu2m-hU4r=!PC`kl1_7b&~Ym) zCQWMQ5xQjanS5sW(Xb;P+JxN;Q#li!eF!xf)+6{07`n`Ndb`%1TM%M@+?CS8g_6de z1+PVdC5fcBh6zOJb}2Pjg+Qv05e47%apjNbfvI@yKdhT=k8tAr_BlkZ>iwGAIc>z7D5L2vNl-GflS@#Fh$n~CP4U-p)SUPd%fpR~tl_iJc{{M7ma z-%{nZ2v~dz`ph-gn(9t+`_~Iy59c<_HlvKr%bGvH3ejPzu6h~3t4|gN zRiY$k{ihYD=CR8e<*mpd4_NN7Mw%eAj;A2s;`7d*k85$-&U)X{l75e&b{pq}XKu3m ztZr^GghP@si~3v4;#nR_o$$INPJBhnq(%{rxui&=nG`lPxZUwhSzi^))OY6cAhoPWJrT4=?pJ6P zyg+NC(3_7YFiDAsAKmXqE0nkedx;N!kM(i;%+vA5J=G4NP4VT=!3`hqR+zS&H_Pky zP4?B;*iZZBNV)d%zNbH*c^=V#V?3{<@%bVnygct@mn3t<;v=VBxC?PlUf?zIcpR|A z0IRn3WAvcnPqsNpZ~K|yp?~{%`xFeR5c2k8s=xk7s@XyK)+C9clKL1p7Kpur=K1Qm zFN4aye3ro7U6WJD_EDd{oxGn>**FnnNl$|Rt@mGKsH{Nr35G8@dOC(VOMjnN?Os1J zGZ61r;x;oNCDrNg0FrP7_|B$IZYUm9RyiT@N4)tQCKI%Rvt}lVQ($d`teT z;qQNJD)!VfxkP1vbMR;RO$*J@y;ngQZhwm(IpZT;Cj8ky^^aB2u+w*WYGw0OdC>kG zTsViL5+fGZ*+r6Cp(FFAu*xHQe~_3!)@xt0?&YhV@uL@XQh>v0?a&W6%^)k&c+ajU z6_&=&A&AdJi(&tw{$-l!F|@ZW@fe|(@hdV+t*6H^%{jo*#x z*{YcD``7nMGJb~uo|YU4r`Ba((BWLErTAxfn+pSSOrfGG`-Ji>NMGH181bH~ISxoD z33_xN1DirdO2JHRqoMtiQG0!=bOLBsuJC(!DUWqf;;NN_1d6>PcvXM1g_-B%%kidT zm^f76CuMtQe^NCT*cD2RGETok+6V10{}Wbyk^RC@doe7no~0#dz4@AM=D5epbKO6_ zX0P;bpQd4t6P~xvq+mREi8{tB{yxX1^`et~l*!*-xn-$xQ3W!)mLnVNxk6y@;O=Ry zuN?RNNb?8eqWnrArN=(rbjo9$6#c}_JPHc(8}mAVW7J>HGsEjWguc% z1>eikdfP>WCDP@7Q>TEO13-=bQ{=PeqcT!0wk2z z{C>Vpq{(-*z9Z!3uj__lgFJCV;d{{33=I;<51>BbKKn#zgrRX)XyZ+nliv5-} z`|B8XS5qR+ddpw;JAu4spa0wwGLSu0jCvE}hZlGXe|e1hOwoLROIe8iyxqaWQfudL zb{DBxqP$<~^-?`BE?@ce{Fhhy>&f%jU0=-Z-MBs1^*5V*i%xX7F7Q&@inY_wPu@#+ z$Q|DJGWYEDN2OC5;E@PSuV$%jnVkh_;B&qKQ^2HJb5iw3W_{>;c)JyE@|WPz{t4e- zvBVOeTtEI*qmJ2pI-#v~nqwU5)qGKl1OxfOhVKhpo(^*(LH$`D`Md|?qs^YFhR5_~g`qey9E14|2c_Jdk>*4UNwa1?k$2$?bU;+R&EVjFjoS{+* zkFM|NCpRPd#aFK^6VT5oA93jQjj;6FObSKpsbH+avG=uM*U_NIWEq;iPKoK4O*6Td{Z$&K^OszIg59fFMdT+CsR_V z1Yd`W9U(fRGF>S1`So^BlS?mH%pre4dgV^V3DU*WLZ0>~98VwUMZI(WW7(n_fd*66 zr^gQcX77P@dz&D<8FSA!a(`wfiigfhs0bOXf%7l z33LJ7;-7!WrnPHVt1bI65D!0!@ItF7vDa>-O7L*N8?-C>aEEl9R;iA2^X`MoCC$ga z1Dj|$gJs~kKt-XJ@OMvsuGA^5`Qlu-Qvz_`V+M5Mjd`h2i)Z+?c1J{^5sN-55FF^@ zcEIPYE5iPCf~S(ihn+7+Ejhx+cRevo)WM(qz3+j_d-IoskQRcmgZlSG5(PC z`)d5l7*iizYd@!3ofNtJO+AfiqZ7m3j#-yFKB+;p5pk!}RCy42zZu?-qw+o6cfynL zQl?({YJhNl^jBK~+zrFhzWD3Vs%Of9E(Uf4D+sd*W-`5YgiUGPbh&B8hwXLb-D>wi z2F@Ot0oi&Fj3w|Ax5KFCkMmO+GJ5-y{6Si(TxIIG&foJeU-nO7j>I)d5a_Lg_~P47 zxlcIrd!sg_6_*YmKs*OeGy_zvM(-g}RxpFq`VPpU2w+0p+OEjn_l5dmw|}OMNJs|EYWMJRNW-&q`^&r-zwe zpjNTa3HKAxT{Wl?)fM0uFS~PZpDFe^G@zWW&HZMJ;I>TNV(k<6eg$X|o*PN66mGOY zqR_r?=IQ0_<)yIm#j6EFD|~ej9|C+t1hqQ9L1AX?5=cws4EyVpWR0T z|2B-rs0sVND#ZyF-<<{P93y47V6T0Di{psmZdbw7Jk#5r*zvhQb5~h|)7CI3U~}-# z4^(UdyiDg;ywE#)98X+yAZ*uN2?gMgR~5mx>?AM3iVxfi`W0p0z(hr^eD|7vJIh}` zohBtpK)6!L9skwl|Cj;bCDPVKFoS7h(J>TV?@I9Ci|yhvN1zyQ)b48KApYt1KdmCn zxOO7q=B_D=@C?@>r^T4zN7O=ln;R^EfHmJ5HfsivU z^gO>Dh@5zu!5*wkjAQo{cb~O#%}O@&M&$9w7oFiutDl#kIusKT)@L>_b~G~^Falxp>2w? znSoC^z9b(y4|?ovvF(@RXy3kRJoQB#&+mhBu->TzC1uM z9C(1ZHHSUk=gK>xn=iHs-q2=({4?vwm!%Z)55S#IJJT6{gloP?IoYRWynI0-efAC+ zjoKgQ2Z66DuTbn+<}#u7@cqK6eg3KT-{`SjzuBE{4IE~F++r}Z%%2IiqykIS+#b`t zd|)lrWm%*0G5f4^CiQpruIbrI`v0_h1;3+mX4i;ahjK-`BnE>a4 z#Dymr-(Oar+#&lS8n9O@ND|Q(8I0AEL6(FbVdlTL?^B&Wo@%&6fLy~NfrMJmEk#r8 zs4~e=^kLSxnHR49*bo|MGmv#o<>-zhrFMx~J868-yJ(PIqbw@w#jeMaO zI;@M1V0FRW+2z#YGwAF6 zx+KY_j&&v&hw~{Kd#UaxKh*S!q_p#{z>fy9@Y#hsixg)#r}Q%Yd(W{EKpR?>R^iqS zi)ydjndukf%DSCpMCH-}gdbzWi0q0EU*& z{fG26!HuAs}y|*X|U50J(Z{VPBLJ6v?sz?#BF#&8v@WGbI`P4L_UXt{g-x zccQ5N>plqPKVDTHO?Q7`eO?UawPLkrVq zlr*v+6?E|H@?e{C#E*{{p-lb!jL$^4Im|1}{fD7}WyyBGXY|LY8NN5Cw9m(Fe|4!# zT?5iv@uGox=?DJ=|NYQ4h{InN zsd44u(pQsP`%W1VFo>ov#~`l%{4d%0<97SrKgTE5SJ|ILgr2` z_1E_)+2X^tQ9)6r)Jk(#p_XuIKezs}N>?eP0BTp73K4UR;^8y#f?LPR?q`i-vjjg@ z!T=D~jTv@m``x4FSb>LPa*F9hPu5I(dGUwiLVk1;udq+FN8?g3he|x$zX**QnSv?O zy1=pYmjkm9!_#%{V`=5(FZbHWZl?$Pb>2Ui*&M8XZZhm;_`%S*CHG5(| zM*yn1aXe%E<0l{6eHCt~4o!Mn%9S0!aQPT;%w6Y_Lj z>X`HS=fZB^_i?8HivIlY?(cEkp!L?YW~#^c5cX{S$DyJ}IGc5TncW2Ql{;M9aU5%> z*XUo`yvDaj8W80iAZfxg`huAah@{o`U2h<9cx=Nj#^lfYamKv2h*3;X#_C3NEhznK z-@iU^OS^}|!!+)Do+&F8In^i#-Ea`U7e~h2M)wf0y}UZ>?+zR^#JP(cB%D0N;6?Uc zs@t|OB;@;WP^M#_2=UhvX8M)+k2rl;el=KCW zLP_t_%gb%y6>hXWk3Xmiq|4KIGpq4|$5O7%DRb_e?@&nlWM=eP8jC+Rga7Y-Fiu}y zl%_%bM$JWuhDJu{ZQn;2#4PW2-Xt(dA%)gTcbhx{hInKE zPZ%@nmc7P&0$Jh_^`&#}>_F|y4J z7WfRKIP-3SihoEIS4BK=FWBcz<~RwMKLv9MXDGd+&dbCa;CFY(kZD-#Ce1g)WDizI zdB3Wwz~VdCZaWsrKOJ+SBX~OZ=PNsZt7h-NU;&zb0GmZ*e1$b3YEJPp=B%Hes^nAE zfW8ON)2DX_&uue+3%jU^dX4d8?i}dPf^%=!qjCdwUIfp^GW(&tWiZ^!)d2oM)GTMe z3I4vP&R)`k@PBH--6)$BiYxzAhxFVAE4kd&|gR+V;JuSfn?GrcTj z!VE4GEEX~7BR!FI`9r$zPmx3)#D@a4)K#!pq&e$(?86e0!8Wbp=}&*Km6_m|8CCoU zqSxV3KQcIRSKMXK8VFQG)~f_&TCmBO<)uME`*G##`o51quu<4!6iI(9o>|3ZY5oqC zhxeg$v-=Frny!3tU1i$`b;O!^|O9M2gTVVw*_ zM(oh9;Wa6kYDLEAi|j{L173}9Q<;yuR%++J=T5Zl6I4^xc07xO|5>q}1M*6&dUDDn z7q=bU;rB^9E@5=X7bbk8rCfB=f-hUg`A>@IjZ{|1zzAEc9Oacx~K1{WE*o z;Q8{GUv%is9-UwCN90~S3FjNnhWeKTjV7RE*jB^C{uR}?Kg;MZCH}REWR|Vvp znVq&3rQUrD3bHHxySOO^h0(}EOW0h}PVP%T<3PJv?X=4xxRa8ocq7(2DE0O4-ybdP z-eHqaK+$}=;~{2W0FQlvv7NQSkQPS8`k)rg9eMlPt#li+?75@hk*F;8fQ1^j*O6nd z=#=ppfMs|9T);WcN}soQH^n~ou{kMOIlsU~aPNwu1=+)hpcGwL{yCi~CF0s34d;6m zLlnl)$aUT0!8qr}@k!0o*s1+*J=UbN*|+6pVVgyoQ?^!p(8VGZHr=k@fBu!f-znc8 z=M?tMOs!l(Lnx4YxHJcI-#pLF^t*FJIgaC`9{llBg?#|VmvLgS=OYYX2R6uU42=DM zyznmdpAMwX5{dvFez{6qU`^SV1POocg{menK3@h5sWgb&d(e8l z%hkffld|3rE2akdy{HY?=aTL+3!f7$^UwsKd=l0h-mNq2hDYm*B|jhE{7T=>+`t<5 zg3ozRU9p>q7nKs(WeZHO&il6XWEXU#=zfjZB_T;$_xxa#aYo)F?6HN3KUB-f&M*z0 zLZwgc0|E||XZ$wgoT}KkNtlq0057GwbF?&goj9W^n!IC9^QPzEr%kqm76%WYbwvUM z_OjAkW-E8DS}a}eu7<3!j^JEuobQ-r6EEaGB|1a8GsoTLl9}?Xj>LZdp7e?)G!UE- zRV=ECRr`(`Z$LM9N7>m~Tj24U5N7SK1dNyaoviXo%9yE>U=KdP#&AnC8}Dfyxbhgf z$K&N(v-KvD?0zy6Cytu!e$BC^sF-oysYM>w2uA~dwyalI{__b{&t|uqD;j>Is+z#Q zeXhumdI2cJ2gm^@M8n_s7>}4&&QM@dA-?9s8&(WH`;OVTBDA?* zrf{Wq&*(Pa8I*{z`G!zkjz8=g`upz>Gy|4e8A1`Cw1#|`(mcNKV$6!8K7Bl)e_Xcp zHElet1TEQ#H(wPnzNP(!0C-EC@Zjr7xvKs%EipbkK10qUz##IhcpMXmiM>Aj?n6VJ~&JJ0mj$# zmw&v!V~V#g!)h+V?vTALTjF)!_;~5T-HFx5t@QKc`#iO$85v>jmSt!9^YBWnih}HS z2e3LsNS}f|ovul+^}!)Jy{_3`M$Z)*MWx?C4`xFj+rRyLoa2n;o*{PENq=G4odt$8 z125@`-NVxjNCT4t>(uy>U_^V^X_1{iInn;_VPlVAy0n+)#k_~d_H^q41k=l+7^mfm z<7ThKkzN6}nt{*EWTXnm@y350HolnM{Kx-YN^#kbZnY@ewoDf!ftr^O6RNvI?sGas zYqxOlc@Bu>uq_#2*N9n;-a`c8u-kTXZDE6wywG8A+Dc;lbO_cRkQPG%8+-C>lY3d` z>Hc|s_uL*o=p!tSC-!6)fXx0eYq*z--5?5BBgmcY^TV-xMEQ>{S;X)bjA`F_hSy_{ zEc^6pBn);VQgm?ioC}zI^1WgV5WL-Td1!oQ?mSd_{}`O$p*;hU>%{x@-tXY>G~)Jp zW40Tj0pw1W@*u8Sy?f1BX_jsLT%AvvqhcgKz+jrz{JZa#`cNnXw&kpk36814W|J4^ z;Yx2l$5Jlu65X2LYcH7WZ2&I24^r&605riI_CBn^((7UaDg&%`zJ&KHd?-I?vGuV+ zvU`gH<@w@wWW>JW=#v+qZen_%p*zDJc^&-$%%_ilfIUDmudjLIwz#oZY^~{P9Prsx zzZL&?K8eV7HD_ffJ`8jf&bYz7_rk~>+bc+Lv2bTvX{Z174WW^c`t^~~aDIK1*=L_; zcfTD;1L2?8h7J!8%iX&O4wg5BP1#}7spxfbOAUImoS*h3jXpwoU_UKQBVz+J^r2{;c|-$SiW)|K`ZUA#=M{sPhTKx=7WE{;I^bQd@nV$vQ0HTDl2R;b)GsFbRboCW?YfjujpM%aG`?r_GhD&*R17Sx@i@y1NGwZ_2>m_yIS}zIMkxC$eZWV=`>%gr&yk+ zz2J85{&$2a?|1x0loO3U}A1OVvt*wQH`w&~38!{5GkAO1l;n>XBLgM~;aTp`~ zp?=(3-#?EYyW4HB%~XViy>3bb$XtZye-D;%STHTnVIN$$%vsrge)6yXB)SNTQmU->^C-%d=2Fnphy6Jm9|y+2FoTDD^# z&uc?PH@>dQ_qEvNiP#k6@d3~OL`nq;` zcQxxBWB;OKe}8UP{l;YYuzu?tow3s?R0J-YIzkk%3mQ z4y%PXCc1oc3Iu-Uc`52tXt&}OU~_}Wd|jbGA^io@<{yhsy8f~&sM7h~Po=vlC{hlC z$l*K@AXFk(IJnaJLBG4MQB4$ghx(e&8S?NCxJJ3AL1%y2_pJrqcD{PkIvUXVW zv_q`h1!;G1_*--kJR~Ae%ZJd^s4X@rzaV?8qMHwlt$*Ns`OnX^oU+dX8x_u{vd;xu znMKn0t><8o4eaq0Woa?%vTZ5|>q3o5Y$G9^kHtC~CxC5~;fLq3CSh>Fl!|GBZLHSN{7X9x-()<^ge;8Ca}O4I5l)dr4?Eo=~YNA8~YI**FWnfVe&t zL=kNcyDG)G8$YEAmmR`iM=V1#x2HmxT*1t6INxD;xayltitu+`$q$^ zK+ZP*A73>xT;KQ=;Tp)zlToq$ARi<38zlAdgBBg%l02tLa>j!?zv_QJBb){Ctr=l; zHvd>;4Zb1{({RsgKz-J{i}*e)BsRco@maG8A-%6Ou%oZ@8E$s$yOEphZ61Z-wuQAx z@SIPBG>`1cQ9si+1c_?~TL?DGDglvI5i8|SQ?p^dp=9cN4lX1uJMt4~5G?R||5V)P zX`!4Q2oHF5h{N)eT5QD}Ke!!2NLHVV3UlNU!Vt9s?0Wg`fx3MKn7^#u-MdnTbc&E- zd&Ox5rFQOz$&R{|DPKg(hA}KLgz(s2y3C>$jfxZ;TOMxO!ZBZ0x zi=seLH*HWLbVrVy72NNe?NXNpUa$G z?rqBEN28uNYOkCc1nf0;d~7=5SzJj5(;uUNj(Kjkr-)eq#*+!ZNuuv4K<8r!{}&o~ z7St_|JLW(jKc@O|)FBoQdb|a_G_7My3VT8oq{e7En5$8!S-?XNI-};9HgY)|8qr|9 z-lVmrgu9=Q9096VybsStVxD_O0h`SZGe4wjKsQsD=+$I?fN~FhPv#>i(rkwVW^Z(} zb;ASY;dB5?Dcz%s0--P}qiebc6p80CFtOlAWm?;XctO)wo2VmbfF>M(7xT zHs+Vr5bFoOke1=af}04xt-6gZxfLM3&|CtgQlWo-84Oy9ID|fkxF>pv)1NLOtVe_S znkZC+(C!~wZ8EH!Wxv}+sqtsEWyO4#|Oh@!=_C2KtVsEA8L%c;cLBFm@;%~ z%$IM%T618cW!}})OgBie)#yTRHD5yG1FBqe&4ghJc+UGCRn~S@nbPS}6FfY79eF=R zg`b#RPV+u&_3ULEL;$$Y%*W1dIcJuzc@I@|y%s&m2U}w^>Z}FgOv=EY!&?mSq)rb) zJHb^e3+u56_tO%Z6)lzR$5bys_}EXO*-xMX$?C1++A@#KcmknTVH`4tT3cBc)>03r ztitwOm1^E2$Qn}1GzuR&G{>m421Eat50%Uz7XIEFLZ`*=3gcN`q3{ODJq}`l6CRw` z0J7g!m`yimPpk=T&GlJy_O{^R%)s27oWK~ruq%a`n!WwPtS;y>je#~ss|4CY-~_Ia zRf*LA8Z+6NVRJ45(0?@XCKJ`(tQm=JO6L`^N&(SI_g#3hhm)Pqjv)YjEY8alx$V`O zjp$=}^>fWcJ=sst0YG@IZYO8K>8$L2MKCN&1((DqL0ZXL9_0g652kvf8ZCV&Fr3MlxbrOLiN zpHyszjD+6q5H8bU7kun#1<`mNdg=7AH#~N#-9hvx%-Cy#*EpJ9a;46X2K|=RR^(ot zLv?0LC^weymeYY%&S=tG(j0fLcpCKq8t6sJ^4ql!!FZRAhTy&hXco4>6gNv4|(d0Dsl^ASGD*;tM=cpcll?raF_Z+H$T8pZW%-{>kSp^^s;x{_hm zvD|YdBwziJ1Mfi8N+4B8j`f11SBtg@3)DVK*v4)FD;5Akt(@X{UUO(Nn+p(J8i0G> zjF;kK-!fSdR#skrFrdK0&pa{6<}=IL&m>hT4&@wLq0na@fw!4-j*azrzq%YU$Y}0s z-*DN5&YQbjWUk**g9*?gweH&-qUnlUTH$GNI>It`yMejlJeb#GMe!$geqt4vuJsU7 zqvlzL%Zp!YNZ0(-a;;o|B7rHDL|E{Xk(ZvRi3@S0N#!=Ht8AN=pRtyt3j-_7tXltY zc;@g#XbyKFlOH7_?G5mFJ*_)r*z~CAr}p6ROE8hP~Bm#5uECCouhD}v0F&o*Br?W(3Y`iYLG!v z8*pA>BoMR#RjqlqgY$knGW_@`_6vt>raI~)oCd3CphbJGh_4^rCRDEQ-GJL2Q4~=< zinEXgLs(_a8ciX*xa01N7HgtpdhdOQ_n1Bbj2%0}7K483!i4?T+9w;Lcx z-0E~q9X~pv6seEWk&1~p;?^9YhJs)96<}+-_u8j*O*KqUr-cA?!}ug@S`#!$(#4W) zU*0zw{=vT5Ab`|m-;Sc+EBy_!OpDf zS3`$ns&N}QD_94Ug-xpGO2?%|v1&dNEq+^j1I-rA(|E#d)=@B`#o(|Kmvr;u0l{u? zLap@%ABP$G_1!r4!U?_X=&SyukxLH5xlf~6Yrm~RUg23*N)XJqtOK5dh|M?a4TLaP zUU~MLBebt(}f#@hdyx;dC%7Rx?P7<>sww;|d*#FOVHJMWA`v3{tyz9`(*q=V-vTF9067AC1LgAMBxl zKSd`%)*#!y@VFBfF6Y4KbOVsi_d~%1Net#xFass1O{BF1+J(!TLjiV=wZugfwLO*j z5ruXeROx_gTk`;XRTxB%odAGg;=Bmo_J>WH>NC55k&fScK7y3nET7z#Nh z#|8LIWgu~LkkJ=#g~ggAhTfG2QTsg(rN%<siZE!|P z0bI~01JwmGW>U{P2;h4#-T?=juWS7^2dWx0;0oBt95=hvI=CneU>!&O6w#RCgIT&@mVziQK5F6-+xvfnYyVP1m0 z)a|2Qj|y2<+?eIRy)V;bW@m7hh?d@I+J>2<@9)VTf-NSoXU$sW7 z;i}DYmnjqsQ79h9E~0iZSG;gM#15(B~3_ofRtRAQRc;gp)$R?IlFi zIk}{aIm3BW3a~BoU8OB>)^vL|OQBXmQjmjd~vDPUH3m?<2)|?MT z{%wwhhp2jhU6iJTf)`dGl@39Dh|@A%b6o#&($h>`t(vW2HtUw81Pxz$vYxUWF}3Hz z;D`b<;ttqSPR~j-aXkp`$M6|ndP|@SR&;Mf!I8Y0c+fUp>Oe+{3YsIy zniM6(@M2dK7AI!d?QE5?I*TVg2zGgLzX1crvlE89Be5Q>W+Q42k_${C)UFUgq8E|D zI>oX$?vt@%tjvL)+qp@}90Sjc?s&SmOdHz;xUJa)suMcsIo-5jHF)bhS_@%6wWmGw z>rN>CgvNfxr+o*alfKCJ*u!?a)_HMw*_ZOl1vlX`^jx{m4U-z`xjZ$KL8o=9H8G&& zz)t)A!Lu|^?HFxe?u&_;EYkjp)7AhL;6~0#shHX6nM}JhJ3v>^Sw}N-JrP1!1gA$) zhkRxGG&S8-Lx!53oUI}<9Iu;8w-&!xp?yLFoC(!ol=DW*kZY|^^d21KAXp5A1L%9> zV;!Ka`=K;YDRj^3#C`!g#Bnh6q@s_MwXYwh1M!%5gA>>8uVPEv^^dz+``HC&W0gS4 z8G>}RFPEY(si3zCXWNbnYT%mo;gNTio4^R22W)ku>p0ktwtaO#>+|)9ZBSN-#`w^# zaHk}3X^iI6ybDGLxue=N+1v1n2fP=39IwpDq%~hDtu;t8tgbA8zp1hP8;1cKhlN^= zjglw@`Db0kG3b@L80J2AF|bYeEuZH(Tlw;6AyW%KGoX|sd8b) z3xFg_N$yq$geE*x;WtLrXz!2q%(fH*>%y_vrFxgi6}591cOu0#BcS3PL9v@5z4If( zf^6=k-xi2qC3Q1RgOSB#ScoLR7*k7Z^ht5LZ4W4K6C$M|%pt)?3qyvMV`vT~nj<>1 z-l_{^@+_gI>k$gp^F_lO_l}dEzd1&mQ8b-~SN_ zB{x~Qo25AQF^1JN?q-|-8LesTPez`i!!FB)H!zRHMz4AIhq2?0Wy9YqYj8zUHDA%d zY!I7KTZx7{WvN#US$D0|DyhIrp_jG- zX_4))8p6yM&)V7>+OvAj8vZa!8%%9hsCKAanPa}P_?@(<{ZRo4oiz^YUd0Gq`m%ry z#tb?3YS;8TJJ+^mmEG;tI&A1W?BY4}%S!=v_0U;CmT>{gHfpY?DV%vi2r5q_$3*KH zlrt1T*9~ys!P9P(va7%@iKV;bBJ#1G?YlF$9!<9Xi9rEM0OpK1Cy@l3>3E=Rp&rfe zh>oo=nEx@YYgG>G*<`&k!HMT3E%se#%}j)DY@X=V;F!3zzsf+t?dhf6d8d;mCOm%5 zOZzx#t~m9~)$*eQaJKemVL7h7D$a9xb@eKMW9krP(S5XR!U{$l4ff}0GmK9wUB&cI z6((IZAa_$Yi3XRhez2jkAW$*@ef`>hnjTMz4IHYZ9ZD{>&(C<=yKG9?L*IG7E{}Oz z*b-xwn^yb9xs@6w7S=vZS|UFe;&Z$e%jIT4ANNxZ@PUnj2ersST4Ly z7o91SYhr)X9R&Ubx+e|&M5>V6iDQT1k7(djrMas}S88UkM&C~%awBh>iZUDV@X3M3 zheENS`~s|ljxDS_Q%Qj-Hna0J2#-vu>;=)6r5oapJBD+5h&w1pZmB6;n zQ=PA1AhiTy59W?Nb-OP53e>qpcSEm~Lp1Km&E+Z0TGJ#Ma#jbO>G3p)C;A{5SiX5& z!#QVVg8}BFffHQ({s?*HW_D!7p|f#E1~9F)ZgI?dKtwJ29;fiZ7A)QpSYu^R)oXMW zR+QnKmjF<#qotik7sftNF|N}Qx(X0?UdAQ8C03Ua)}~1c9S`H&M3k(!s@;kjQpFqs z_;$NucnxWL$GFtmCP>k%O|m5?bPB%M ziMT)qw2uqN*r7KGyeN>NokM{H8^|mQ?^|d-(i8p=T%g5E$(@Dbtq*3aCNzgE}Hc_)+owRjfPb9lDw9kb*yw^at~0SqEje>_9EzYDx(CfY|(7dGyU zPn^X?Q^F~DD`)k5I0(nKO1BKCZypDXZZ%*}RyZb9s`g7`xJ%7)soP+eny`$5K6-mA zEKxTYwi6Zde6(*4`$i6d`hyCN0hB_dA>}~$H=NYG(~)q*O79%t!MN4`m_G3vp{ z%OPSm^BT}OrUw(~cI}XTu9!t1)l486;EULGXV8L2k*;d@M0S}VPg@#`%MzBhgOh&r z7nLh3&`xUTrHlM;sP-oCv>i^p#LD4(1_NwV11WH^rQGA^qe&t5L;AS&fPLw=ggGB2 zgPqqfmKEYB;BmK|p%5@S(0k zI+nUY{R;l=bB;U+q+A7Stqtg<0$JYyRD`M`495tbBGgMy&?Q*z$ZoiuPn%wzh(5g1 zasp=6)3r?O)+D@<2yhE>cvg2uzP5Ln&Y;`EW!C~mT;$?1T{ zE!5?)e@4I)gl+rK^}16M-}g+>n!x_B?i;Bla!JV#_t_*FcY9LPg<|ee+Z+T|+01HP z@N9T#5~XB08-9yA-_i^E^9kHk$Of`R-p|u!AN-yZjILBizxv7syz7;4K$)wxujaun zoJ>GVxvyTv;{Mc6j+t{D2&cR<2Hg$22Ys_}`nuy}!9H19N(M?{8AoyE5=SwITw*2^ z$4uW>LFY`IyWP0AH@#{yR3;}q@%Bt9Pap=q?DdoWcvD^=D=Wf)%50*@q(#H!bZm?_ z{h5EOSp=AVXR8l)-3IByMa|6?CA@A{@Jq3wm&N#mHFKLVrSF;TWf<6KGi$x2;#Yt{ zK{1kXNPt4=kl6Dt2pN~w6 zzBOs?*FACILM~PTae4@&t^-87ReIi^ERIL`DW8M+W(w*RrA#)*-DU)2sOLC478)dr zc>>sbM`>N6SZQ7)cq&qJW5);z%#R`3zZ}9_*amUw^YNjqwlvRWvrV+2$)=L$FwBcW z-zh8wpOKwO%JnlC28%vt64Rx<#`7Rb@NgT6pq+p~#n-yK5l^`R1Oui{2usHn(>cVQ z&A9e+Qj_y=7o7P82x80GY^T8tsvVhIeFGF6bU}P^3=b#SF-_EaYhTkyI{^UdNGM3& zYJRQaCLI<>cmkkYMS#Z_(}=ObTO@PKY8mr&&Zbe(1oxgJR)al{cwL-1a9fW(D35Iz zQ1}`bq|5#W>=CWanV=cT8DJ_siEfn3s?nLmE~5-$<22pb9lNRxc>Ml442`qJDO7KB z9*#RzXWQsl;GbIr1H(~qoawx$zyly686Z7tE~X5wx`gtE9PX#64eps1j8K8uBgdey zSb#ulNwnw2-ez65XHLM-5)#)?T8@HFQv@T5q=*q?3_`a!56)szpfnm#%#!iuJzCuB z655&CK1zhFHm{0SmU!L~L^?n=x7m(Us8*b&laAg*8%7)Som6&#=qkH?LtCiW#jF=M z7E=^|cNxWSq$Y7^=v)j*8RZ6c2XkdVS0H_nA2c~ikyM>DMOaTuxlAwQvpld{4?Jc#z&B@D;`k8<6qI8nGX0mcfN+DnvO z7qJWEL4*R!axW|nBkqh$PM3h3&LtE&mS+G|D!NYC``xn%R#d0f60H9gsze4fxm~ev zkrNgaiO3kX(Xc|HirCK5ez~7qoZu|%wbl&2%5=@Y@4)#*@F@1K#;Fi2W0#`iADc26 zfSeM}IQBS5(8e_bTVORCJxSN7V>=^K;&D-o5i^^udAaD%Z)5TP{-SpkdDc5zplyI; zqzT=Y?Xd;;JLduI8>#)CDJ7H&za_-uG?ZJ~?ODH9O6L|KV(H;Ss{MzE7)BC-sD7me6W zagq^^tfylwJ9+A1vMo*IzZFs}{3V?rW((aEPd*Fbvof?D6@m&sGZ?f?5*%C_xr) zwj9JaBM@?lZbU8g^JwX74FO#W#b-pTEP`5BTX7=pZoW+E^M+=f@!1y$c}%W>Q9H8$ zKR0N&%*$|9?7YPP7+%!J2k?n!?Q1o56wCFwE!p0Zs1$q6)_O74wOv}#XB~Nn2w>Agwyj;;)GY#S{yTJRcRSa zkk6d1>+4kk;on3aHc*&Ug{?miIIBU6j^4`{YdQ24p#!H-f2Gh{WixKWx{pIQ15FgM zb8<{0+Qy9)aE3fH@`k~LD;0>4ExSZ3H1baC${#+Lf56Z!)OLHD5o?1WSbJ?ns7-iJ?2i-E?&tZ8RQ&R4|x6{&}&Z7mOMK+O`6P^umc>X z4&W79YLG^fyQ9-8&q@9=_v8W5X(@Zv=cl81Cd17^>{)?b4$h@qZW5l}Z?K-x>+{}D zTkh;aRIs~&p$=`6t?jT8ouam?u#*jhu4MQOK^q2&G*GKMA1d;YQ{_pH&s?7xIMf{{G`CZaRY23{a_Qa60D>Mkc8dLZP=IaGqd&{UNmo2NCVMO4{ovPNUoOQeF@o3lVUivMxc`0u_<9w<1{H2HG>AXa5xiPBw zHDpwRh{<;t$=o-~F+t9RYw>Vvra)&!q zll2a+W2oWB^QPZ#9ic`sGxlOn6(;6*Ep_Xf4mQMqpV+Sju=gQ-28(ty8IMsrt*Fx2 z5ryMJb-8Se8+J0j2oB6%y}=0LXc?TFK(QwMPU~XX@Zl?AQz?<#EwEbvoh#*V;3CwT zdk0`VcHpj4&>+(p(pl1aaLFpqY(kaF$Ozus1>K27Zx@@<6c5cK@{-#3Hy!N=#I={p zyf+&F)8TgXynliWUA&=CfkQIxRB}79hNhGGcJzhO3MYT6O(biJu1_a?3b;j_+OyR z*V;g10*;Z(u*evAwZWB!A?v!sq1?8nyTf|WU;^MDiN&gE*Z#&JJ-ADFZ}@vLM(~jJ zGw|m@jZrsd&ytHIOq(r&tT8u33h&mnF%|j;nbZO(mDnfgQrGNGM=fYN(B!lN(QG;i z%+yc%k*7dgt#6EA2UjAk2z8`%b6 zYutod0$9DCj- zkJhQi{PJyj)y_B{{tE}t{<=ez9+mri7lmou>%->?IZ$zE%e0`7)KG*N=Y#}KzY;^F z`CI5VhYbhxq!@(&MjTN@M~kB_p6`iI;CAaE$L>@qI5SSSJsQFOF@UedU#I9+ZOQ$2!PUJOf|1S`2B4;__p*(O5quix8-PEucA<8Ep zN++!lVfirIA;p++`;*>i?IL0;Y(SN3T4yOxX2U_ONeXl=_CwexDs7N_Hk92_Wk-iy z5b2Gpa|)lGfZR-a>=9~+;3Y4B<5@(6nziY98fV3@4Y7@cLI0nci%ukj4r!}fQfXz- zN7x`Yc6wK>)A4}w6a%#=`x?t*?QmC+_67LG&hbDiT(*c!L7rZU^#BaBX=2P) z{$hajx-EMkGm11oIC~%VA`n`e^*RvyFJ;s&n#Xl?I(TL;d=0nfLxBypZMxoUMk>X&N(m4?sq$LHlCZu-j&|*tBgDHpGxo7J#?Kv$O9jch_94K_9xk zmgTc4jb@|<0{hBdygpY7tg)~$)dX9@C_|MRTAJxRCe%g>y%iPFNpw6zl6W7uM_~aixzSZi zR>l;uSxzL8by@Nd#!@H=&gjaaQ(Btm4QHrRIAQ`lu&A+KH@jyZSocoEI>T(`S26B_ zh&q6rtt7)EF)EF%xrL=y(4F#2%A9yH)Lqb*JPim!YeM{g2IL`7y<>9)CysoU$@~G+ z8FYe<;b!30pe8zAUpQR8+n5JFjmw&|Ppi-wgx`)~0rO=`W!_4gHgBd0XnJ7_q~N&C zgQ>FV_Ga)xQXyU79COrL$8D%yOf;zJ8sJeiVriidSf}!n1K1M6;jFaK!@b5Bmk!~= zV?(r~Q8;W$qk+&+gpE!+XN%_S8eA?_Vh=du+K=hb7RTtp3I}*p4%%mQ#JNs3_9>x+ z(Li?2x|uwkp87)&>jq}cov$@dp@z=2TacERbjo}*uCyb$qJ<8cQ6mlxgNNO6+TM3p zXh^_kYPKbon|bM;x6phy!Lc=$m6ytfRUPPJOo)rI%Wy@NkV&%;6cijQp)^`Lq^*u2 zX{|K)d`pMGc2CEa4ic`ys(|jywxm>z*6VKS$D19J9Q315aCPJ1;^2sj^KQtmqWTG3 zmxT|Ru=dVG48U5+HneWophl|Ki0I{Dna)!5)0%pZ5tRw@+69|STR3B=OoQDwWV!HK z$em}F6}rbj%L@(5@U`dI9axpNvXCEgNmJ>!P@i@NjTRN0m!3fry=r$7 z_cG6~$q8hhg^`!%9!wodID#ysvJHDqUte#J7f7k}4##22Uc|FTk#^}Buw*=2nNiIZ zlq0ABx=wt=K%B5aLSKQfOeLLK;clMf^7RJoc{9TaHUoeC!yaWhx#Rt*@jPRe-K&>5=q z&T}%8t&@m+d&tig7qfR6LpIHm;DH1_aB&t3C+Dh^*bKIjNUS1pMQVF2@bDHF3jP8R zqOjPPCbX&Ix=%#6!5wlGuRPhLv30hc3I*wf1w0KbZWz{5mv*s~d-J501Zk_&(LB(q zX`rF&Ki$s{mtwl%IbzH7D6Uf_xv_FjEqkSqlfwwWINxbI6W)(}F+(Ts%$);iF}jMo~;sRj--x+4!)9woI)B zs|>N!SXGtKbK7Uz1sOG_I=5{hSj!=GG9omSTAG8?P!l6n1#K9(T^-m!h=Cpu z*#5b-zR)a7Gpr50QYsKrDr&XX&Y2pcEMqQpomF%lHfC)}-Oq-3iyt;E9UR)LeyekZ zRX00Lv=13muCaKzv|iU3AE}_AAE5Qh`5ch1fz(VB67h%GkuFV|E~cv-bQ2o3!TR9p z!3LGa6g4(}Wlz#HX>F&o&Kh)=j;(76&kr=_!j7k>@z!816egTuWv4xZnVsp$X?*T! z&a|39-CfaaY1e{DfjTTVI^U|cIsu(CSjE+#|&|y`30sjxW zbS>2UfQPr)Ah+6ZV}pGV1vSHENHE><2_W#`Y1=c*ND)q8C^({(gy#YDgKEF5vPCrQW@iy5zCg{JHtTsVQ*mda zua(+x?uVSSi<-uiUhlHTmUV%pOY-B^B*|8{*UOD&F$)+-9HRVT4Q%M{y5O!Jlcrp` z$=nHmD71I-G~#D{G6Urx@6JSh08Zv|cv$gnv#ZQN--Zv?8NgSF^}=_~a$*kF2xn0d z^Vr}VzXx5~6{Jvg7QVWGb0O6X^cqh=+tVQ!yz+3Qa(g#VLwb^LZ7#z%~3)kwP1wr`(6m~?rdt%TWh67=o z_?x~YwP?PAQaa45M$Di^pi^`)n^S?QWUsi$(3Q_NfJ%!Lv?WfSm_&m$^f_AzuOAgk zR&B@J(M#HQW+9Bzpi0%UTlN+=uG1f|T)QMDX*hEy4IQM3Fo4uZLu{|p4rRCW&3M#m zj7Ij%g2fsI)$2Z(ry#h&Q%OcGvt&}VI}p8J%u0mkin|8|SdGeB_d{4LQBrAf*{-)Z zNm(?(6KxAbLTZqRHUgzJ4>A9;6<_4Uh*$H{DVi7&N2>O<3dO~PU>j-#kn9<;AldNX z;T!6zX`$KJ04rxT&wlw_XD}UOk>U`8btl_SG;A?w{e!GIUVo)#N%mzA& zmn`XwWq=w46V1_9sA1K4znliqbPCy5h|Rhb|> z$})&drEJ|N7`m*O!(%4d480aCD=VNOPy{&0WhR-N&un_&c`d6wpQ2mlhxSIKmj%46 zUr<}??w&VVQjewp#tzAPY!qnaO~U!lqxMNIRVOP@vL=_ycMnN$yPX-S8LL(!(H zjg}YPj2nuijN4VbCHicv9r&PU8)9SOvZLG}Wbk_+--4R33|B*^&@(Dcsn$9ibjN}T zhBXW<&^VRLt_KZYXofR=3>e`94nId&0=1T_?yDKitFz@$(1!Kk(F-lS5mk_ld9i~* zP~VM9F)NDd9G=e70a`EvMePsBULccROOXO^;g&c^wL4H?H`20pJwfsNLf7xJ-6gQ| zx4$oY<vGr=sEht(=Lw;@AQ`F$DsoZ!;+h0L6_nBAZOVX#9fMaPai?TU__U*GI{ zeW~`ZGokT>ws^2Wy} zpNu;zQY{jAnORUh8ArB$^~7t=)umCHo;p7s8}!I+z&q0dwWKI6;&Hk^qAP7F|J{)dl9(X%`xAP^~L#|RFIV4%&G1Vc%Oz#3jP$y`K6sgx>OSwRN zp-2)<8?57uheHdDj5}C!P;iCHe!~DRPzp6uOjD*@#FLQ}(GBQ5&!*E3H`tD9BN9;q+XryJ z6EwF;}X;uw0Vk}7n zIS43{rBU6ss-p!0R2Xt&twGN@g4}u?2syWA+tUbmod$%jiCH^sc9Ls>sa?;O<>lu@!WyC(}-c-ej7Nep3*vItA!>feNC5 ztaqSolq4xLCOQY392K}H+Z;E{SWy1X09g))j*M&G@FXoWE?cu0$f32M3oU6-{Wk0{ zfkm;-Y_3j*2&=BX2H60y{AjjZvFX_+%Sd-@*q=5HG{JR}UetnB>)B-AUx8;Lj1jNJ zUfLnR;p)xjTx&2{=1`J^XXuQi!J`5!8s1-4-hw9Ge$$c*D#gWEa)cX;W4;5~i>?AM zx>_i|9zaO*fM@~fA+br?D=4REE>1IIpqE2|u-3sn=%p<@2;i{5muD4r#^V;61i}Kn z7s~=*ozC+;QBX}z*4)NED=?o5NN?@g^Lz`FB@-?^43ow)oYAQ)Aiv2BkIe-jT4;qs z=*nVG5R5{!TFB$kh^f#6N*ja#?j+^U)okwM<6aL>=^E1mPSu8|xUmWu4hD}=3NU|k z(}|8{6m&~i&W6|T;GAceHZ95Fc?0SkoTy0);0#hoJrkIS3{?I&f^Fx$k)`Y}Cfs=h z@KLfYTbRXxhC=S1j8G_VQ*=1Tde)2seIYLyebxg7 zjjT0=IyO) zbPnw(+V0NDbrVGaciB=V?KbVEBs_46u`{Dg=&ajq@&Jda*0O@rWBbs^A@dG$V8nFU zUO_Q!iyVi3hd?!VQM5-bf^ERQ0)i2=i%6PDPu%=qz?;+uU=f#xH>qIMLaXNSSWqlk7eE=%xc1Rg#II}_X_F2=;$vOo{@ zH{m8T#1V>rTNlb{=Yo^j?u5|}>;&D!D7J&ZX5@*7&kA-6%Qe*I8m!KY=i*xQRv1avg&olk=nIi)pYo9{W zG+H{yT$@dHemp7LIKfnHz<=iqucq}?cMO3ON=U7lnhfCI*qIJ7dmZOu@&&{IoMy9+ zuLFexc4^mZv(W!n)3)qeTwqe&3jjG`PLjDt`i8boX0FKx3QTab6Ep5~6I7IR2Kw|B z3NBo?=C)>Y`vBUaTT4Wi4Cr zI*7bdg5KTih{!?Lf#%R6sk5^l2;Mdm~YN-!4y z6Z2WU=65}?(~!s&B#haqDw58kL-gEjMZiM7)v<`VdpI;NeE_53Lxum0$!c4<_=-bd zs*g7Q%-JAm7_YDV?yLUU)o%saQ1E(b!&;#7VXc-dz zO>4LArG+K${ga(Cqiu?&y+I*ENa2U;3s*iT5m7;6PaOC`)N#3~gSQ~HS}k_R5ls*} zlfJrFMpX-3lx71$v-%~P4duAjhg8bgJQSFf0P zJrHf62%pSz9d^>xx*YV;UP4wF*Ora5$2sq#hm zo;)ce>i~97*F><#Qqv32jb<=``v`2>$TkZe19SmMF|9IT%Y-Aywj=S-rE;SA6jqnS3+54645dlRPWC4ra<+vu{MD{7A$lUH2~D05y;x5<3Py(#u>dPjGokaCt?t&9DBs7B@b)OhUWC3 z)v?WGTR*=jh9wq=Y1-hh3x3VpZbu_>8S23sds|r7bvk|78RrQHk%nw= z#D9f<$$}XHM!!w_qAy(ko)WMG5g7tDO;KpK&kd=E6+L7#9bj|Nby&Nt;~=vYL=KKC zhs1njb*iq1w%17g#XWw&fk~=*y)(H_0Q0@%q;cA07+57n5wh7NC0&o=grT~0w}5V8 zYo@2bfOKWJv!hbU7U5QGITwT-%q%0JHAAkNCW>%VlWZd@p;!+3j#`;hPSGwf+5mv2QG^g3mi8T3KA*{UyGapP99D7%6OO6!47PnFn2%V%ErtxV z#YQ6uVBE+Fz}m|d$zcsiEEyWJ;9fd`jhV88l3d5BPzogFCUD?vq1TcWb13AnUK2AQ zU%{TPMb^(JGDj>(>~L;y&_PpxZ6h$%Dj5a{K~RxW+C*5FLe6l8r%jp7&*r3vmsm|v5nI%eEcjo|qqMXmW(Wo6EPp}zMg7Fw!ae#>+&)%tE zd0d0DG4obLt$pe2Cw#Xqkzdgdlh~SQa$XXo3NC=@Cif#}s}_b)lsjpYb?j@VrfWm6g{pLdF3T zx+qLF+K#jm$VR&$tILH!q>uEMsTcy>fgEr~q5|I?44F1d!Nn?rphUO>0IfnGDYA4r zsHm}XkQ{d8WZNi$CC)X*jhiC@lF|cqjwA<36gzf`UO-QAkLMF@d6>Xwiy0>! zBVaz;9*z^R|L2zyAxCL%4_dnm3zreo9N7{n|rf%^LR<6A8?aHgDJ4t=e((LfCy!IocMZtgp2p)Qps4w*i-0w=W1FeWR>_fB!^($C2|L zZ8uNJXnEws8d1|7Aa&UjVtz2Cque;NP(EBOCk!~eha>B|?t z8t11ZKmFK*tI+A?^gFrt{owo9iub|fKR*8Xcf9w#A9(KuK7m($!TZG=USv-<6-@C_ zo|Y!v#wouQ{Qv#ySG+&}u=kHY`K^yWy0PN${?EXN%&L$qt3Afj~!GHQw zPUyvVojAqydFlO*_gB2%t;hS>?M0mYH}yHb8`S4_H{_4!`0n)_ceaabhLg)v`<3JW zytiLHwm(8~67x@bCTTljJ5lT@E%r6-TF! zGQYAvzG*z?>=5tyKN{YiRrBMUPe}hqpODFqt{3P1Cl~s6zu1oMFX|Wl_^X5TzupbM z_x{}bVE(py^DUS1@7@mA|MaFy`I~PWi2GxI`-fgV;{W4keZ;?r%X5EV*WX{=|3CfN z?EiZ&o;}_ZFCTXK%Bz?z_hWp2y`J1Jzf|8z_XmFa$HQi>ll9q$_p9j7=vrf^clWXP zy%)ECym|z`@iRGsSD!7fuctEB)9Vd-`#gK&dh_eo^~R2@H?1~re)H86dyW(Pij(KR z`~0;=|CfI8op-uXj7Lp+1GJozA8zF91MUW@6nn$9Z~W}p@82%{pT6&%PD%DLdH*O# zE+{)0&*Ri3e}HF#=bIireq%&8abia|S)8Y)eRDu2XQY>BNg9JK4O`?J&v(6#5<9)F z+>vR2cw-yc<)c&l5f0&|-X>PIG&fIJmg~(~*VmpIk#V?W*ug1|57`aAeRr(SZ$9|o z_7Fe#0N=zHZr{ag_Tl>74{wYc6fwe^U7V(IY2z}aad=~!AP$Fi{cykC&+XBpX62i{ z9i8rv{kpH~aXlPE>7Ts%Ozkl7jQXJB=;mbX@`I6H_xSwTlVidsZcb%<6QYor-k1j? z%k1pK>jVCS4|tsKFf~FO2Y-DFt30-i&wpg0i5q2pj1#{dmUsuB*3<2=KHb|5t^t_t z6;2EfG&4TrVRZK?1QEAe{osQIv;^!7ySY7hoRyhh4?K?S&)wJan;veD`-8qQ4u`nB zxx_iH0`_Iu=i4L06@-T14NkPQvHip0CbMz#U{G)X_Sp!N1Fq%H#6$KyVftM`Lj<>Y6tZ~pmpg?k?h%*{`m8BbQxpZ-~_tv`Hib;3S=;o0Lx zu50M)p9%i+?_9(1YpD3*ySMlG{UUt-=fhg@_& zL*@SrAHq`i_v+>3cB#8R)_jLE`K_P21U#PU7y0m)9xr4Q*GvG%@>jp^V)Nuszy7wq z&YAq>hZX0utKO4?eCx;FwEyixxOsmM-Q^hff4=dSI)33>st&|AZ@F9k=AW8{`pGE% zDLSY(U9-2Y7LRwo_uYqU_r?{)`)b$z#W!~CfAi)o55y;z9^Ij`-;b z?31YONdo&&l+u}Bc;_7rLAMjv2|yih--X(DvHeZv#btJbA9Vr%eTEO~;J;3gklNjp z#s$&*-mX)mIuAflj^yw*QMwu1sJ`J#mJNky8=2ZQ)5=f*y-O)0^N?HkR)i!8X;X4A zkXB+7J3qZ9Vb}cd@f9R{b;{^x_}~SJ*Ilac^mq{br?z3$$MyVX{Ul*Y?Zh@tNTcdZ z6uWp#;;!eD7%9?)Ym#@(v!V#81#*(x73rF1`N$owc}o5D`b-}4iU0f0-~c{ZEB{QM zJ|^Qoer+}Uit*P!d2KVFpU&OK-hcd6mfIYuk zKyL0nc=zYC>*Dmoo4@}3p`T`_n>p5`@1sNV-Y|`=_`MHr-s|~?-SrpH{oOc?ZUmuz zd0~X5pV{wy`r)&iyZ^pFGoEgtXSirbWY=xn|N7nl%VlPy?|t|lTzOsl)$hx%tc2IY zSa^PFU-x(a%OHhBMvN`s7_P59Jpx3@*LALUT}rj*-RJdIT3m*=kMG9rhZwr&*wAei>-~Cwe(>yWxUPTQeeee_|DAtky>tEVHi-E7b$GZF zzTNBaFp01E;jex4_R@0QR}216OsU`>;|CU(;NQYz!W90C_+s5zdu*D%x+Vqx7e4=) z4(|Vq7f~4a2mjSW8R=K=wv+FI-$r}))!7`p|L{^k;MY%IJ~pPC`1{4L-rZ_Hu5Y}T zL4oec{U;v#+`&(ty!`%)XD9o7A`u_>(M$bzu^Qk)JkIBvhk1GP>I}QN`RsLi*0{Tf z*98E--%aDYSMkqv);<1t`&&K5o;B}Yy)GH}{R@5-^YU5i=H_vI!B_9riMtKn+)euX zKkxsoC-XDv`q$lhRR4avvfO{>a}StpynOL1*Maj+;o+tJkACr;cRsQF;z!Ruxi%ti zG808w>qpmR;(2{F&>RjC^X+2ma&t-%iG3VLhs#H?^HJS~zxnGA3rbhrqady4+V>HfQ)kk^yLNnT^u zx`SZZ=?~v)zIU7A|M0zsGkvc<^-mCQ8h3;62~PU1NBi0BUq4+BI=+GaEPmV#NB#HX z&G`B|5BKKlZ@hY|zW(lKZ`{XE;J5U8=s1YGWB=rO(Qc0jXYusV8joMBkKOvw^~T4G zcPD-O;@z$M`peh3@cN7OTzGs9zuoQcwH-I{&XYm1esrBSk6-)>JH||U{Ql?exb^yu zA7|d(I_82W|!VRMHVdh)#Z-G}GJYtNgvA%^ZBGLH|dS0m2g zOCpBuL(s?Vd5_uhU-`~E?+CYDDq{yn*E+q0=^h$l*QgWu*H7DpF`9v~}`| zcdZXTxa|-gj1yXy*OKWD4a<5#tv_40IgRw1!q!{Ai&7sqd(CTab8ZjKSOdLG1e?f7 z3j>lK==;_GUbAsDRxPB%_btTh=C4PcUB8ndaYnvu;GnRGJhUq7FX2D{c}np<()rU3 zJ`*8(zD@FPPb*7ov(AX`dXhH>11-R7xcxyLIpY*Rz*G21I{(2ndH(=w>SIFx!SiSQ zZ8l%m9k5g6=IHR{hZCv+IQP2Kh;h3`Ora-PL$z)U6ct>*zya>?@TqG6wqxKcH&Gt$ zY;*&!!9Ixj*OueSOKtm=o-JCscqF8DXrry^9PG1?%Ek5Iut(H4>cU3d@icM4&?c|< zaC@om4*&v&uuDaLwufF|ud6Gu|v`->1lO6)ITO7t$ z%sBPeHuLIid)h#}&p7}0fAkR4-5CNxI{Y<`-SoekJ`qzh;_Qmy$ zcZ0(>cJ$O&n21k~=PP~v?dSewim`w2jg7+xUwln@a{6E9TW>n+Hx#q)@2v-n()V6` z{|hc+PY&!0f8|X_^;j5x+Cr+~u-|*yRC>HCpMLD!;NSbkE=u>hCVici_{orb0~>q8 zUHUek_Uhw#y!Hk*|G#~c zr|}EVA5S6pTfcm*BfMcce+Cxw_?d^^)c0OE-%eQYc*Nh@9^Y_Pv9NrhQF*=n|2T4A zD;W4^U*;~jLO(ulHo7%P@Y{&39&hSzZI^GjZm*^puSZQac=X9Y-Dedjg?)v>_KQz3 z*NwN-Wk0-zo{QLLr3nH)H4zF*xjya!f z_-n?TzkPQ{JZ<$qKg8zmKLnMZ{@{LmzR+hM#_k&b{Q<`8ZN#U~?N0yXVX*Hr##e{? zt=;Kg`)d#5Ufpfy+lv2QX->cR+fPEihxq4d@Auh5L-5Jl*I(~e@Mo_!2A_QDu1Wk1 zW&PJcAderqc8u$a&BynV+lMz+UC93U=9Y}~GJEW@KVgS|tG%HoSL@}&)q3q}zpXcP ze~G`9_4N7rmHq()ENfa|ow(JUq?6~PsZn#=?tqo|Q|HTm?O_%L|7ATS?*{+H!+iK@ ztO{|Nd?}U+`J*{yh-s z{oo2!`Yvd#wh_JzsNV)~34Z}elD-Vu_)GsCOc4@2SrQ0qyqp0~0dJ%|%{0QGod771 zaMo$RZ(KS-*}#8pL~5j;T;L)Hmg|o@ZLUd^O{xV+CWdLYpC^}(J0IT+^O{7vHqx3R z`SA^cE%ZIHn%~{PHz2=yF-W5sdX-6P1G$Xeblpn1QaE>YpEX4P`i?Jq2l%Uj-8as4 z%iva~H9__H%@R%7+xMR%EX0N|2eK33B+mhBc`|jr?lgLeMR;v^zcv!=4}V^3xjyf| zZms3n?Xq`WMDG{&+ZDB5y>EZ3e|`=!j(ZJgc0zaO=RmWc7i+#vu-ScG=9Oy6{fj>@ z*1QV#``X2t8o%}m$nbUndA(M2zm~jSEUH(P*I)Y_*xQ=`fB5Q`fxUgJLRn9OkAL_O ze0+)2=00NiHc}h?gy-j#%~gZ1Vw8j)vKZ8Y>MBxQK&qKlb+zbKI2p_muS_ZcfKR|{ z#lX8ofB*x=m4d|e;Z0Q6mg;g+7-KU)oMCH_Zm4x106Dt~h15w?UVjwv4S-p9IIk;a zr6!Z*38I#({KNU?SfX`8)PE z+z3Ct%=~1oe!UXXi~siy;f!THx#|o>$+NqczBGp~N}i`!$K&uhpu^Vt%|_z~&wd*) z;{T7mciDAa*|NmSsE%=f>bmO#7mRzen0feqr7JH$NqmT+NQ$KRDkRc;N~B0i;!7mI zA)EdK!yWew81PrP8O!={lWMKB!H#(%+S5+j+P{|5Vopa}|6&n+c6z?JZ4v>NscjZ*~9KT`w& z#}K^kgRvM!kq^-Em*!v3=HM7E>Q&0H z1Pjl3-wC5A9DmlP;efe!+b{w){e6{EBs2+H3GKx>66b!Ro^5B~@f^G_{tHdva1PA- zEpU$IpDB*|cpsdjXq^A)^>m(ub4Z~M_6tjb9C-Bs$>22o>H~n&2$Fl2Vvxf!KfT|L zASk#Jo=#EFG3@j1EX|Pg>pt*4a45>_j&NotMgR0Fr)Zjgz9qDn<6*<@XKjYZm}hPL z7ZPgo&`SIlf`nGmKXI=&r{MmtwlfTev#+~j7_{^CW(*Ek4SWfx!}Bb~{q(929fZLb zfEMu_32*dl2T8Lu26ue+5Tsvq09Ryj>h}SC|LVW|3(oS-MT+?nHFy5gmuR;0pT0zy zo&WSD?Cbm|>T^8|sn>e>ne%6a?7>iYPkyNXJ22xz6nxG95DZ`QVVu3?#MgoksqvW) zsUL^8ed&qxPajg|v+g^-?L+2#&4;&ni=gicJ|xoDf@iCr^64`lQa^6pm(P;^310av z&^{}FH193mzUD)Px3v3O05a(%`@YYAOTy25Nd35VU&hF*b@)qkepY_B?%9f$B>h?d zo%fQfU-O^u`4Y3=6?{nD&jLvOxP8AFvhQYmNa@eIANg;Y{WTw2LC{YT{)2+IME_dw z{N_*j{+SP{AGeN%lHUP*Flt}=;Ir<>r+sUOulXRkU%KLJKD7F+I=&WsD3Z^7Nd35R z43vCn9Gvq2is!TLuV3~<4}Dkgp_aZDyapa9sqgbaQGm+&%!kyETgO7l?^^d9E-xMT zS^4A3zBS(0{HHp4DZn2Ty!GJ^3f}7Rvj9>*Zr?MY_-p&4Nzr5A!*8=!f zFJ=2RAKvt>d%qTZXyVU&Nd35Z&(Zn&=7B_hsqW9p-$`#%N%{o;1}-m3}0pl`3{KmN~8lN34(XTh$--Wwi3zk=~cu+KdVo@FrN z;K0w+=>q4cC0XJP(Ti=y9;UvDY0jhQg3x{VN;ukork$#*0Pzs~qc}Et^SN?5s!^0`|ZKiW9hZC=IIFyfMpf>&s zN0Ts)o^{}8DT;i*E%vnH-)1^S@znE50>&G`v9Bv>mVi<7tiv$iGJm}z*xt{<@k$dI z%W?2@m@7CVL z5jG4Dr6BE~S7A8-Pls;`i>&8Y0fhs?2TcFxTM*Bed_EC65PIvG!k^!i|Gk<1e?rvM zU)-VpJnqo{Z<*;IbN^FfzjXqPm6s&=nh)m4OD=rPhk5#v65r>)WyWVdq<*Zheqf}( zWz1*YkG5eiiSsodjK-Jz`I`SBjed~-l1-oakos}kzBAE3WZ7rkkG6s22g&w*{#(v{ zpZ}J6-{-$&;AcLhe%!Y24D=6~`C0d)ZNy80e$D@or(g4747{Z4*ZdDz`c?&SfocBM0iSh0`t7X|zUF`E zhp+h`+T#0skjbDsKJy{< z_`m*FAHk{rX-+ji1HuNlpgs-^`MaMXGWfZz!0DYkDnm%%WIdlUw~Tj$3Ju#?8DPGfB37PpzxPA zF`-p;_wWAjJSFRwJOWzA{D;2>=Nw$v7Wp5brO*LG-3|30|LT|L=l^5%_&ZoL01Ll3 zJx|s8J9weLW}Y6p=YwE}<3B=sp0%DA{}M;-AAf)0Uqc{N|93e3+X}x9qQ7+bf3^<) zHqTy;!v75>*`MVo{9nJ#&OgIZ*h=3f>i_cVeg5Bf6utwB06=+9@F$c4yl|M|&%}F% zKU1(DCQR|aI12ybDE!`0_^l3Q~i~H~|?!#xO1&oC^s|BRR7m~o6z}b_d;$Lqdyl@ImHEg1+QXWk{>b>N zzy0awe+?gV{W?qY=b!a!^Edhc$6BtRhnd4h@UP8WfBxxj;HYdkI{NwZ@%#PPx4(b> z`6qbd$Nuu*2*Mzkq4}HUd9&xwhg0*#+551`|H+HMKlII7|0gd37t}XP{-3-Ed{S>4 zAAbK+7rp$h-QRw+@sF^ktsx+^ogvTy4u}2QpD7ML5A3kpXWdWGMcJ=gn8PhUU-)+` zjc<+rY~lOb=VIe^Q2etsFm;|k|1n8u7~ng5fFt*GFaziI z*$nW~y`9RhvbRV1o%?bpzjEKcirD31}KlO z$^=FePX{Yp%M&Dn5kJ+@1b;p|?1Oa)$zN%g&)i=*l+PL;1W(9)6A*s-^IYxa1ZOzF zW)lni`p5qsg@9v-lCIAs9FdzrF!VG020~}h{p)w8W$&P8G`%LM2B06r249%OVwuNcxhl3wSS5{qiwSDc%Hy!N zAE!vMuteE2+RXRAy;iIFIh@fT$annn-fMrg_53}g_%r-x1_$|SGU-ha;L&0w9$brR zj3rIUb27yZ#brDZrC#Od?l8D{5_?N>7h-3#i30z!ZPt-((K-_+rmn@OtGW(PcGd}j zBP5&Kx!&xPI8AEA)^>gCvUqPIIXQ7R<#O*YEgRm<4&9u{j9AikY)W#h!-{mZZ8txP z^;Y=x-}a!Uk-&cY8}cc!D`ra+50>!yw~gklDkg^&WIg|#HKDkir;rsg4TJDOqrF^@7Yd^NPZSrydgSA-~WqeSaQ+j8QI3{7jVTXH{@Dx}_ zePhtu3xDBU+aDjtHZ z=o{;issj<7)JYvr@vPOWb7avUeE`2jDHJcoetRl}*S|w`vh095t+(6vKUtF0P5s+` zb1&K9+x4gO%CppM`R#g94a8$@eD>@4Y}p6;$ls3d$H~9w{LS^dn-vT ze{IjRUT{2KjNsSyK)cQ>x4&+>2L@wmpogX*nf0Sn`%uoCi&_M!zVJSY@J-*3BPGRg zTW`GBtef6v<8&a-;XR6m*!0Pj*Uq2sem!n4hEEYeceqvs&6nrPqBdGt#z|I@Q<4D9 zD6K+IFU6_zaz5_|GA4YMzhl20H}XbRE-SUl;hb|+P)ZETIV};nZ{uX^*s8itx2}}> zT#~?LOl8w(xgk+XS;CR*isLRdDc0lBg`(24v*7p!4F@tI%Y|ifa-nJawzIF-Nxth8 z9s1R1`d(79Xufh~$n|H=Plha8B3-n}W?JN^_wwTudvj-+ir@u%w*x?6b5&A_$A3)q)~ip6}RL zDkoIAUP-hd>TV1XQ|9yK=1}~(=SP~4;GKGgAIJtlyp%>!+$;jdpu564HY`2U$oy<` z+m3_Pw=u!jNN)F9wXn^FrPPmcbZ7-_n4PAhaOJ*ZN6sACw!a1VkX%B(o?R6BojwYN ziW|Jr@=?eyv+em;H){9RT}O>RcTltMiSd!?4@}n&mSW+$bD@S2C3}`bFTWj=8&>NC z*S@+t$*nd`cezd4ZEzm>X(cBG6JC0u3vby><=KX~18 znO%eC_NrEI%m&@Cr1h(;Iaq5ZVQu$_=gj7`z8w&LeWkixvgH{KneArVY%UE(#LMpdt@1WK_{~5M-H(a z6pPwT+@t>GS6Q_e%|;6&ZS9U*W$Xm4?0WBvAqQn}GeqqW;gBNMc4skWO^cwMN6Bf6Gdd-O*kt97mhE3@dRmtaGua^mvyu zJMB>8k47P4y^$CMLCW3Cs<^CT7dRi5$Z-9D35HgR#JP_T{%N+8Jmigfb5Pp*{FYm% z^CltT37Y6@7k%mlqtNBBoTY1?)VJlp)L1}hJ=x=$k)<#`$ntcG;oB?sK|$M%f-*{J z^=M$2vVxXNgQCLqxK|LJ9P^VYup~>$Jk6(+KNFH%PQ!{A*oI~Mp5pd0i=<e45!6a~OW3E6Oz} zdQBW%#q2hjciIVdy*~}vN=~=2D42Q2Zci6$dh1Tmaq?EGy~r-nYRlB)VSRO_yUj7Y zqR`b`*NJS>ztIExt$<4luV=t9UvA^wQ7S*^&cndPq>z-q)j5yT#yPe_R`u;zrM)`1 zi7h(`b=r`#7XJR8SQU|F6y00eiPvGX6vt~IB|?UT^fo#DIyMFCd@{~g;*xF<`zgZ; z%t-7&eMAon#W%mze|upQc-Z~ABQcYc8AtYv%&wPRxnB|Y8-EpeVs%Q&lFK&?gYsuK zy^Y1T*usREu6WqW8y?vFHUN>CSIF9JdUoX>KVo7P=5`J2oV=;W#M2A!5+&U>j z!G7Fdqyu#NakX0{%IZ%ilWz>0SCx zmKW2^4~GX8=tqJ!oAMFb;r6KIN(uS{X=LFcCva!K+r$*e8o?rUeKpC^{Nr8}j( zQU*G;fI()+UPS>kr#U&NYf?H^=s{FPY92G;C6Q6A3Ek&Pm9G-MJgN zJ6l(Nbsv31)oz(4;XHQ-eJOD7$PgnNfjuA6a=i zg(%Mo(l87c0*os{wLnd;YXr^1wkmkk@tb{2qU@U4l>u6sInSPIFU>-8p$dP&Fu!~0 zU8P~ELlW)fq}Q~WEGH}TkFIJj8?ZgBlaL@$vQRvVZUo^L9joW~D^YX5pOHBkj!LW# zgyq-f?bzivb~iV&uaUzR^jcw8x4@In*9aM_@te$X&he!mj9Y%j$vrI#+CdhPaAo7G zb|WZ5?W2?$z$#al6fdr3^>CC{F<4mFUbK!rjY#am*BRGnmuTB=cJcGv}TZOL$z)TkB;+n>7%~pYdo4W6ydQPlka?%-En-T z3p~1+nB2O#8HY6Q_#R%*v>0Jt3C2yJv@MIQd=zCQn7#{%T~r$sxnJZdcd0mDy5nJ~ ztBtyF^JOr*@-m0dbsySb81Hl4sT}9yeLj`K_2Ow+9*4K6ba#WxpK}@nGmr8#MKV{;nG*ROClz-I$=9ikeJ3O`hiryWtJ#V7yWF~D%@MDZ+J#LA4dyl9AOc`shXctkp z?(CG{TU*{`2pC?}W;d7GJ}k`}W5*R@>+(jFZPjS+_u_P&*Fmj<{ySZ_NqD#=zVz5% zhC}hJ4|;wkWr{pdiFZ*~4}D$h+v0M&z)`LTHX*JtDovL_9!?|ZQ@wCTp%Fa<)@d%g zh_ziK+FB=NtU4Gc{CJ#FT}v&|4h8JAN37v~`6pAgP2ZOm{!VY7a8RWIt%-TF1?X18LS(JLxy_Qt@DoviAiO!41Jr&1}wX0 zSxe*V-P~j(za&`RPzVgJOXh`X8!Iu1wEJC%5#@2(hcecy#RD-cp~uqTARP8~PF<)I z`A9kW&@F0rwJUl~uKG5xR~08LQ6q-kV(!^AYlidaB2FXkef&PUQKQaDaT>IN z#Ure+%k?3JnUs)H=|uY|vc~Mzu*4bWk!o*?qKn0}M@P}8W~aB%wr6?%EG9V`8a5+r z#A!~?x)`4KqHQ-gDp(W3m%Mp%Z@XaX0@;UkOJ>msBIOtQK&~lNI>6XlS(l&|L|9Y! z3(tpXlII>y8%LdLNs_!rcps8m+d$o-(3QCdds((BI$IGoclQT!$iuW@Qv@e$#;(Kr z#*Kk6G^?({_ZJsc0?L~PAb14aXiUzFCsnv!n(A=HA$1=ik*3G+uslcjC;mlHoMF;1*R~-#E-#40Ja|5BQj3 z2i4wPCK{gvtFg?3dGc)g7FXm6ad=&R1c^@Zebv+s&gv&2&&Sgltf6_mL*N5n9Tj6c z-0OojB6&Q<5C%kmQzN~HCVfMa#5(qH-8${X_?&2W&~^136}xj3OcuMbPq}$Py|^bfsbcNjiRa^E=lSYV zPMR%Sb?B%)p@F(a?T{HmMp{qVlo(0@TS98mkyWcjY}Oif4le9Nouu*9B2vQ5xVMQ9 zu31Y-+1=>*UWg5bb~6m(JvNR0Eg9#4ZdsC$Z#>7HKv!Bfl{>|bndixMKJcR@8&B+r z-$q~Xsx_Hx3A!Jzsjgc5U7w6un*-T2FM=6ebM#2eS>wavnus#*o zA~gyJCo>+0#|G1!LQ}u3p|^*V+BBWFz2y@^Tvyl%kQ`ZTU2HqJMzyefCO+BMH#qLs`iVbxxCicETIqk|2`_txFUkZCo z(88VK%y^%!PvUt-jX8>@=m1Aq$|v2tg$A0D*5SBWi;Qu;3&JBOL@V(f4A=D|P|s}` z+8D=C+1NcN9hja(#R3KfLW<9?2%jlwist}~^dsdTm<;x0sl1##y|9Pq(DlL2ET(A6 zS^2=_Y09jpVwIEQ9t?mU9#vqTa2VE!5?;J0MrJV&yc4y)N*Z8Ad2(8_YsFY+=fe{_ zbp-70n}WMa{lt6Pc}7X1T&HS#)7%Jtz7Rc&Hc#x!MBG;1;#Mg8)m2w8G)OTj(IiU@ zqAaT`8ig|>kAN!+$`-2>WtzRbrE?UAYMc;`p$KA!z#KWrI% zeVZePNufwMCm{c)j?3gK7wp?|pHFr3v4H z=x;&=+AUfjQDNBHy6J68hPNxwLbw3HCb#(R5M9Wsrn!Mq zoivV4-4qE^Vxrc)^AKq(oSWOutu>GIoRA=0^*l?{*1Amj!Mq-$OXg5Sq)ImIi2;T% zy`~SSi`z#Ch+|`rxW^-11m_(Ji$Qd|1Vq^xij9^mn2q}&6N&KH#tDSaEmoSh>n*Ma z6Gl8Ri(R znRt!E$n-D^8H!=;khvUQ98qrB@v16eK-HUGA(Qhm;IwopJf!7s+w(pa+=J&A7r^%?)1uUYW?;fn#8OSEZ^r!fr8U!P2J-fI<9j zIC^z(ab(vfwML8#(;Udw5g+`D%9T0aZrF$R;4*RCgEU(9L`1lJDm@NoM=zocuA;PP zR`AH3Bk!Es6BA;@o>h0r873X2$~X{UL@wP))GhPgK<=HN=a7i-XSAhm8o98tXS8$uG#LQ zqDq0J7Rsq?Zh`r!wfBp{gPA}aGtXg>@raVmo^Fb=dFj<@KHs##+gvXkC^5Fj9IA-y zvrkL>8nDXH1*vzkZh;9@J>jmn!aMK1hNkKUTI=`NhdmR@lKpAIhmHhT(RX&rJ=kivunsCQnHDEElxcCjiB zTxWLprVKm;fpIB{Ti>JMGekvrJP$D~!i+w6w{~ieL8G4)e68kby{Qg@7ZLp_qt=u; zSdDe`J)u=^Y z--hWOlhcXg5PY}JYA{HYB$cx-BmX4zhvuGA*K{MtFwAomu<9G1tr@_WH^SqEqp#U4NXUEAuB&{a(~2Zk zOUjT=satYVD@H7bO#=*oD+(SSeulMrN{&f8AR{KonrESk!?mX)_p}J|yK1^5048w{Q6znEF=fLq8^Bm$La=+Jxy& zH*S5h7|mXmjCemW zcMzT`4k5ec!9oW8({L#*r@Ap*9fkYGvuYuSIUJnE5#>>O@%ywL{OowpY9pxXjGLwWgbvp_bVPT+xk-HgEA=;R>eHX9F{Aa)M^vzJ?W3 zV^8AVVGGwI+3qGAVo@3UUKA)o8kMB2Qr6KhvCy|s9q#cHUWYWH9j!xjO&Rfg*@(@& zZ1vDWEwU9hg6c*V^r2%F#eO8Ck`zw(;DFt53f3q%;$gkj`MT$c z(^zoqdB&7wrkd><*K{W$BaG|#egPoP87k|ZPZ+wvMZ~{EzP*-$_94yS#*XbJhG~!y z2VEG~vJqZj$zKsoqAJn)IbORVY1PYScV(GvB}^*U%}(g%Sly7`N%dhppkww|$nTr*f6EVTj$HcIB78-VRpS z?3m7Ii7>^3AXCW@)>Bmi1I`k0QwD#qw4B37b)s67FNw{LH1yM2o&>ZYTbX<=Fob?y z&~vimtZ=Jz&8B|jSpdAR@b30j7apY(fm|=S$}P9p3=zMrrOgrs8TqJrugmwP-0WwX zmKs?)Il+;uj~mh7!APJmx0&yx$bv=BUoo;kMGSJO-I26j)M-jnxnDw{f%DfZ3{ zNa*`?x}13De&oG#E&-orsa0{3LX6Sn@#bow3#M_u@Y=aQiH}`<2~vtcZY~OT)n*-t5e3Xz>^xv??BrMZRU| z^+9{+{@kvQtzZB+vkpylVM|n@`Gu0!>gIx|CsqcGicC^CyD}YfOTs%(W46?w2_^!je0g5dK53U;;&3Hx|Vwv9JL-?7T=qSnXnEs z!t07rEsr$fKc^_Eb_x?>q2bSz2} zt;d&d$jd{dVhnq_8A0kbL4M152RJw;>~KD!9~B6!m?E zLx_RNusH3!sJ`E8hX+Nf5i^b=6#!;ywjqdNfSr&S7D0XfoKhAT=#C@IH}g;t08%Gi@Y|7f#5PUMRCP*yGbQF8HhXH+3?k)vk|>z6XN9hAdwM} zMaWWE3%Bh}VJDt8?8=$F7$5Gep|{+CQv9%EY0o;0LT%8xy$o{EY^nyenDgVzO+X{L z!dXFAlPgk`Vr66{UTBxmch^Y?bE5CD%7{ezw5fs1aGeS!)tY)EmD@7UGI=m9!93M< z=YdZ%99v*G;1Ln?Xhgv(6oHdfxK7xlsLy5(z){p?Ph~-wOB-%?boDbiU>l4}9%pZ( z6=JIvI=IJ;{20Y&asw3N3T&C_5!MbMgH(C6l+_KrdMozU9!iqA65TQj2zQ@7+=y2( zS!#`|GKRa)Z`byG#?~-V7DCk#=Z@UnLRmaqxfDkNuczv{_EGOPn2x(-h%DtsvpDj= zFWM#9RDlS}WK(+qI^y9aw9RKDyGb+7nC5E8hVXtxRB3U%{vPw5e;b zsWo_+?cpK`*)0sjX)6cg2D}6~>Tw%RZf-7SUAJ@rv_3}2=9sF%cE}ZtvT;}#+&!#x zWS!O`3ujE9QAozGL}p9F)3_3JRKur^;oh#(3#*7)uZTL5AES(mnR@%00E`FGe!C2B_*AoUz4Y=-87pEg)Ct= z8NGWQ)oMR&HYl;~H4D>g5FV^)YIpC-tT!!QV=GGJ(d-6qz$aJe+xxC-X%1w|K~~Os z!{T?8FDNd?EA^N@yxomX!}hNA#Zt=Vrt5EBi%q7F*{~i= z(+AS8j5U9~BCsAjQ97aTPD)oJ+gqHpI;_X~uGQ;%zbAnZl1Zm2$0ANtuQOxYs8)l& zrSIrX-0W)SMi~Ab1$G*OSY!i4 z4|W0j_`bXOF>p4+`DRnP*LxfS9JvWPvm0SABM^pZp4v*!8o(k0{@o25G9yDcJi_CJ zB@xt-)dG~LkKXp+-(Qrq#WgTVAGiLND*iNu0-3wYY!3K{--kr8pX9~zA& zG6^=vpgs(=q_|ZgjEZtH4^N0oXFbo=))JL!=h0+IMn6%wB<#-Mbp&1#UBr5-k9klY ztY&py&Ql$RtTuF4J(qk!vvGq3>VxmQECD8GHA(7$y^@TC@H>Cll6B&32cKfq zSV%WlY7u>MyOQOpIW?yWfSJRQ1*yZtz3q$80t-#lxjjZ_&sK#!A1(fsm#+1~O%PG;a%Z7-DtWu5xF~&acNdt4Hxz&&QDkMtS7s+~$Xrs0O&lAh zCkos4X!+`iKl4>cqc^U&FxO!@16?9Ao7Al$TDIUbXZj)mLjEAi!x_V=!-C7Jkv1-= zhQqSq)We3uG94xk-l%)dC^r1wE3swYAtikYG5nnO!DXCT+Pb$RfCQ*Oj@Qjb?1E+Q z(cDxdmlXnjqM!wTiIbwI9&m|l4Lu(_j=FNP0uNlDPSOK_SgJE9a&l|$26$Q93$@ef zm^6~+<~QYuQEdDH!$1IlY_5V#n!4B`rcT=?ru5Kxz?$~pEWvknqu$IVjJDu|i%DME zavpRNWd`Vl3iA8nPYkjiqQ(Y14v+fKjq7yH?&{-kT+XGSoPx1Bwvrp{hbf~8P9a?j zaDD3Pr!#vV7X#Zn))YjvzXqpct0ridZJT;I%T6R=$EQ7^+h^Gx50MfZ%QE;$38?2? zXZI48A>c1wa&ttci`m7?0u4KJTF0alzZOp?k*)H>)eEDmfo>7LipMPMcOoXuj{|w# zYynrgRS#D`RYqpiq2OLrIZrCh01M;pRUHc)A_N8?LEaElT_33K(l|iDI^WY2t@%6# zgYk)hB~J`n!hx!&nf0DHF)end*Hda^g*(Fl6IC&nR}8V<9(UEz-{jct{K!^#e_xZ5 z7T?&EKU`5|OiB(^7%i2s@dS<|~5Y%n@fuw zvwEeSn?sX6=jdJr(MwOVOO7kw%n@!BVX3ihcQe;-og=on?!2lFBTpK&W%W=p(Tap8 z2dEU|5kDd=?ze@s%q#XqX|m*%n>PPii?`J_5W4i;NlW-nW208YYTb<|bvyaW6LxaZC)^65QZy1ModLch}` zuLlW&3}`0XR}7q0vK&eBWt$d2mPtCED!-agzEzENNlEhxjpoQ z6nfH4wk|+n?hl5bxgCL}&~wH}7F*uDYQAfL!u$-BQVGhW<1sbz(c z4N!bGSF&aMfN0g^K2HOJvepR|V)qkH!kb%n@6v-0ARwCanzsq>p(T4nx~NhEoCMW< z$$r~ky=;(Eu8@|0TpwvtZasPDT^z*VZYGiumt((W^O!eNHCoy|>m!O;qacYr&*#p~ z#ygHZ&sxz?hP?yM@uE8id;f^ZyXxYFA-IJ%Lg(IMfybnWl9UF}0oftG-Du?0DMC!7 z7g`102?zXXGtgxW2b0h|*;pN9*--f4Z@~fLVW0hx83bNo05XGHVuHC&YDHGxZg;I+ z9^yF|{JjHwktALN1M%nzkF(J@A^BWbC%1*byVzL!Bl-ZxJz;=D?uw43GJ>8kQO0<3 zWVf3ubFPQMOP0emF=-q)eom!W!1dB)j(c%n#q-u)#`Cf!u_it3eH0vq0T}OYhwd&N z8Xl`LxGosKyB*tAMXu+abrz7@4yYmX01G0`pR>n|SHk`fQ_Yz|73^XKF#ry2Qj+2@ zCwQCT*#nQ&O%x7(Zqfec*5>W(6$ss3Y-E10wd68xdh9Eo9M&8>%hM7hot<$PT+MkO zPW=-5ejKe!)YIzpq8xAgC{V{?E-;}7t}M4RD8LEne2scSw_0&8tf@IR``&3|Px2Wl z15t|XEc19{!D*kZ@{7L*mPoS`IO0&@N(8*Vw&6+crrwInTm&$Nup!wxv0!EcpKMsU zk2@xlSwEl8cjGj{*7H(oActL8yT*pxm`|Jz&_I{siwdh(XFL=fE)Fp z-5c!@PWqw%zwPP&t{AiJ2Ggd`JKwnP^Q@?WM0=j|^_*t4fUDQzNT zb5nr4g3c7v2W)K3^tn{3%SjnMld!_D2Bx1I0Q^Uut76Zyr4FtQ3Ep7_UjbznwbMAV zY2ejv-)9G%QD;M<_gs~O`;2q0=@zWGeXcx4_oN92AnX7>g&u356WcZExiVUGspif0 zau@tNUdCfcH!9FeZc-sGGaAJQk_(=gT2bYVUUWbfZQz^U18ke@c|)6KGdZTw+*I|% z4cO3Mg+iAk;7kVuSE!qumza3lB~Smc1o2DrrS!z4vTW;dNT}onywHsSPOQaHjbqBL z!U!k>LM1&`R*~sjAj+OZ;12*0V3d?!KHSqbF2M3sIMz)jwd-t;xNH$^grV0_PBgAO zG`w5p6~-;c0eM*<9zrMDSds#njkWr^9}58~0X8Df3MMmyOf1rHBafzr0L$+_F5#w> z>ixtA1gc1z=D}Yu8kfrbz1LhtK6OklW-(Qkr(p@7vc8mh*g^b3{Z_I*4h4iHYTuXb}id9!9L&m*?`Z*c0r5*i$N@l z_+ZoE_uRGE4vjqpDJ`Jyn^eAZAd9y3wq2E6MZLx}@NM{Eqk%!09i3_eHkEp!BcLT~ zBMt#sI<)vk^T=F|tII(snCR7cv&&`>ao4N#!!KfW?&vD2+6x-4+JZ#ht~IV z2(qr3d#ntm(0Y)Yio4)<3r`rkYG8cb`O59tRLR)GP&7?QlTtR7;OO0Kq^?aS&_ejp zA0J^l87x!a>GT9>Fz?)F661A2+vt-Dnmt^mqki(=YtNEmy?oGlJ6W9U7 zxLs)=I2OHh`)vsng*mv2G=4C-fCed)k>E>)voP373qS@(kFALmS94)Er?`3|dM+?B zA2p4JFju@nU%EP&4Weg8hP}aVwtL&7;y!>#Y|k5chiV=dXGcjRSWP)`mmZHbFdUvs zW^l8arYVEJaP->>2jl<)mzW~U)FTJ|g)LL6mgVYDgeM2xkuI5BeaS_ZTW#}uaEL!v zb2x3m?Ydl=EIv@#p<98KY;p#YwqPfX9)Ofxho5H79In27_2+#=8U%+dh6~G%puW3x z=Wg2Cl7Y}8@bd$6!B*j<<`%VKn2rI3U#vUeB&h*J4bL$dU;^Zfry5)d#Gy61Au_5j zBGQ^dai%vBhV%kVM(V(CVwK6W2{$Npb6@Yt_MEgMjYRMTS3B9qAsB+X6Hj}t6pS^5 z!NJm>1lfvVMltTj!+-!@-|kxb3Hkf*KKqixNQuXwJ?shQsVby1)_I%uo++(T;1IHP z5Wtt)4zhAAx3nRz;&XGgL0G~X-Qr0Bj1dhnmn}|!eR#BO)td%`>38E3pYpS$T7Z(_ zUX~)26ohfjZm`enxQXrSuEZ~qW*$sZLluz;!rl1OFv$j{zQgAZNA>c+yK)6-W)x++;vp(MkLSJr9uO=~PPx{^SIsJa-iyh&EF7j_Lf9^GsbKtOrjcv?H4kP9WhJt-zC z59F2!fQ(wC4eRoZJD^+&y|E0Ud1vt5s^yjdo^qy4jY%d{R5=r>Lwu&oYVFu@XSR3Y z#Gai6yhfoxc^5Gg-F5M0bp|;j)bl#aQ9%1mmO2*UzTTe|g~WwaIbb8&=o()?0(UTR zvIJs49`-c`Oc~(HNDX;>oS+x8WV@Q%TTib2CF$LW-Z|Ki&m!NDBHi9`aJ8*Ux|a{Q zGDW>`S;M-hmwgU7_ByNIhO>F~cl!(Bg}CSET*9oX;Xv9 z1;?u+vZKK=pob$-;JXvq00q3J+8O*FLCu>L?o8IQvAh-nJCOy#{i?7%o`&u*ozQ8{ zk^^!{*N4g;l_4V;?+%z{5IlKE=ZR+SB5DFjBD4?pto0w9(pAlRhN~TRfDl`rLD3Z( z#uS=CVz)_RRtgg5o~rAtp_))zR30@T4mZx4sWzB#$BhcC)edz76fR1AxElC9?2It) z+{M@f^|LX{LWal;_fXKCwqLpR!uXpA(X>t`UVIK-t!`Mk@yw0lK#Y_Cct$->U|P7N z+E019BWUNiwjW`C20zhxz|aGY@6%h*7B$rf!wdnhE)l1k%U)K1cdj*mr21j2;FUWB zm-fB_Oa^R{+y35|Ey-bYb?r>@P#NON=Fk51`WVHL7y?clZ0;F%fx`keX<&@eqzj>M z5bMRx^v;k=f2R;!F5d6$9W6^BGn!96As2w|5|CIY_{)5=Zm1cNBatiRWe4lKdoGyE zkh1)$_E||~@eSZiXw_l}#4$lLtLM~l7OUP=*`HJEK9S2M|;b3P9a-)Lph(`Ur z8e;5hm}_eC)bWw$U0ec=smyBD_T+{t_RI!QF9mmwGMDT^@YezMBmwsH$mofa$Z7+j z&XYK>6{(Db-b+TlauVg90F2RW{Mch%@L)azkI|W1bLw4gut%|&BDhT-Lc#9q#C&4R zfc0Op8<(jT8a*I})RBohv0pcp^6cm1umj>=8}jXaP1JW(+R;t8qW7+>USVSk5rPDm zG0po8_-eMm=Nj+8t6t6o6`t;CRq8VA21c_+*-w6cn=}ue=KC%}^mM zno@4TkY}(b|8jl=8-yeQ(5;G@uAlK~6bs*!A{X_FuNY%WF`WbIDAgpe+5e6KtAmO| z`-s5f9vj>J z5gitaJFWh2n$Ba{QLJ0S55z!tGdyx2ybvIX8CC+}jWADtQ}y>9RCIK7br~7<+AA}^ zJROj;E3wUIM7)%cR4<wMi-3Z7%i`nJ9E2~1m-%5?BbW$-;AJ0-h^_v!Oj?D2Ul!l-z5THHJlz3`?N z2TP3S=v$VO=6zOZXN0|B+o#{R^la5W_Xzwh{oviS7|#f7UWC^pd(v=1zP$HW&F-=O^DGk2O;l{?X|%^S3{| z^K`E?w1{_^+h5KN)K1Kc(7h5z$b4-n-0Z_!<4gzgji&bDIzFBXE3Uz@I_soQpd{Or%&Vwez%POc=w5olB;R! z-y+DKj56qO=2@ip=cRHj+8tmbe{W&R+r~nWA?c;#&y#?KPJslSj9*Jm&J*5?gbb2E zF*afPvoI^({b{MYo&M!*lX~*ThJ=OQ2T`&qvZ<;%zy^*IOs#F~QvXw%GxD5pV-)kd z$c>ft$a!~!3Tw_6&a6l7d)eyFW4F&w+qaR^b3dbQ6EZn$R_VW`9n5m}JG-48^&cl! zd3(p*eV&B|QNr!kk@Zrn!CQCQn;jfWYqvY8hhU$h z*^aG*klEIS-{-M$s1Ek~bacAt5u54rUcOut+(E~v&WgaaN?hDvJ2dviN=Y7@z`U3E>wI7To^ zcvViqRqUpXYFr&T4oW4Ip#C7F{7!h;>F0j(XfIoFF_cF$#Kxv3IG zQZ~=8?+({w`+YVSI^4W6;uI<3YF;klH94IeT;whXtXSTkBKxQ!UCkuj3W5#;GS?cb z^v98>M9_>#SBl9+*W#ePr>nK;9z02ZDL<0tPMmPftD=2)pY3EH@0rV13KAhye3)j9zMo~O zMmp|Vsb4(xVXAjVj2r2>Iozv1J~8Xq_>>2x=(-27{=BG8gYb6-R1NMe^$P)f+jX!P zBFDem@$y6j*$Q@q9YxWum}e$|cK*ePVcXctDB18SiVd(5c>jN%Rs;OkAa#$&gochf*MRq9&964_KvOk}FPXEP&6Y&quDyoMxWL^-2DwZOmj7>&` ztM+%=yL~Fitb+X;SGx>FaCiE*6OJjxQjYRj^Hvhi4wx>h{Zt!RL_eZgnnv@r@|w)X z-{$SK6S6#5YN*3jRcBUm^DFmt+~y^mu4y|brt8AakdECwTaMA=ev`M=)NorOi6^sP zsHV-`AgC+8xML>aT%}9xb-2ajoSS-&(Xx46KVugQ*@A1L`HW2NpkLs)ugjUafbZgQ z1di^Qo5RT}2MMY$BD~P9>D>f*O;p}*X3s!D8RqE4g!d9s zxg-bfyPe^3%>P=Tp<;K=Q^t;?ygXj^9vovfOrW-=GD~TQ?jlrsIJ2a3z)$ilp_}cw z|4dvet|PC^tT!#on3%uF+?e+TXG$GA z1e&OBa(+U-==Z$1Nwgzz?kFGB;*GO+nxBF@;zoYpqKS6D^k84#LoE_=jp2tRpfDEF+H{B z4S`D{<P!>1Y_NSdAl+ z({cfo;Y_#s2M5Ud3ZhWpP9?`E45oP3{a!_gLdBix9qhM}bDc*oKGOTR7LU-G!TdYvliK zFliVEpZY6e>$?j-@MT+=wy{C{X!DWBNa;X=;N(_9D6LRm&6X>ie<{y@Cg4gS+Mwh(uGIeqBWK zd?P3xHb!20X}b9lt4nI-N_{4UqXI9CvSg#8tc~{;WxUhkbXl&8s@eMmCKReZ z@}burcysgy770e&v8M;)VlXn?;QK_r=eleaHWUAL?d7uWU?%;RuazZ)3@%_u-~I#T z+JE=gibTz#Wws~wZ-u29fJ*Ls!-;M;Qq4|(hNkg36_+op`hPWYQk<-CxCjNFPMx8Z z2F8=>_Y^FRr48-(?*;aWX=0A_dh;GWN7D+hfYxh8MSq7SM&n=!Z|OmONkX9ePvSNy zQh~=ZuIhB3tHw)JFR_A;)9z(O$eXkDi(}LxDeY8n`L}g(1M2rWJ!_gBk{ntGUP}Ot z$Vcj0MCNqv?N)u|>5?1uc8`b?t6MwP*~4G;R;y=-r?Z&p)AT5%0Y4_r3eQPqJjG?uL7z3ONLHmSvn4pSTpjsugQrcf~0? zMYzW5#KIRDZ9dZB(=Rhw@8AJr>-H<34<#^0T2FkasNLiB(JRu7`C)eLFXt^lpEw)n z6A|JzaE;u`af*&$4n=e7(avqhz)bQ4M-aN@a@iq9)R^@83%udC4wl^2=MKwcK^=^O1p!UadX%mZF+E-Sd`+upI2)6el@f zT*!-8jSE7E!2e*``5rpR@;MBg&RBe0epgzq+FiiiS28Tw+yePXvNOAl3;R^KOKwBM zW1dTD$~;W1R*dT))6yI(71|fm1TSMbbiUIr?nKN`3d%(&Fw5bSnM%)4Gq$s8h?m;GwvWcHXA zEzHMfTPHm~WCtQ2Ptk-}fWFVpxpN%j6*2E!AlsM=2JG>SP4A2lnV%DuT93;?wk!{$(HRVo{d&=MJJ~+#c?H@8M|kBw;kq6)fxZ6A9eU z4Z8(g74M9(;^_CJ$74MAca5gyPF|3wNb^(R7uS+kRuM+f0=~P=0Tjx7rNw02L?qUNjjf~Xv+U&moqb)o=9tZ-fgkM%p*t5!? z7G50dcY4RIQ=6IHHP=OURo7?QRLVmM4AAaWXw$Q6ma|-EcjTh$Gd+Sud#r`a(stA5 zrn=5SV~TP)JX+8g>g&z@%pVI;lI*)3;b=#6xUpe*+s-YVVnZTL&HKZax_#MHVmcFb zy8Cn-J{?J^0z!;zHa#|yB0t3?EPyeEqjfse@Ohe~=zoX9yM15&T=k47=62Y{Zqtan zE{Xl6|E8AvIPfH&)K8tGjW}=`(d_c>HLUy=e!2BDHs&XUw>OT%lNaCW&GY-*+EJ)F zoS1T>t$aK8gt?@jSKLk^1R_Ttgm(wRa0eEd~GsHjS!x5qW(z@n;HO zXuBHw6&n9fp)#GKNTKZ@SrfBx%0yOX!k3OumF{cB{lWPddfaN;_e$PVr)t{ z?DoY9pJVGuk#O|coD7tX*4Zk?sGbGkvG1uRQl^CEqSqDV+VHYF$>f;^3(^IDO_iU} ze}Ux)tP3tZ-pUEKN5RK3CUzEg2|0a&awj+Hjqo!KkJUR^lXvaM;%l!fb?YFOhJK6%dWlCL+ad^qJM$jQXv7}@15+z4r(aop=}ZZ<#y5SDZXQ@GpB~) zr3K;>i@VgZCzb%y+`S*vU;~%J;$xH2tPx*&wcjJ5 z$a6HExT@$Z!q>4%Yya7$7e|7BpZ;T$#qRHw^s-ClD&ZoSCu7H{XV4>}{RuBOvRQR~B&`k5qA){R3+Ws3Kzg>#w zIFbWyxpjJO=7;83v;s>!p=~{|2hU&zd5+mR4a|8)T4r(%^pyUEU=%+3)XJ_7yJr+O z{q3p_n7J-s3K5pUCl!8WjH2+Y`^uChpZj7w94WFh)+2NXjm2TpN0~|A_FWno1$$$7 zCzVDzI7#YQ`dOsTO0DqHt=p^rE>?2@nN9+GWPwMxiNF&IBl`+1p1cY&FAHn z2my3Z?vDJICWfhE>`wEXu;OPN=1f@vA*emC(;rYp?%jXSmGc7G&-$13BJ2}3IG2G; zcm!&37i9hfx$fhZ-D&DUpzMR1?_d5I0>77ucLWtD@?Hzg;1Fj$zntS8F5xh9o_9+> z(cVKRxU?;o{Z|$m3OVD->XH38Dxewm?o)T`)LF^QrTonihqu-?q}fuUdxFQf^kdhw z4q}rpCR=bG0JdEE5;I@QCXiJo%%mJ(nHQ3KY8R5gGc}d<^Dz<39c_?ewu%==NCmpS zo$>j_JxO#>=C-&0i|L;EP2(r-@W9q#jR;!6FEWQ4PXw^f}g8x<*pfVy@nGX*2?3kp8sGB z1b2@)^V0kM7cv)44w0*#33K@EiSqT%v0CK%<4ZXEV#@x!gEcTBBTGnodD?2dzW4XL zHfLtQL={kh+xN<0>x@e+BvXZIabGos>#CSY{CX?KA~{rHl~1PlWxJCvtd58SLAfVb zxq%)--9Frl0!C9fUJJsL%!y4gU-SA=xY4Dx{=3(k59#`~`D1Qj3m&ohc3!rU^Y-=Y zYw3=3sHKVO`ozqeR7gg;h^Ee0Qt&U>;PrSo*_C~{R`KotBNYxz+27!Q#ZJCh@o z+@0OM$-(g=m-J}sPpb1s8autu|FnGQ#@}mQX~!ip-bHOlDSXdeRwxfqX0*9A%Z1$C&RJ1PMU;KDS^}H^V2?u;8H9eb&Y9Z;F7`2B7ejES@1FYuqu~8& z2GA3(jETs=*jzqaw&zn=ytM505M#C!v-5k!r}1nNZdK`2j5y5hcs`!QknL^e6#f6s zF5XSx`{m5DoYGHVET`p#U6T!0S8@*j(i$h_zTAFily85^K0&0k9p2VEak0>7Mk64y z`|iD1@K^12F_K_qlyF$f)lTTnA?TC%PjxH*citq|krpCC zUX#*ie5)kY^D>{V)#Gt#Kr14U>i}bQ{nezH9sPU_6X(-~%cppGIC*c%^?1HK1Mntd z>_tOkSfL)R#C;EzZJJ!{@PV)@zIcT8XLtD$_s-*aDAWE67;3J%H~gm;s87_6C6Bo%|KDXPDi&>|A2KEOkdjwSBRq>ve6^(f$pqA4uR5>O1tpIBYiu=}_|Rhb7J~xe z4I`_tOqG^Yv{g+Bn2_ktx9L{kUEEoCmSsZr+H%Xv{CdV-jnj1}GnfZu?O+r86exRr zKtQ@mn7Wn=6EDvWe_*S?T?Cvu$#0Q{$XoE1&Jn)#baHzx?WbzCFRYjvt}l}v)t!RK zGbzr9RmLLx#?m2S6<;{f;#9C-Fhzj8`Ww!%0$$J0Z+a1+x5O3j(S(Pr|J+p)dtGR zt*Xk*_tX1O?xY@y4k9PggB^f^Lo&z(ADb~{JspA ze%VHO`E`ejyX*5mZUCGgEkJb6t2aKM_ne);(VsnKwd94klMJc^v7QXSV#J(Y| z%F``9OHzhOg-cZ3LL;!c<(j4~)=J2JNUlW*CvrODU~xJn zO6_TqJ%Us@W)HreBwfvBIkmCurQ1V>;z8`{+d*eMQV0>2*wfdqd%B+7;ps^%)tMe& z3j~~6(!zn-^wpQz!p7{8m^YZ%;0y2GhfV2THL>gEzw-C`%EW88Yfcp$N-1ZA>t9?J zJR^gOhW3x}v&}TuUAukZwrXtd!EXk;-1fQ%|NR5K;30{CyU)kWr@0UDgdr+HP+jjF z5x>sPj)>Qj=dQ;FQch*7#JApk#(C-u|4g5ybbipy2Rru4s$9b)B~K-9VdQ=r9M~Vp zl@cu3Koba$$%$JIS$sXoHoK4l4I%P~Owphx>CG2{;Yprf1P7{sTVMR*iCOC0s?F zsCO#w-L$dx6x9n`s#Vhr#g<1K+>hUGakv`ejGK%XnhA1^B)(jtZ2DCIfqg+s z(hh2;NcYv;=z67&5~w=D>2F@nQ17TR(v;nxZ=e1fp5oVuTZ`aaRZ%kD`H_=k4pd=D z3`;-I;Z}S=LEae%ptCFtC+BVaX@>&tilC5Y-X2?Th{RGixQwf#iDt^HhGST;? zgDSTz;U4;U>8b3>D9IBjzNMZb{d)sY0E;4+Sva^fv@b(hciTliFjqr71MPZ5eMF7) z#2UgXEk@ug60cNUXm~sEFFrz*L2s)1+daQcMQ#mg56mmq1^G*F+oBJT++UW~AW^8- z(S2&M#OEJve4hFB-(C?g5L+uvfvX}ju$0&Sfz7~nJl8tG-l?gj^Gt0*83CdvKy)WGb-B8bjn8?H}XJUuN2Z>2V_6q@ZwTo;%DQoAby> zN*Cbmyi>T7fJ4u{&_O^d4T&x7^=y}Cu^)l6JN>i}K8ru5iM!)p07)8pXJEriYHBop zfaQ+;MT*r=MsA4<_|eT(mRXk;=hfI2Yu$4)3Uz-1)2R9S4jOJ)EK@z%`qcKkR4Y@@Msy>r`dc5ojBn@ZxLFs{8*?bczP=pc zs9{~9l!WxJw~o?ReSJVBxH6c*>0RrKkBm>^FYzfA3(@18Qh99u-tkyAB-U;G#9Vko zCfgg_(D_=?f2j=k%#XF>;x#7WF|c%Z+hw|T9%k%cu+P^wvfFW^eeHwib%V!lFL}np z$j@bTI`U_^{ij#Y+3|dk4&WK!JX{b(z*pkd=%N*_ z_j-v$x;qTbsolm$4Vg@m_%cZhnNlk!G$fF{a0gSSQ{F*;W#4v%Ps5%YYGPD>-l{b# zm4~D`k&_G6t5RGd8yQ_VCc|fJzuUXkTktE0+xC$JdI1VY*_D5HxBzmaU;Rl~(Otx` znFF3x*z)-p>aN&Ck8?ZVBZMYACVaa43X-qa`!84ol7r~0*F|J)kB7;cjoe>ih=+8|FeVS;nAAcu;1}bBh6Dd z3ZpVuBW*TVl8D)M;yvbo8n2b*6!Tt&to)USn^BprcHb@8i>Gb_MO}IDl%Uh2gaGc% zL8S)stGUK0*`G0@nbTF9Y$=266Q@7#ikz45uG}eyo-P{*V>FSySJoz5`U0c>jg5Yv z^v_^0&1Qv#oO*1O;t~W_{HWj83Rs}vNM0Tv0=F?e+4SdB;cgq z$-#=xi$#r1UkdJ%JV`GT2)xWk!cOzYE!S_|P-Y{0R?i$SJ6#bmp(ZW<&tHJ>Re$$2 zTtV)_l}F1Fb)t=%h23G4#|mz5c^mgByKfWy^f^oOtAORnRk4p6+Pl;g(V{In+&-qL zjM?jZdj9xcGCk45q`&sig_zul8%dps@Ss=0KpE@aEm;fv3{}+k`A*i93y4qeHVXuN zo|x-B2|Q#hZlB5fy-sUlDw_uK6Uclr(BkeViYO4iiVDR>Z<(~$BFkD5W+gI4sZ57XuLb+Rzq zfa(7{1EQM6FBCWIxP+v4!@9^%G8=x|7kg#)rhO&J|Gu9)!_)S(RC`r|DZYgSst(Y0 zhgDBuI+>HOGj1`2>C*@#p-zxIMwdNcXI2LeW?IjOqrMuq3n~zoe9U-ZvixPdpW5tW2f{foah_^$`eHmhvP$#R z!Hjkw>x}<2+E?hy|7k@iq#IlmkrWDg`wDg9 z2b`-csG8^Knr+*TJ30GSIQuKu>4Hxe?)YS=Pjj>QOE#Uyuu}>Qj(HDA&Q;AF+C{)| zdpQbZhGEN4)KqqeVJO$+nP^uQ_{sh^Q_1o3OT%}-Tj&SL+Q$PO3`}^L8*ENF5;_C- z+I3*QlebmLUHgw8RP1ZQ;~1_m*!)<)-i^6A>&-sVU$r7TM{a#BU(4MokeB^c8GZg0 zvpe{xE#3d#cQ?VB`8~&D=7$ZgZL-0*kh6FF-OKMTmN+{dxFBA6fbKeeQpHuWZ8n!FqD39U}#)>wLP2Hr2v!@lstPxgdxH zoQA4)h95_)6TT+1(Rs|``49`KC(#YFV5rBVe5$q9rpW;WW~Omz@pQD4wopH0g`eJ& z>V!5O_?4GOlj3qf|9h@M)01uL#~T#;5^K!1 zx8B-QE_s^fxQ7h9Y{e6Sjkqm-wBoU+Xfui%mz$#;TAN*5# zNcSFB-nYjN!-tWeTmv(dfuLJ#?%b|9LLM6LPmBAwC@PTW0zb@!Gk zr1SOfWRm6=iF9sTB=XmX0g|GuO2K@eD(KM&9fqeVOkF}P!EjR#!+dVRJuxYmP{JPA zecY`&59?)k(*L%sjQ-Nw5#-^2y05jS`0KRK&nrz}tUW7x@AesFS)25K`FrGdgykU} z&(r$&zVBLgA0Ei%RN=>6s@b@8zx#zg-e7?U)bfZGd4@!z$6DLGjfbC~`}^n!sy#qy zx||>Pt40hj?`SHhr|_Izvi0I#*Nik9I&7g*D6|+k0K$G6%3dEUeD-0pn9fON zpZaKZTO9iCQfkNN)BSqyM{i(o9x3%s6rnY`at*8TEJfmeP|W^4d}X-V#oi`oZt!c7 zQ&+*bx~;U@-V>i>cG2SHk}vIe&h8W}&%w?EYsxesEzT*Lj_aFD~$i5(NA3=ukjN4!Ty{`#S zvEX2iJuUF|`>n)*>am|V29#!EK)N(-^NySxc&BdF8A_X&PRK(?;RD{GbA$OxD9ZXA zxdv_!Qsgh@)qN>P;Bo_h}pc|H-D#&)FXdB$H(9@Yi(_OEBF@3$++&J5t!lVC% z5iIv`Nuzj+DGzDA?=|@1{8f!qG_9&b;!BY!%_6(;$v-){KB=GGwvrXq@p6jwLe|ia zV>-PNp8Q+7zvDJOClwLLo$);a>_yNVY(vU}0~68ea&j(z-(48BUpo+5t6@wF9v{Tp zJvFnI^ca>_u&5@4+zK=N(s}bc>_-1=b zwx{=B4c-Q;5`VhT(DS&sj-Ndrx<5{%@A(j=kcWz?18kWq@qHGpUu2KwJZ1n>U(J)N zR%$qD@9ZO*xTk=TFhAgZOFbUT$Ly+CulM&VI~Op_>=!Gz7^HvT%ysSx@AJV%!IS>( z)m+~zYXTlO-gZ(Yyc!EM{WglKnI2i7wo3Dp+B}}EDV~c`zq8-Reh&kTw8G2Z8KeIF zUcvn$4#djofk25AsTL}@wB$7LAk3wq>xCu7AG90k3_{C|=e9fOw|(SpPZj%E1n}X% zRk~;rk6$f+n%apywaF?wLkup;7(ST%%U34&XtVOYjQ{58h%t)+K*>%zT{sTJ#rdIt ziH@A)Un_)+Y~x<=oPXXR=o)hJBGz8Tp~fbZ4*u96bX~^^=DyE4aW42sM}EnLi$Z-E zMjEZOaQ6BuKw6UkZ2&s9$^q_2F8z;XfbX{1^uL11z61#aVExZ=YnJXX)%)Hs66ilU zTDg*n=?@#HX!V8*=3XZo#s)zg}e8)Us|d!y&qlGL+B>) zhIY8!yqziS9=IF2Vfj=0dA2mMEavY3a@uh`>gJ|?udQ~WP(O_4z^SWYH*kbcQuv5h z)VUk~?L9=y;w!kbnbx2FalN-m!EYy$a{=y%fwBRlrN*^qUbSB-24gZ)-<>_hXY~SA zRo4gdjKY;ZW3Hh-_s=I#7aQ)S^m%OQQ?D!69cTwXp8k_xkDF|qn`l~mY5ap zJ~X)2@6Dp7`J79SqKyB|Hh@^6?ZNncKK}**qM?tYv)4bUEbpN!!feNSWn$ zE&~yfNo#O?EOCI%kytbv*6LeSYr2C z^@6X{x76s=pT^_2)tzJ*rpLHSN8DZSBY+>ggllGmex$U6$gptZX1e|rMB-$^AaSdr z9qW645Tfg%(Hq%_(49)wlfXNgk*!Gb{Sz(Bc zJ;x#~?{495-&3ZL!mcqlQ1Xck*Q5b@swTzCMIuPeSAtmgNtT_?RWOS5`wu#?k%8@h zOrmc40_^1al-Lvik8|9f5N5mCEKEa9z)WT7QVb#-c$@|;wPhbkY1k3+6w_QN+ktZo z1YqkxG~FwH6hfy@xV9 z`0#btIEJeABCJNYeEV=;FdPhHiBhQaK1>e`Umopfzuqz0(0A48kxQi<%|%(3zvbi< z%8Z?fuI@2{hF&6ji1phaJcTjx5`upT?(Cb=jXM1B_u2Xi2q2(VW)hs~AHTe86|g!8 z9Livb9pdKGuIM2}2SjgWpJ0mC8I^R#v7^H8Q5`;9S3TT))n=&&mS%}^u|UC9ZKqsLnb;fsIo4r^G$w~_<oJF8-``j3C{VqnKrUv^|k)hM-f;Azwg?s z*->z}c?fGIxcS25zW;q62Obw7eGY-K&;HeZtCIC;dn9n*)ONfp`)=2V@$C%m)ciCQ z1HE!+2K3?|tAI@aNi$#CO5|KmqK)gCbSH{OQ1i%tt`f*(sAEa_d)VGi*1kz1^Ewrm zT)GhZxcHMry$FXrM;z!xcqf-Ts%nnY(ark$v6;~~+6s`UypgMzl=tO;w0@o2;rm|4OI!i4F+1Bv>d*HbKl2$;61IanFGa{$)NAl{y`43VWivHJ;G|Vb&%zQ7P2DaLH3UxPucSU(6jM z2{Q+d$CIH}BPA&Z0x~u!mz!Jf3Dh}(bu&x)s0%W~6eH-vr@SsDjQ1dLn1|%pAuE2V z%p$y%%QI49qH+JuHGM%4Kz%>ZpTL*Xz+enANh(NptjEgQrrg*p3NE2+?R3~fN{)~|xsUGg zdp1Y(=_4o&jxi4ac+U*jhW6<36*A~csyRtftt^n5IWBhLBIE)i_qet!KA(OrR zaUX2y&=R2Kr?<)b+7+ZJFb%oz$Qq;XN186Or{*IOABM!O!Kk$YIbkbDGWlYCZ_Dj^ zFHTmEiriyO{6(mgTfRJhah>})()#%7Q_76FmkbLJUA%ahZzBI(OvS+_eCc&~(aUI0 z`gNcod)0^u*-g<@!c+T3k{dD|J)jm7fy;zD z0jHK_gR*ajkm{tQKMrUzBNOS5M!&ImTZw5%k(1B&OL|2p!dxz7e0bvS?!Q87&HE_4 zi*4(#xAj|&39G;NmeOu&5gpq=3(wRg#8egJIc}1kIBQ2T-oQ)R9Zvh22Y#uW%>AL2OM6*|*d;#UJ@CUvW{5p4?G8Nr&sS4xM`-0z-b}Ky#aJX-&WHVq} zxAR2<`~#cSuqIGvZftBzT$?|XDeLZ<#S3!qto4o(;WAtVOndGpjFV3kABPamDmt}y zLmYYMshJW48$4@-H|5ru-|1_IZwm9e5L_fQelMshMpenmzp2yh7MpW2MkgKP6j*fR zj&7trN}HUHduWjvNQT$iQ@HJ$rSgtfzg@&dJ%Hqr3>} z5O%}1!`jXLetvyCQg@Z56OI~mF>Xe^E;R|IihWPXP*D!&osFmY^qKrFuW({@ z;q%D>O|$=~FG^Q)vYqxuI`W2k+bR96@}x2j|Ex^pFP*ym`eweZS?14hEtnCn<#e?I z@hOV#UlMo9e(b_}Z!XxGqcM*6Obq$=tvKWKt4nsYwBxdUC``tqc80j-&bLH=yX!}{@}1^qZ7A9K9kJsbBJGJKQ8hkY(o zv1(ozykY*HWLTIuI8x8d5FH`?Q@4nwEJ?8djzAOM9ms7DNin@99XXMa*>xKqzK6R1 zUZtOQgsh}T)-S}LVOV+owW;v!C7{3_ndI!yVVUWCB0lFNb`8uEytzUECVo=qZpt#VuK%FIe=A*Rp!7`bK_k zN8oOE6tT@AJ(Ib*wGV*etwNYm|FAS*v~&Ya(E$g~=nvv&fu>EDRA^`SBH#~^@yt-$ z`C30b?j5xwVqEv!$JlFOBZ`lwRt2QGiPx7~zeJce;3FQf{)LgjPg=tQMPUEh>G~hi z5@sNk`;2&YWA~pA7G%n)d;Qxr zySDTk@)a%KVFa`2V}2r>ve2$Si+q-)F757FTj=`t{tqUwDk%$QFf1NlW_ip_la_KduK>=DOy)d@~ppwB-_K1nPbG5`|$8 z9CB#x)VisT`ijK$0zSUNZ{;vJTqW(OBHNX@O@@0h3NvH)>`#H{`Z9017vD8Pf< z_vJ`;TY}v?lOR?;j}K>xI}QOccXdnVSj&ybAaHYIeUC1z4R8n`m%XrX#s;IeDwd0~Nk3Af(H7dNTf6w<0B}^GN zMNeqr`x7@gNW~pt{?xYu{M7dn^LL!G5Pp*DLwV%yL)=?Ir_Su9XMa1F8H8%O{L)C* zEaAht6sTR){-ph9NL2~rh+BlmdWd}NMAIj%B!AQHASnA7oNl$f3_l(lKqk$jxz{3- zTzW~~LNp4yCsG$|cmB+khJ6UiN_0heWLFYjyd7}xvWs4C97LVFEz90QGZ{7GZv4j_ z$kDfV0$@Fg1aPF`TlMFmyM6dF_oYOkmGg)Whc?otYB*X;ckjMdsuaHrYGI8=^viE9M+IR^OpoNi}n)kymO@FvK>!_(ZdAx5C`b{$Pnem z#Xqaz@s@bYC?T0`DXPi*)%@Hp@*wc6ZWClb|DxX&CQaaxE7Nh7zs_jkaT z3cru9-}!}jZ=?J<9D?exKa=~TQ5Fk@T`fkdc-RZ>5t9j-oZK@ykI;konojUlZrXhv z;kl63u$cI~T1`M@jz4c4JPMqIMJif@&!5TA*}w%KJub$mXP6Gvdz7zL8@uH@S5Q8; z?1zxW_CZ2-IlORy_~Kp|xCsR`7qTPf$AP@}JCoxyaM}~b)9=~;9HF8|=z9EhNTTa) zdqhLsa<1!MfrdsM2rXe;|3nxlln2iIHm;d#&R3N&1h;@ zjJicea4n#w#i$zf-t5M=X=__`!&(3Zyth#0U4GO}{wvRhLcH7;{pTnFYI6ysD07^y z>cI)xPB|l;mpClZ1|=?T$tzv}47IqbIULBKg_qg~Y!`RW(r}w8^ zARUz#UU1|F1VSF@%{C1J(2~QzRDix{u8QWpHDTMxvCs)?-A-8KwkWT7VV9tDZuEI5 z$A7w?dyBHiH&*;ye+)_6%^}=uHuCzxD>oV~Hg@>y9BfcLEEat>*c;pTGun^+&v$mP zw7Xdkjir1oGAw+TQj_srozkvd2tgsA*LrdpJ0xi(SWt)s)8n0k{^yu{8CiZT9=us7LO zm0i1RFNgwpR#XvC5fCMj8mR~Zk_udXjJdvZ{)i|=ZMWU_ZgZ`<6yEoF#_0X}cdn|X z(jAYe4Ww_(D-2AVAl}_t^z68k1ODe}ZJMeHi*6Pc&Goh1e1->}QOkGnNsgV_r@imu8Hh=7MC zA&Zf$r}vuGX^a<>VUi^N*KE2m?V)~_OaF{2)~!V9sqrycL|dVb3HQ^@e!@~P-`=TCik$x3+Auo)!U3xY>(@<3Dn zc$i{o@n0tNN>VfAUN&4D&B!)@&9Trzy(zN2WudlXW8(T$9iobEAIHR`cl8coK>jM= zbujJQ*L*!q-Q&mZa@Yxe**pDeHJf&RoH}*bY09|Bj0uTR8s8H7UMD1dMa%m437eqP z?p;Iv8aOM(ouWWMV$hvTyy!WXZK{LwCjY_iHdh(h2g@h7xVh#n!B74cBsQ;aLm`i7zzFFJJ)LsIEzco(V8G! zq;G`RCLXkGE*V6F)=IOUrqN)lgE!0fh^E`u7Xc}_BT6f(J#!u2R}k0iVcf*-vFK=V zYI;>)`*EVKlT7%LN#`@+g;DmtioP)F`R`zO2RIrjyach0eE=g(KQmg#8v?%7u6JY>p7ojz5Vki zGt&Q+0mq;#oiV^Z@)z@k7rgq%Pa*@*mMkA63zFdP$Y1l$Wgwv!_^Ve9w;Q~Q4_P(@ z(|E_4-=KGa?-|U&D%NoO0A#V#<9B~Q(jO56l%u>;MG(2KlrMMUGv{#g_zh*<~D_?LqkS ze|YT0uWro#Jn$two$(BAmruFfuVI8cVs*M7UPpWz#(WR*<>${~E>Si{^uVX$p;HmR z(1Z*Ae8Ssj21ul~w$~g&YG+`fGRi7D_G?n*U-L|9xbxK&?lu;^qZln?X6+65n#2Fc zPn-e;6yrhJ_}>S51D7+~EI!mzTn)G6uMkdw#YKi!F@k>4Ei4n+XRY>6J~ADLv-&lK zlXP#YlFhUkK6Jecd2wYs^h+1JQgeP%a z2($aq7fCw6#s$033Kft^D^np3Z&qa{hv=O7WGQ%;8Tr{U4{t zAPDV(iAx8Zz_TkyZLmUfO|tuKo(sV<%R6iVx5wX^U5NMp_6#MEy&nTM*LlDE@jJhP zvTLF2YA-KA1u-pwOSefV8s?{b%<5wp!zlT=JW;X<*E{~b(uJKXtIo9agIK=I|J_tDvyJu^^)TSH3X3S zrV;zBpWulb9J%gOysQOM$&X`vYm%QP)q6>{NSw{UvAHy9IoOY3yRTm7!E;nq|2=o^ z#p&?A|9ac`dw&i0zrFHCc{NU;PQMQ#b5`cY#-)sS0}A^uD++y-7yE}-x7hcr3)>OmZ_QR6MV**koNc#o|6)0Rj!(QO{_=S*%o|M|xq=WTEy`pv$SnmG9KMkd#)Vh%ixZw2QG= zP$5)`u5evTCwv?`;eKfHUN!dP8GNxzpkAtv8K}5#D#yp8;@p@QEM)+V(SR8*)my}8 z&J&RmGl@2OflHwI!{f;G88$vF1om}7XhjdRe3xhk{dv0_J)>d|Fz&d`2 z{Jdw)z3>uA@PHqcW`TXDe+&uN!;b5w4a5>f>R;c+S>bIiqSoLdSyuMmVOM~+}R?jT=sh9J`sGnF)dkZJ4rLK zn-pP7DX)OFJTsVX#~Y@GUzD}#Eeij9#nI71QNc4X0ih6x`{=8>h5Hj(vwM<)6IeEK z0)FBWWsXYRIPMfaUZ#N~uHJZd69st9a(`!*#7aLW_&0Ho;mX!;E6IoTEQ$vBc@5q$ zCzh*(s6$7F5g=ZbJBA&kBmZSE$=}Qqd3d7OBBD0hS$x*A_ptgs9S_A;?EFHSf3~1O zDWv~FRwSF{`~=L@Gbs{nbEE3zT&}Y)9`GY7*jf%m5}`QfmW7<)J>#{9%1F zY`Z-%>s6aco~;J%(0_CAij;Fye;qx+C<6lUrzL|g_H^2KIQZzoaTxAlLzbpIzU@PV zxq9$BsG2IV)o82N2R~k+93i($MfyDEKlvpCfp5!{#4SM+mGE6xfXTYb>5W$5u4cY1 z{_d#U{vP_NJ#RxKkM~XFUdkwUBEnm?z50sN9FK)7o_OZ`>)7$jsOp85RWf^b*G|nU zo5Yx`W`Y1I6g>li$!7oFWn@<8yX)ir{Pwl{5KWXwWb-DMsxFhrhx~mG^?cM^s*i*L zd@gnP=I3Z>9p?|+^m1E|83QgqXwMwM3AQl*a%V734#ge&DcHL8pO8Q$#|q_f`#a<9 za`)ScTDpGmi4-#|`D#yReSCGC?SM5Exp1EhR0MF#6|HukD^+48pzNPOtF*{xj{9aR z-L@AT^LpG%!ROk+a%Q*%urxa?3)&|g1^v(4^>VyWq|R1*D^|08w)aAEF;`?3P&7LD z+Je_q{CEbd$Z@|*D67cP8=u~e!{i!b7)l+R`)kzyy6OZdyg0D#ulCp2gLb<=*(s3VgwX|skc|0aBT%&}kL#azjLLmtJx+b< zSf(0E-w}Rx6!sQDzWkVLZm5*;NokyE{`%DF`>!$?jSMOh7$#KN!XWu1lt?lc1cnw^x%FV_s+_8c%GLLt%L-SX$*h<3t(xzgrCB=ZZjaX^EAR_6o`Z_oWX zH!Oh`OI%cO>O@uRasii`PO;^<;WF4mk+=N+SxEx_;eWD{QU0{_9$876R_$dJi6gp} z+aN+~PUmAQ)?-I18jtY{)P&nXBX<_RzYy|F4)Tk9kZnBOS}jM zNPhQ^c)Hu~)Yh-#Q+oG$UEM%Pr_v1(EB{4P|1#`<4oI-ewlX7nT=Wd0pD}IX|M_Ik zouA$Khc&Osz(7Tr`3);btveBu9t;dF+#f=8t)El}R|x4}06N}?+{_ECDWAyc__<0S zQwdd4x@^}A_22I@x*jayaE`Rf#JrQq7?(^*665G_=PGV(X!z0}9&lTZoxsl`7RK>| zrhlUuelxl{JV_&*y-AzUO>IUu63mm{-;Q&xA6xppf&!b&&&X#Q@8aXW`1s>IwFACB z?}9Xj4qeTE-*x<^pdg>VSq7{@A5xz0E{I5|c&$t~=bCEuc>C+(w;h8beVG0|L#<3B zjNJh$r3b#_$+BT)`KMsqdA4uL#apbl(=cqsr#us#*&kI*$C_*Aw-6I)ev&QuodVdx zS}l^zWhfc z>Z3UEJOJz){$(5d19*Wt`tnamsU#gUQ^+ps_5J5n;9X`i9z43*NhjE-C2@ySo_p@` zug4YOj(WwJ{7YJ=?H{*bb>>EN&iCk%gX`p2alIGQ<$O1cBbPnF?kQLojE)Crh1~OQ zuUugX>qg_xPj&F}W)HX1G_rw@A&*QFCf(u5g|R=*wz9A|JFq(piven* za9ckEGfbd&P01-2ksAa}3rYv|ZmoAOLe9zJe%!`Wr+TJGmpMW`U+T?e^*|NR;2TAR ztkT1FL#%DNt_XES_{IyTg}j4$!KYerAIKX%q)WyU={XX-6^fL@Dp z&9m$fS7tA{Ib4nmEe07(Z7zUW>+5YN`bpt}ls^5(qgQ_gvKPRZ^vzr0YXUsj5uQ5# zkxuYfjEoH}13%y#hMyDqZ$Ip^`p1Dk16|Yv%KgW&P}l!{^=r~ntVMu%VxBMP5gRg6 zM)SMKplqP|#CF^>JHHt(O0RRIQ^m zJb=S6FmmHt-blcjSnY*RAjuwYREp2!8FsVYN1hL%eN1-R->cgyN?E>233Xvj+$pNz z8kH_4flMA>)HmwD0C6e7c)U)HbBapUc|>=LBWbpM2y!zxZKbw}0 z0Oxn>st&KGJO1qPy~RDFtZ z=*L%NnO`cM%f!CPu>~1J7}wI|-Dy1*hQ3#)B;mixrxd)URs@`~zP(upXx@uNUOF=S zff41FYpywX(}ss#yDp&XqoN-_=}rILOmy>@nX}a_IZiV0j_)re&Lx9q)d{f9emkDq zsKF~_JhfgOE@+$y&oyFx0&}l1);Y#2yy#K$!a*L5!|W9NWZQRU?0|%@lm%g0%40dG z9P{zN>~X{64!zF^FkUk&kjHDy6ZZp`St&hszY-C=AIcs<|O(8tGDsE020#mC8W-tGQoBiDLNtS=7si&f}s33n-X zn}xZ!-T&UR&o1~aAy}l$_(pGf^hRXkiHT$P{@Qs9vfI&IRQQ{i$vr{yrKPja#SGlf zHNsPP7ValU)tt4G|2ieomPO1ULWlkLyAtXzGBzpzZztF)qg+TB8u2feqqx!EcM0Uv zX#u0!qfPkC5&(^|mQ*p^}xIwq2~!9xGh#Z9M!Q;C2}q*bocG$k>IUljQAc zto+i)M-;jDK9)a|JI+OG$GNRbLYu3)dJPpIKVHoWY)&uo*Og5Hnm`QxSTZ?DWwF0- zy;%S8OB6byC7&+OOT*B>Z5u7tq*@o54=wO5MH7;BTc)MW{)biacr(@KLucSGicZuq zEBmzbxYWNW<79`oj_2-VY$UrA@arX-#XF=Q3c;D(_v75E{D?M*`vsOx#JPLrC-p!1=YQlLyb_ad5-M~(_dT$KJWUUdRC=_drG1HW-lyK zaNooHfRALk>KuSV;kRz+4^B-%r(YKg_ng#Xl0B?QpBK+qFPRDJn|%&r2feS3K~glu zw@^4G4qb=UYsW`I=RJbZcA_A7O65sJouJ}t^K>5`sr#syfW7S$fB=oA6V{Ssc)}Y4 zabHi$Q`dMu*O}A!m{8`6{<04&(7NuLm@4=DytrsA(L4IS?zaH{I_!HEp;D&u*SU;!f~-%Redh+; z%h2Ok)w;BX@Ek|hy)T1&&2L@rl*qbyhFvA9A~rmlX|c1)lkTkrYc5}Fz0WG^ZekAJ(0Vd=~dP&v)nLotID?n)T(>t8NT4gMTqt`H6FI4(yv;h*zVeAal!u^OCiFs&D7-=`OjocK$ur57;; zjq460gO*DF`#nHpo$(CpxXbA46Hc&IT+kK&1Xn!as|8VI>RY3ui%^9s6BPTo_b&!B z2P#{n*XfPgjDpgoI-}aVK1h_ek=FyuW>a^gcn5#7?`J)xVZ8V>M;1mWi`Xw_e*G@~ zooVvz@-h)|gY9dBFtNQN5hZY7{!aS&Iq+U6AUG9E$Mk8GNi%XeK^OKYyWe6tE2xaq z5;bzhx(Ke820PxJOHnf;5;GsrOJ2_IoBDTTbM*KAM-A7Z*9Cs!()n-!y0~}z(&9=; zFm}ZsH&Gy9C6d@M9~R7)r1eOx-lna(@-b*pHMzQ{Ww);B`kW_soF4~?6vmsESEt%8 zh0C}-gqI?L^S#y;WfG$O}x<=I#R%2ULl`hE>sA|SIni7t*ZhlL$Qao>`bVp-P8p~!od zgd-J0H!g78<1x%Xy6*W*dG2cCD?q->r__;Yl-)i4NdQ8|aMaSJT7`Xyi}`9*Df?|7 zyzs~99xlpXCsaH-d1(pWcH-)*3wx8lKvgN1Qc@jIMU8{{IIA-f3 zPc+2|J{Z-PR0`x}co~}|2U=ysvm>~f{Oc&zO2r{Z-!0^kBh6w$nwx3T`S|(1f%cav z-oEmc!Wi}gq0;jQ+nR;Gu@2?(;t*>=N$1C9j{nr9r^hwU{cnf9Om2}+4)*3P>#~+k z6HM3gfA^#ZD{)xI&FWgjT7??Ln+u>c#Bjch{e4y1^m>kZX=ickZQn8TSbQxMMA~(Gx6B^KjhRWRf~T6?+3k79B3An9tbr zPuXxLRzS45e?>eO@cYcz3a*B>3$woM(0lOFDytuCK-3P-Vk!>0&dj>fE>?UrPmw+L zTMSlU6;F4SrMRXZZ$Am4Wr!9poquZ_)^1*fT4BbiFp5=-WZxr0mogQ62H2&u*}-Jf zS9qmrU~1q=%8$>#9rxq!885hv1`3GaY^mlP|FmH5^;X&ca^GRj8g&Y0xHjv>gb!?0 z*Vqr5IA;i#9kCkW;2%lM1TlbQHkV4Zip*YFTSXQ`PWBJTz>d4-bm469hufnA`{xZU zqk?L3(BHDx#^0n3Rey*?7=>~+z7_vT_{x9M>VfVqU1fH3N34?>@6rJ(JRMzG)@E;2{XtW@msWxi+$dI^%V~OZp5> z;QHk=3O{xb9_rb>7>G3g2!!l9-7Ag>FLK(;#yMQ{J^ca^w9W_lK-@kkPW|iP&mj9* z9;!e+2884L_l0!+;HGT#t!SgQ6Z-sq*{#I?xgv@k_Vjl7{OhN34GPi@KO$tkN~Fj~ z2On3q8(2s=7Kb>!d@j9r29u^L59ct`gdm&LY6$V42=KyZ-W7Y~`lz1m1V(%Ic-Wm!sk4K#>QTbNfYGt+iM#zDUavLbw~ z=ncKed1&0-Q)9AkFfm1%yCz4(*Sl(F*U?_oeKVInoi?X6@G(03QP`8qzfZ&Rm;~63 z{s!IU@Cx4fT;|fXIgNJHOy_Iy`hE1f-2G(sR(9W6Z&>N$*k81TU~JYQ;&Lb-xb(gA z6ih771^1yo58=mo-wKmBJI%E*Ren(SBZM~MQo>JcP7)7w9IdmtGhAXB2_r&vHpZ!aPi3otwC2B}ZcQ7^QdoExn8VNHPaDpqE>1Y7 z$jtC~4res%w*yPi{ZZNsT^WJ!&-24PE6a!|)l|gHkK<_}-?wQlYxLhmT!UOp8tD_+eDE9k|!vo4>AE7{t;l7Jw;k*2FyD{5ZatLvfs)jIeyjM{p)S$k z1^3r`V=pgVMipXSyQ%Y1_9LFI_&ej)bO`+NBfW%6Qi zYB9=}<}|2K-)O(f-%a}y&aZH>`3vz=J|6Yt1ACPr7Y#u}2&wSxOf3Jh0MBSv7R|%5 zo9b7hyK(;UVE>G5QoSz<-Dd>1S#RYke0nT4Qz~oplHi85H?9D~Cwg{pbCJlhHwv7e zvSQ9JoR5!JysF!L#<9kl``(-mCWB5JJ~j;N3O<|PUDE=dZRjYI;;C;}OJ4Yv!fzE? zp7vm0zsY)oZr>)@~R>`SxhA>gd}L5 zelZLhX9Vbfb8(*SV#GCl^t8X<>(#h3Wx>*2G$~UmuSNP4?U%UNZ0kyAb0R)(w07<+ zTnw+A7Ak9fDL47OLNw$TBDV#U{`kJwEJ~YJ zZ^iA0N##ZD2jgDDVzYh?=9(giB84w0!byRdBS%}Eglvl=)7$CgBFu1mdmi-zQ3dJ} z?_SPjbv)dgVm&S<}i~l6Pd;9Z5A0LE#BM85QQXF1koM z{7QTM1Prn<0CGhlk*X{|@ zXvf5;Y*1x()t5^P=hHr6fE+F1^wxF@vhi~+*sK769+SGip)F)-uX4-v5zEH8B`$fo zIp1q2Vkh@&`BHuqqHCY;7c9ky@;)@$ovP?HR_(@@(xWceeRcXBcI-6?kd*;a+C_>x z9dEw$;~405MUG&>Xs);oz$@kaq0G1_p?aW3w-ha<9Dzw4lF2=3AMk+oY^OS0=VxTP?tfXM7G(;`89PRLsZDX1A&K zuEp{rzEITto;az_b^oHfr_hIA%933YwU}=@r-Vd23qnEw*5E-Q6rK>ugC4x2MQQmi zUr4KkBR@iS=WT{T-f#NEn|- z!8vD*|GXa{0jkT(dGWsYJ)9th`sPb;#v00T)X~VP7IQmb(#~ojnTI@8|9%e$NGV6B zMGf&%zD5nKtULpFBMZ`S6uP2y6vWJo1s76uSOBg}n1^KsiN$r^xcFCJKdvYFYO%e< zYcWEe!GnmKljZ)MEbj=0Bg5tG&+tkwsUYU=3HY$>hLxMSqsN{62A24)6qyMHlykv3 z2EaUWTFmd9p7JJ{{yQ5=&taKzp>$+rMI_2c@8)Co!YQA2qI71w3#(T0C(YSsAX{sH zqc-oZXm82$-hzpdx8<4hBb-jOQZ?`QMX- zf`tNACw+EUOavs|MbT%Kh^{=@Cx87ZC9bc-Ch;x%dxU)&uMPA-~&1OFcr8v$FiA!h!hO-S%$ye%^CZI*;WC?u~j!acmK4 zA}6rV$EBFH(%RGl(4~nuvKu+ZPIJ&3!MrT1aa?jB(VXIjRwau?UQ6aC8ST0q9%Ksj z`pHVQ6$&%$CO1G7jEu+bsK5>Lr1Zu2`@W&*{(hd0y(PpqjaEaZ1)m2nKzere6-S|n zV|5$8*xKoSzQ+&U)w&Jd1crG zhbolLW(4yy_pdr{9K`5@Yf8!>qcErVeOMYuIv$gL_zXWwafnn)_eJed=ars_?kaBI zGX^T3CvU!$mc|lp=I>DDHxGit8@E<@Jo4+cp!LM*`R2VFYy)jEitZJA2dQjaKT*LjY3ZG6< zw0M>Y{&xrhtGR#YWrf!1awciu+E*r2x%_EH7!_8O4%TYJBuUKR?M}$|xxe1$YV8Im zg;}i{;!1!8_eu%wm{~o=F>MV()fo|#JqQ|$W`yQ3L=v=V1nYz$vbH@XIO}cYaDb)W zJSzqtlqnTx|%Kn(2ojBvga43TIx=91gkY%;7|52zgz4cK!sh!<$aQ$ zZ;XjpmGL$%^!FKROhhsGmzb+Yd##X(pv7^J$-wT-kN31L2t(~27e+7tjtgbi|3Gr} zQ=|YN@sGM18+KBsM^)+v#K`yy#aMk~UadM;d^j(Xi}qe(tkRKc%hv7sv$?Bb@CdCE zAS)dLHPTC=EdTU(-(-Mmk&|cb{60Uwzh~olpHK5}DxwfAru>Z;`r^Jp-NDkF$WlAY zu7L-wJGe@k@`WdhtxYx|P#Wi1v3%41&!<#C=>*IbK2DGxcdUgFuA3VBtPiF1<+~zPT>idSh;8e%xy9X-Y5h7uCceDY&Fv>jd$4mxgBq>LZ<_Y2#7jxAIXBthIJqQR?}z%Aa4RPbG4 zyLl`x@OT?T#dm3E1bt<{-X9MSu5JSRCaRn0mkui)pEx$?`@Wv8c5)S-iC~1~6?w!; z2>{O-82(PO zTVWG3P59vkL-ha=HFVzsV#1wdn(&Hv_lnu+c(28smZ!_^uzM<(k|Z+q~EBIcmcWn9H>m(h+y!r9TLJdX8%=MRmys42;$#nAoZ7Lt}8=tB0?Z8;pzvOJBDHR3~%rm5UEPgD`bO$(0Q; zqSaO40LF{xVO}_Fs;K8gADex1X42dCejKZKLs_*I>5nvcy1N;eF`UBJ{U!EjHNB%K zPS!uGFUfjntt3dHPC!rGCu`ZI}PdA&0xUN z#f&9A1*)3@?5aO)PWaqPc}mzPwPVLO9pRZ*Ji!x*dr3O18gc1z1Dp$ivku?jc4s4A zf3W6yHT~J=0E%vgqvz2}!RDe7klcO$-b;eHDaLBas0AxaT{cgs+E z@vfuX$WgFy4U-QCySekII+*!r9%^t=h25YIW!dq5; zAJ^o1Yxb`74Ym249x22*9A`PFIwZ7HQ?oIZ#CO4yCfg%m}mVNzB6_GLUO#LHo|3M({rY~D2aX)f1i7aB?1IfzKI zys4;H9|~(_;`9yhatZ7S_{@=BjLPTkgS-cozbup`nv2&=HC9-xcR9?VaNkMbTs(Y?PD^>!@!pK&?Nzv_d5!h~Eg!3V{l`&>+%{1M+@M*8LQ z%Y))~_8yOYFaPt*Wj5GrKlRBBSh;-G^L~+~%D{Q<=AU0JUNbL)5>4(LgDi@0y?h1r zf7UrAhfLVch#}S~*(x@Cz3b>*iM@$`%T=vz1?To*rHlY%800xUtl#bN`9?3irsv=I z6WLQ+RjWQs2<_%}?LB(`)=}zw`yebV^uXSz;wr1 zT4Pegrd5YM^-d?O`6y zM_8m!O5Zq%nZqBciAK*T*ua79X%w}7H|81nl1_UII4AJYIOx)IB6a*FPa@AgSH4Cn z>#Qw>`IfI(Umkw{&-4C8-Ri&>I4@$F$ANi_kIPs%dtr-SVe--mHp?i!ufGyTbBP%I zH}lGpE|zQqg;YK5-Zva68et9JcJJ?2{BK4*qwu~26$dq*Pel^i$MvzvEu4z4TK35c zo3-W51MfvncIW$$U)|o3yuz6nerM@FW`2&=Cy9en`0Qyl>q!}QlK1WP#yBy!MYDh3 z5AFlXpGKc$YQ0E&_L=Z* zVkVdD3LCu#tNA^8bHsaT4}!(dx3o{sRJWuKEaBapC=wYq0lc17Ltz4y!W(FS#mB;@ zhS5(axC2M#cmhxUX&&rRE!5@aVF897!RaQ(1yy_Ok>Qsv&PY8{1z1+JzP<$zN@oIZ z@9hnt9MoYH_y~v0V+UJ2=KV|ytF7+ZqfM3z`f1B^V$KG$S!wQ>Phg_XlYJzDad$d! ziAo|m%c+uj*FWoo$$Iu(2^sOy4USErZR2jVwZR6xu^l$nRi<*RQ zGTw!*eF2>(^j6vc=swHX2QJSsb1xtu#)hwKBp}fV!$?bJ)A3;M5%MA5A#4tB7q`vh z{;rpAkKUQ#gZTI|U?^er!nz4(nuic6~^uHNfW>Z>rb(@DRD$VIGClV)JPY3@)68m2chTay> z<#9QJYqP(ctV3+`yxT!YG(c;)&KkxtolYFo&)+j~JJR0OI5j7GA6;WHpdqkhY+NGP zEO_?r?4SkP8!{+-A@~}<8&sakA3oDoU$x@Jz{iEDqSMR zprm%P-RTB)0nQNYUvm;V)CtyG_*c9X2dBQ1QRbk;3!c|l#`~u6Ns8~r56KJkpYu=hm88rLg+;mr^IqR;X0_Zh$P9`2okan)Uu6xnmQd-FJj9`nhJz9cIS zJ0G);n#fBFjAP!IF{!>^XDhx~sr4~}qF>~vSb5&61)M0) z^{J$Y`}NccE)3G<`*{JknGN+be(&cW0D^g3OqD;MzLZKTJVl{)3F|XZ_~N=E!$eZ< zuW^3*E1Qe%b0h*dp9T>vB3Gz9-zeVu=HxF&V`}H52yTm`+Ipu70cUkmxVs=C-RKb2 z-G|zl4f&2X*JqO_7av-V8lMklcYfO%8My@9$Zpe5AM@IiH1WvEX&=xLb2=L z_o*H|S2gm2yMUZ++{%H=RA(;Ua)!C&X+xsIMK+Tde=MPe%jbT{9AH=V&wJOe9Hvv^ zl}pQRmca^p5t%gmdz77@rN{-~v23P+C41|B?N6iDzfuF>oi4P2Hs3pRAQ>XvuZdq_ zZ;1IVUSJdA{JDO>5b#zUrU9*aI-RD!=YlE>Jmuro&;PYX+YOpQNHlnkemJ)yWLy`V z8~Q@5b{GL$@RVo+P zd%Rp%L9j^)j5>t{XXVX7BZY@-csR2V=;Ck5%)6eJ_+k)?yO;ahb5hY)IWouOjmQW*WEqJzK(?PJchd-gUzeXMTD*l;+V zbw0mP73SI=bJ5km?8Q3>g~dPsQe9S?7?PxrnTuG700!b=W|Dc4GDK2bKSP#@L?>Q; z_92fbUx$(hU7c8EgOWoA!U1Xc6mt9YbqRWKNBHI-XuFd}w4TL_H!+dlm&>VcQHC&K z5y^Rnk{f~OrX@C6-j^+~IbP9iB#IYNhT{z=0VlK4{hHF?GNk(Jui$y-W(Pn z(+;c^(}JJucAG~>2ESx!5o;42zvgJ#HH9f}m6Xw6chhes_rTt5bb^O?XWwZ6cEEgx z1InnFojc)Ug zABfJdA17ycA?DB~>XTww0{G1SbDq(?Fqq`LDb}0#NN2FVJB~$d7y#jc!|I*{*G18&+b7wS- zUcLH$uHE6bfL+rw*bg%Dgi2g0o&B-i9l`U_f2!ZN`JLTRH%H&z`jk^8iyK46tU7!+hvK zj}K5@6<_X21`3jnOQ3=i>^br))V7batb-*YKXteV-Zls?}>dDLDJv4ynO+%+7_ z!k(;h*ZS+auLr0`=Xhru=q{cDFHu}^KP-2SU16}ld$bR zUb_SsBD18l;iCPQ4OD0u80B00avcW)aAE%vhi+p&rX(4we8*9&-=~rMr4FvC2+9ac z<=?EqxS7a`rRUH6kjv_Q#Z>uJZL!_))JE-(z$j2y@?MA+`M;<-vvy^*^=SVf4nTSi zL0Xko6y!`1=@gVs;MafUe0TEgTscX0lC|cX%HSFGRNa*hC696=LwoHBaaB(2?Yom+ zRsz*9b#-j(_qL;1$AF2>pL>3cI%-rZj_f37LymiKZ_Y%EyDR$%LYDtMONz{;odV(4 zanq+TzFx#{#b&LoNGh^wfkNzyp7Zb(D%nLFUojN^Z@y=P#}r#OnR?CboK8nE#DH$4 z+A+DqaF-f@g+V>c9jv<_=4WVEBD%cFZ_JR$YsoG&cq88{D?JcKxfrDEWS8yUXpZcJ z=jZLmE4^rY z;IC>Jc);lcuMF0`JN6vv_r>!H7@mbxzVEJ!$lKEUkf#wcGa6EHw|7U% z_rpt2m{lB=T(x9xjn(jV0gHwv%s8$oZV5;JdgF1qazXv&*HA)E=JHlyUP|)~Q{esf zbQ}VE4q=*S;f}CZPB0D00OM+h*!y!J`of_! z#ZNAJ@XUrhJmf5Luy5=d+p;VLQ6_*4l1KLfUilBND^x1UAzx(g_x#$b|W<1@7dfj57(h} ztAIn8>@PeqiWp(I(uA0qC_F_Syqya`4&O z)tlUKj(Orc-!P@;vFHp1Umt9?;eelMvsiHPh&ykmg>N2*@xnJI1z$qi?b>@4_b{}r zUw65IQOb&?Ff}h8tLYf@e|JVve$H@g;(Wks3bluTC03kWl&pxgfv#~FP(}i@Nvhk`gq;$Kmm)g zq3^DXaOZd5yW7lNw!E)?dBq)+u%#OqwLc)hvpMi@?dOwfAnl`D`Xi+|>D+GCIeQ;S zDsIasT4UkF`+KY?&VF|d&y$8Yo!=o^!g+j6QBmh;bkPpfOg0=eKrou^*W-A;E%z*x zOrcLh@kM`a)UaYHLrF>Qi(v`5n1|Sly{py@ox>FRtmwDjFBYl+oX~zU;$7Gm zxOR=mU=#MNEk>GtCI2Q8&#(>iDpk-B6FS) z#DnV1N~2^5;76ksed#uv_m%c76~=^hho&cN2+ZG^{F!BJ`?vCH_lG*ZliJrfJf(_w z@-P;^d!1A00m&J4h#cdrI7c6sv&-CW^H2Ex&Z(xw5_gzME&(=H-kAzBRUG3_uKT4w zA3|tQor(?FYPp-%>#0VY_VN51wFME-M+BRE{LaExma&z;?z~mp{^m`sO`lQ5gJG8* zxip%XW@DJ4Y7+=UPsu^*Qv{iTvCU+XJwY!DJ4(uDWIueX{kD%0XS4RbcPLSL9nQp~ z`~+YiOEk@*@hqc#bh6Mh)`5~TzbApkUYyD2~ZR@ATs6<@5T3)xo=`y6$gb^i4AY^9AAFHr- zxU=Z`fp?>MuN<}<9Q4_zEf{ZtQvP-JCFwdhZm~N1UPYI%yT5Lh+&EUP-v>9Rzwi=!yPV}OKcGR#l+umL=PtDlJbK2VrLEmavS zie0hmKGcI5YJ1-&4lLjJRXmNQDsOT9nZ93cXe_`hk!0A2DIPHOiPPDBNDZ)4@caAE!$LkpQHeKb9arUKb8#;d(8MSu&o5qbDM*GK?q}M(e(_$+1>W1@ zs2&5Ylrth4Y}MD|(~0}}9jM*q1cUoS5g=B~`v;+%PwiDd;>}RbQUZ(td9Vv}tC=66 z5U;%q7ga2m$(niLzGC6S>FM|3eg4f$M%WARaP%=j=&eGfBKIwAQ>oA0P4<8Axp*NE zTLm@l#g|xVgRfCHk2JqH&t)I8`YYe7OdC`LPKl@ZSHhkfoEj^-3jZRq{>Ss_UTtQzuqiSwg zb=I3zb|aDN@`aZ)0v>QkALm3cb+2aqKwZ}xl!(tlR}xeKIj;JudHVY^&0Lf-pnfH& z7Rv2}XA-*}FpW}KBr9ReaBons+A@-z=IY3e?D|T6kRSYQG%Q~UhNnnB(Pr7ty|TI7 z&=&TF0)yv_gb|Ow6VwG&hDPBZi0zyfcNqHB9VHg;n>Qe1bOLTq6?d z)+1R4HS&aC!?4!&aWbNd>e~V`3V$i|{d>^uc=a4fANla?FVzM+Xo}%r6ZPuj2EiZq z3eet!bNs!baKpXnzw_@{*KYpYPT!6IoMZnuq!WS7q-?xs;&!;dq?>nr9gZ92%aNok zmyxhR$PZ33jK?KjObg`3EBu6Qy0`Ak3YlZL&QN{)O33J?2r}WQPn6G4#FXf)vMW{r0Aa=N<#z5{JOoS8w zt`pWx!zbemP4iFj!XpVlpL-)Wg?iD8Q;hxSjViR`?L88tY}~aooKGxILt~dOW`g4g zcXomYQn-0OTCRU1Kctf*(P;`~Gv(hL>SHekAHbyNt#0nRsRGX?Go=khl;~bLjwEK9 zp%94~P=)x2&y|x2X@J;**o~cn59hR6VLhw9U7`Lz&eC+OB!Wo&gTW>e@ zh>0|KJ@06L&kvx|d%|Y*x}_PUn#Z1{F*SmiOI~UTQ?C4}`$?zDs^fdC13j zFNxib+U*mU6lF~4yNtMfPXuWV(%Ngpz617`S+(E-zObW$M%7+g+D7}_uV1@yGJ0`< zr%^C@uSm?l3&v*&mk~c}7edHw(Wj$E6-z#({BvmcfEWL=Lx`A2?J|xnl}7Wt3S)Hy zeaRc^Bs9M@)}f?($O^yoQG`PEo68}NU&;Npm(w}L0mzJkkI4l-l*`q1tHyYjd=Bkl zu@y#I92^0}3RtyTfSZpbqjXmc^}GuZAvN@*KU(p_F(AHWJIlNaSo}0{TE;Wz{9bVC zdg6Z1UpX*srC=oR@ohKh>vF$}hMhNjW<2%|0S{;*k}eruzMuPgdpLfy#meWz&m%W( z42IjKi!_18)ygQ11;o|Fe8Am>zP|thk3|5>`gJPDFPc+rkUmk07}nx`4)tuqLXg=u4!wKOk2Tz`Zq==A2AF`OFjxOH zbkVaxso0gO^j8oaELsV9f)fZ3_+P};PaVQ52a2f1Uh(;9;v+5CFbBj7p8;AjL=SG^ zY$>M|e*Y~Ci;jlZp+r!&Y?8QS_&%QdIUV7|Os2}H-pI_IU)*Y1;Yg22OTJVz-CJY}iVr^~cin0~+ z3*pIbU>sQXu3>Y>zC=#eAdsk@zH|m zS#VMKDOvskR$fUwl;OeeOA~>6CSvzFM~o9Pc-Q?`?$6R%nG=;vJG+U?;}vDwk_6V0 z6f*30Gd7*$F$glC-Db6_L)A72Nm%*ntVcIglH=JQl+TWV{{m!qloP>LbTNRW7)MaR z%lUg_MFTHSisLp7}km!ret`VSj6EQ^d%59gie>|5+`dGteE#Nj>PVf}-ui^<}kpOH`gswB#Tc-e3d)pS}>30|5e~jwW^Tpi!Tq83uxZ;tB z)`!PnE?aNr;htVG5LcM!2Z=18$XvbarJtEW9RNozS zuT~(9|LN!(q*8$bX2ty?T^p;U1i8M3xkpM8^*lj-m+JW&qH1htMS3b_A{&i|@>ZQYGJ!_BrMmof ztrZ8qe^K0`B>_%Vg*|=EZNvp+{j9yU!O>sJ;olr~C(|l?p@{OR$eKRS=vULlO=w>O z;_UW2y~Ei<(%ve-(CfQ z;)Zz@R+Kfo;w81ic|OQojN&FjNA4BPtIT!s?Vl&(_0N+b6Ce?#)&5$zUB(aaqnhXQ)KRuj}mp4C@PhW`DYKmAYKjhTd8=Wd&BaQg`a!;4SV9N zSLZB!=x!fu8yo)R3vG+U&aJ>QYZ11v2-qOo?jwGMw}V#mm#m|vC+1K`cF+Q;80!#_ zZ42t8L#X6giXy8&KsiOFG%~iqe&NIb!~`?JsvQCI;CdvcD;?-a0Bxw^_4%A01UdZ> zcM2B7D0PoI2l<;SEqISkBmB!Jl(c6y+Ig%z`{e2wOC1ew{$g)@>{&s(NAq~=s_d>H zxxJssa~eMN(`Mf=$N!%*JN$I-&Qje~9EvlTEoc=X_OF=*f^6PqI_=IYxU(7-7b6OI zn(c@$D0Wg9m2gNm8D1r7a)90E1DyZpGGQy8+>V3cij$lcpH~LAz5tIday%;ot-8$N z!kvPaRCX&mQ1|M+keJKHzif*>FmdwjsYcD6)8?d4cCaf8c$_A5G~r)ZG=H9U^;6md*0g%i_P?$J(FJ8z&YQ;hM5;iUbw>b;p z=Ozo!Yifu=iIMIBe-Ghv;fk-~Yni7$|K@4}e97mTKbqF2=mKp}8}bBX2=4bE$(=?& zvPM){KuY07=Ay9O_=yw(yvqUk?>5))6q%^&v%rYyiEH z@+Q=pWA_T(9Iy!6Cpq_lcGl}|v;WSa;BK6hu|J#j2#w0j=x};o0?ktRhDM^f-udiI zJ7~sjNiTG>5XlMq8un+#`QP&r?W8wxH!&+}c1JB-`4uwWnsdX`UyRd#Yp>VqC9;0m z>QBx5%eCcZ9wT+Nt~T4+S>Yk;bh`U=(oB(H%a1?0q0kLrca9w<{^?f;yOMj9##px{UGfmmXXe6?OoRW5?h|79Mh z3~#zWEp>A&?#^*9t`_Wp_HP3nLmuPn*YmIw-Bg_CSI?>_(gky^6OEvEEcoNtry{{0 zcOiIO=LQ5>Bz4iK!aNwa43``K4e|(ep$~GnX!x}Q--Xm?BSN;EVN%G z@LrH%cWw&Ly{?{(Q-6rS82p_-GGNaT!c=%EmDLXQeNJ~;;cevYp(kaDV6)<`EiBIf}M;ClY!3Pt3Bcgaq|9U{SJ=< zc%W6?3XBf!Z!VsZdZcOGA>QM-sZI2TkmR|?DwLAvW*L&p4qcdiNvUec7MCo3x zAmu>9;HTi61h-Or!%w+mcPWo|LY#e~m!DD0@aQXbJ~AvJW|368dTiZH;a!@#h1CS( zod3`F!F^m)z70BBoMajujGlnfO1Pc!FCj3Ko#Z<7*%6q9NRN)=HQXf*4VK5X5A^@XaQkK z;qPc{nhxR_!Gl5m{`OsTKzRAo`mOg|F>Ub9X(_Z5XL*>!y!L?hwyDcrinXIt=fHgS zZh5%5k0grK9o9J8fjF2;ojxAghju_-7YfGmt?5$g_wc8CiO{{ack?20-21b?9*~1? z*01#yrq?f1KJOu3;pO+))PZzkKlRH)q)*Hy;XPK1JUcqsi9S;kV9n8*4Zu&Eo zrc&wg5Fj8&{P~K?efdrhe@|{4H}X0xan{xBXv6j^l*Rq>6C?bWNv-+#}b$jrwd2cI$to=*EnNsFX>S+kSTB%f(? z$Q}j&yZuFYJYY~y?_pC-?cIU_H1^D%V;O=u%?5NQPNtj&Ax?7h^Nb^R5w-0Vw2A&> z9lQstzug+{_Elq_{YN$5VmD{Ed*ExV|MU06UfK2Gff4+G%KD*}-+eS^ zZiJ?q5Qz+E!N4EMwt0UhnDDl>aJ2qQKSkS{D^+U;THo*<1!uq@a@mBLEm80G#ifil z^ejXE5+RbX^YLz^Pdl@W$kfg6;H#n|;eY%LRr(#eBqeOXPvBRH$)c|6Z`4{JyB{Zn zeeQC#A+SC9N?&q%E%Oa2tlNf?NH1Hh|L9N%5jUROf7iBhusiB=+cTZxPXgNzcWhy( zKEpRie~Cu92C*t@Cp8n3dE~B=R0^a^yld8_&kRZ;CFe!~TR= zHHJnZR;ZcFXdQ+vbz`?+;MG@vr$ekqI3RLG%as>|^- zG?xsq zFmC%fZMAE?;}r_mA`GSsy~vRy0>J_f*i>iUIx3f9ZNK#jWYJ%)oz)0;#-B*VvXy-j zk0-q&AC)azjuKbL#~59fE_rnNJ%3n+N2ENm79GtjqOC z57&@DD{JMSpPzYWK0Vvb2Z4JJ%1zwUck&Pb3UWkjr`im4JY6oQ(Z?F>?`~9ro5Cph zX?GxKz3zd+8;uk$lJHmw7C-VBm2YcUUSLlI|tDmKHp2&g>}v($!c;PL%>!>nb`Ubm2>PD z=$1{dzZJ5Rf1I08PLvf+*U*5N^-R#&?$KTGyJPtgxqX9f+48Ys!|^M1d#bA%f%!5} zMC|M=4R8bebnEo%8pkAcuRI92>^xcZZ}zyo<`_w>hBWqmMtd%7so1a!qR>;}rii}X zhE58ba^d$+fTm8}=h6JdU9l-Q&HR13=23acZQn$_H}B;+`iWSduhWUMnBx@{-T5Tk zH^=-;nmQYNLnBu;RE4}jwZExjA|T1_6Fna4!OknT;fyH zOq^%d!R1!QJX6za4JmKmoG)QNyvz4;eHE=J4Mp`YL&9u;iIN`J&vN_MB_;T7#5?YA z#Rh?a>pb6ncWQrRNX{5uJ_^5($@+a#!Qu=k3!m!iSZ}owme(LX8CI1`+S&N#lX|3I zL8-i7e17}F5&7OIp!7mm9_$43eI)pw?U@}nHaRW$Y^X@ZXCtDZ1fy$xUpX*mht}R? z9@*`&$L)U1EO@9&D}NtcRQx7Oxp>X@`z7bibC~=iAN%;U`aUmS`RxUN|IrW`mRav! z`?|4AyFJmeEM^15VL(21fv#(}4Yuk5bJg`^11uk`s?a8W_XHN{t%FJi@*R(%I4vp* zuE02=kWg{`JK9usxp1x`2|iK*qYui?-Zi69kDf&-&jPJ6eF!%n`PxrJUn(2IcJ;YK zKyp78>Y;hE*D-+wa*8vLxOu|6^$zpdq-^qu_dP%r8=;TG_i0*7T|mh-szAzlDX)?0 zVhB{=3oen-Ve#vqy(*A%@iG1E7rUi)gE7P3vViMZL{uV@fPKbb=twI0<^gu61MtKK zA|?+bRq77RMOh06AR2844vM`%qm(SiC2F5G=YIobgdQn(>K#R!h%6Ktwz1L5+x zC%2NpxG2+;xhM5~`S{t?FBFkvn;2h8Hk8$m*fHbU3yAC&2fH}$=kuEG6#sOS+q(`i z`Z~ArEfji(E_Pkax_wqmOlm1ia{eE^W0yOxpU3qJCm;&MH|zP;ZjiDD%Wm|B;hOKE zY!nrhyjCHcLCdb{_;fRrQ+qct8JVpZka0q&cqXz!;@TNO8odlnoG%iomT#aqcHtoT zhvacMyc+!mMmInq6}puchq$KM`t&Jl9uyjGc#S(@Cp9Xon+k3>7cq1?JHO>8Gh6Is zC+BKgurro{!t;6+fLI-%^asUTeda%(;PM7cWOuQ3=5pO(N|k3VG+9_~w_7d1>StQ+ zxs%sycFWzFe8lM8x|#|U0^|eV!cb#pi@N&G{N*fuJ9)sLrRI_uG%T!Ph2K4<+;(&> z`~iacT8_9PaY&*+7CXUy3RX);$j%q^rpu_|_}&3^$q#8?q5Kfp_o2`Se%!Tv-aV+W zrrJ;Ek|szK!ZtZv)7}(>*R#1y+qcf5@nNNN3Z+l9!NlkIS|a8fNa5qoZP5Su8IRWS zI~)GfYPI)H)pqGcJw!1D>9XgiJyj*5c+!fodgGq?6v-d(E^B)Zd1aO~{i6LEnk4Yt zYM##e+ko{mJA&mcpc^*)egapuN&p|dazbGmcan1hOYe1Ezw4QfPM*`8XkjrMnvNbb zV?ytEq?<^%j{(ru)fCU~z~`ETQo{=U(X*OYHs>!XO0m9zB%-?CzPtSHNa3c1Pe-q9mgO_J9*Sn$lHZ`7w?Oy% z`HWYGpLqjZ;mISU_}?u2V+H>17o$I6>sy@r=j0}CY9pjwQ5T@ZMELyT6C-QAK(7(z zFrD(WJH_na({kq1%Rea-gU%K0{W;uTi4LeeX26X}wT;=HmnQ^jX2ITfcAF~WPC;!%o0J|2{2z&WwJAfu1AiJlyqZr#%qJ3x?} zsXL8IF%mT({IP== zOldMtF#G6{fNo~r8thNYHoDdnSlS$j7+qHviTL7xvb{O35_j_g9Vz=h8a5S%@^WbO z6NPH8S$2u36TT8y{L1N!YFg}7A>(<=2Xk@}SRuy(dz*yMDR-@%y{jGTwUxHt22LTG zwPky9q?FlsQ8D(~-2OOnQn|a{Sty^u=Bn?ogwIu0q1%4T`LAeuq3}kV@BKUIudmj- zySpEM4s5n7N8nHZ-(fjnp^?3oiOYaBZcn4TH==v8ooQ6uYsptiX$}k$2`sfjhMsM9 zI8#W#Ci2&1-R@{qOUNm7pJ&hS>#C)mghH^A1f@|xy14(G&gyxaZkOJfm|OqR5A;Jd zdQ>O^ef!;(wK@~j{oJXdR(fxwfv+xO@+d7M2eyzrcmr-rtNZYoT%GM%U-J8DuR=8^ z5!Rc%H$j*M1wYJNgKMVwz4MjPN6WFd)k9k(JGR~@-{F4X*a=*}%^iF=|J-Lwik&!( z%Xl8S<+0#)IV^k7Wv&u&@-K1*_syw48pL81R7Jh4gVw+}DE7t_A3EgtOMrMkpKXj> z)A@;I@Qci<-7t=?X791ya2=2Dm+6jIk=!ZWN+e+on>SvSX$HfsZup(5ozC>d#+mI$ z()POX9-cOwae(ghx3C4Z}RQ+Z{(LXmIzfqs2qyu6OWs_7~$=?ZO|QF8|uX{&%kY@;^Us3G^GQ zqGu#SFgGo%RQiIi#v!M#>V*XyTt%SX8vW2mNWP{ z^$Xk&sA%uYtaww*Zjo5t$)?KT1W6|~!QflIP>8;Ug`(XLCxD|tI=~!L#5dn|n=T2h z+yRd>eUcvPD!gijqK2sn84Yr+yVJSlhA}s>bEEzHdyS~)w$MQPxS%ZSlQPL@9d!P}<(?9G@lCbp{#*@1ufF|MSVY%oiT#)0f*UCUp#RgrOdO z^HdM;dRVFDRRDD6_h9En?N6O8_P=N0!}~Wb*}lK!CRx5!dcqx5f21$VWexai zhQXEGLp81Ysy6X2&k#>MrO4?X zrPKh5a39Y!JwWWye(?^|8rBRAgqNAP8cgRy&O1nB0*LgXE+~)(Fvg;lu&86s4^{b$ zIyy$p0dabk6A9lKB=lA91CZ`_X!GEl7h87RZbEQc6$&bjp3qgPVb9O?!IW;retpu` z-g>iS?`)fG0XM&@71LCTw?6u6(_rr0o>s?1A`S$bo_^78*o+0XEwv6@l%cEQI@fLpK-f!+O%Kq+vr1zy?dtA5< z-zKG2FWT2HEpn?{6Y9WJhsH(8k_f2J?hTBuhv@oh&GJonK0{1-dk)$Ontq~cOv$~i zCQxhCjVNrt{nt81F0W%pvK5@f2TlL$0ThE7=$e74Z=-AK0h;SCFNIy@sXC}$w%1sC z&qb)GTm3x%nyNLT^3yLHQx6}YReX=zc@wiy%iR~#X}~jCuajv<$M32t98b1Xe7JmL zoQ;AdLEnL{G-Fm9-&aB~$X0oo<>oWk>^4OR5w^o9(axoBT(nC(P~rjag_Iu#RF_Un zg?$w%pGgADx)HrV=!vh!?r6K|3rlRUoX5?V9Z%6HbVgQ%6*J!BEvE`K*q8u1%$Z=; zt<`aF7=I1b48*8CjG_#C=*fQ0g#Tra^|NV5c^9vU9J*)x4LiICWm&H6p(b4N78}le zs3+yeQD!{ja5+t2qS=4kW?$359%_yEbj)Z|Iec&S*Zv-Y_BgX^k2!54+{qY3#Nk}= z%Ih2Ik(d~ldTKVCp&uSY^l-!Th7>pj6S!-{-I0;DcD^Y6;VjOwy(P17hjxTv(u$Pr zcUu%=6yGGi;F%Z}eL6fCRb{2K+{;@}3E1>G9qRpe;vU-RUVDFMsgzFh=Y)&j>2z++ z1xChIL|xYfl_K+#e)=uP@sWP13lPGe$a%gax{Y?M^m(rHB^EA`9x2KX#w~;E>b$mU z{lK}rQpE%|x5$M2j?OfL-Z+I!diNpamE$-eK`DMgOAKO^a;{OTVp|W!V2as_i`cZs zj%Gn#&imY{rNl{fkr4{kSdjWT%6LLxQXs@&DeWLA)#;;BI)d+k>bSLGb{zQKkW)9h z=H)F6@Ql%Zz7}D;PvA9od@*gy+m1H3!{F0#HKW+lZ;wy7U=ecGgVVCl$`ue48M!KX ztdh>(ZKV)MU0ao&{{Fbbb;WVSh;-)29&3%ivZ9ivb2;|83|Ok?qyIqLqh zm5AtYtj_GCdIPA}jNJ> za6QQ9UzRk~J;DHA4bspRU#&g8%*->FqSwCmHn%!>IrF^?fw4J-pc)@2iLbuE5j(}V zyC%bTBbVR7KHeMHjT}ea=ubL@>3fgON*oV_3dvhqo?Xib7ZeNh@#u&YI_m%Z?A5=Y z{ZC6+k)=LaO?8P}n=zV39UmJ337zW@R`BLgxb-*irup5NcGOG%vGB)7AN%W2e+Tn; z1{4|8&+i?y_kfL_D7_l4o)%QE2?NL!Ak=d`4sV|Z-16{&2NEXRcZ{CJ?9#9V%W+GG z9W}gHBW52sGOonKU|}n#egWw&+QF!2)%WkfEO6SC_2($HawLE5h&kWNqaV4opfd4p zwEx?~l#t$;5KYG#*Z(4!3WuXGZ36Gnd5!c#rQHD-MqhJ0&(3iOS+CNNbdu*MHNP+;K#QU zrG;4^zBP7_iQwOv6Y98&MMz^R_Lq%@kJ9CNJg`0DBU>Ce^|)BvhTeS2=upz&-K8R? z|4?7&cbZ(c3Bw!L`63GMVD0+6-+CqLft?_7C|$O4S4vHMZCcM{8376X2MRNPEpj7q zyjXIRkWu^Dw}Sz#=z9EPkF;y*Wa;a)0dyrKezx6E~K8IYNoAZC(Y2Xi|sBWZLjy^4xY8M7NXD!faz5Du(CI{Xh^f*QOdB<^_BLT#O z6$byEZBVN5o$+_S?~K(6e5VA@0{DSoxv0*Q533|Z__(e_YRPKT4yjD@oXZl+kR~Y4Yr4Nr^UhuVmnw- z!5_`8Evqji@~iZy71RkHG!zCCV)aPP3iOtyDq2DLb!%Z4`(9I3-gB3_-QG5z04A;Y zXArjP@Cz`0pv-#qc=ATy%D^hjD110|Zg@Rzj_iP8u(LA|i?0#2@1A7}(7*|o$?~$w4}O${eIRyx?^d)cE+cA7imP-5Z)OMsh~c`^CNDre&NaG=m`CY5W0q=6l_iOjJFL4Km?@&UCT`{%5AQ$K!O(#4Ce7cYxvmS9$$l z2;s}^syw~o=Vwqf#LDEkhVw_gzUymcl>(jCpxU5>nOh zzQx_O0>y>l_5cwn81dm4s`_wxjc4e?EOOhNS0azSw?3aM(QR>8L2nK2Dvu;+7GJ5+ zuTced|rUz34v(D?(sLDyY0+vM$Z{k^~Lp75Xb&{Ek z#u`=D<&Jw)n9-22P^dgThsTTk_q%+idUN9Ui{R$d+eYM=-i?rH7fxvcG(zFMC0$Un zJiK7k$@T&s@K^VNY2xzQx}@PSQdWFFeRr(F(C;F%^p$@2b-48BB7_{u3`RQwFnbC% zDI>Z1?OMJtYXp%rzbU5zaRV}1aU7e5nPB4 zE<3+D$6KrWkD~*>mUU|AYr#?_T0^|%^P14v?x0iD}?6P8GjN(VKKs zeCh}avjW+z=Je2l?UbT zT^0gma1t*L5$ngC-=N|eRiK~HIDV;vhK zRI}NdYFM0T%R)AYiHhWDY%yM4Oz<*yHwS8_)`Q-v-S;|tn#J0OR{ltbc;_Dq+ z;*OzpOoi8*ff?p{~FcwypPlodJp z*k&ayGd!YpPmZmPFh>p3GNWD5Auti{7Up*o4|x<1PX#DN+`;T+qQ2aMqHbO_JCZ}Y z@+UxyMn^V<>mK%4*z``J*Z$Iphy>Bmy$jXgN?i~RFug>)>TLd-`7SS|QHb@CR$CBk z_5_0;tXYQ6f6ojl@KUsdTI2X%Jom1Th2vhjmfK<-Dw-K4VSbA6l1H)&R%MFo&U^7_ zQ${I|vzUK(wD3i501I&eF{oSkr%MpF-2(n;drV(YHy%0iMECsp`ab287l%D!#|-5? zBl<(V+8-wg>FvneqGRI8+`fmZ$w^x3jXA>itO>UE2gz;p(zN4q!R;@A3otd&R!*>&-}lN%MqM4Aan^i&)TzZ~LrxuZp~;)y+vu=;DB6uH(Hx>= zp0z&1``-L8m^w}f31D&s8>#GsgA!pBWcC=+o-^WJ_;nVIwevGDfmPro!3DKO7>)E| zXHa|u`HKZ^<#~P~`cdUzMtuyds4xokyO zZRD+})6%)W9W1m0dV(yozv?FVEJl=7i}-p3IjWs0{dL#BT2`g6(BkCX&#(HtLrypk zpS?!MgC^pxy6}1|U&!11gDHkz>bsICgW?*2V3wza6;Rm+iSZZ!U5MX6DII_q63pii zctFz~?vq}@KkzcZiRX44E7x0lW|j-|el&trAC+`%2RyV6mv-c3zLzb&laCZ2rC%4* zqZ;_-?t9X*b5!&aSyI{%4yx`y>vAw=>P*;8!%A2^-snO9ppVAXiL#r!U2-~?kw!4b zFQkMBE0^3kTyMp1sK@)ENm?mL$84YMt>mF>nDGeltRfy>pVHZ3hfL`Ej7Tv+RrkKM z-W!&@oHwUCogB9Zj)e34wkq#RUB8N@-I1B3A3xjM;PtlH@-xooGdHZTSEJ_D9Y_;N z)j~ocL-9m^8dcoR+v<-Wd3$bAF&X*s4{C$R7K{P*o4i%!O>Okx;7XnSR<>E-Xn+~l z`cp;kOG8?Pb9Lp1?*VIw3xOghGq~<3R>Pm`J)XXXM-YM2z1F|ah56K$(vE=5Bc=(< zeviA3T(I1*LCE0DV6j*F4D%I*F-6@f=z+49E6a|ElLFU*k^*L9QG5P`e**nJy06$t zc>Qbtem^dO@6jWq*EbDvd6iWf2=FKv#XKy4lEAFy-;`V{Dd#L>?~ z6O9Y_Xu5vdNSg+y9$1lAP{?m?HJ|R^dwD&=+qsvI#91yih>+w-h)TzQHGv0UKTC5)ZgSbn4{maM@^l@-3HhhEf5tHgALL7%Fp22Yq`zZ;Oe5 zO)d?5S|=O2tz58h9&x)#7UIHN zNBA)6C;jUP_!>JS(=Ox{;v{^ET7GGwAzy3yf#(u#h-(IBjw}Szv2k~_>hR%(_QcD3 zTa94Nz0wtJhnY2Z;ThT*2Lt{es#5PzNqPGMulhCdZi7^8+_CVCXGZ*>ZxBzI1Z?5=d{+>EB1gOfeU zt^2EE{HjIGoz5@>DR|*+{d+YAwssx(79y12E1=n-U?9jwqNzn(b*h#Vp87g13PQS zwV~?FXCwC%z+>T~AMF}~Wbo82=!sD=@d(KEesX$tKM~0Xyv%qKM+zAtP~boU=E`O7 zL%Tp^-|asaG=tLL1-?qF+$|cR2jsINtK%2BAzI4|*V(JWU9mirpAOH*`4sb?k=K-N z#6tMMzKyHtYkI==0GyqL^mDb~k*Tl0I}x&wrdl6Gk(yPr4Udy^t4%_bAJBtnHxtP- znD;jk6JBb3?WxN8UTn%4f9o-SJZkM1R*P$PNYdlab7UJ3n)ltoMVHMHX~$Pd#kHK8 zs#>Dpxoyj#*Lset8pRO8Km4VH|vf zb1(=JA?OHsi}~&GK)VZqid=h(7+pd>N97^RxRQii3r#$8{5!4k)doM05A{&C=neIJ zia+b*q!~{)VqjD`TGm@WpKJ5eqnt;nZ@Gi-0B4-m9!{7rPK1bGF}vqjlHVBiNKX8D zI#kx8p#<)h?wW={tvD;zzkgR-)}Ym!3cB#{K&91D(A#qWa>nU_nxN+(Ks`3+ThZMB z(@4!(DRM_!2r+bzy`Wx?J@BmEw6h--50mK14Yqy^ z!vIb*t&K-}uXP8KdAD6tQF(*z2~A=RHhiYyc;Pmf(0=F5-!oL=4omk5@1zyE$<=Kw zg7hy^(!{5fsXSJ_PAeA0eP;|mh{P7q8qF5I*6V{BUv=zWTs4=J&k#HiPggs~(XjWX zS(MxeqH90MxjF!ic3bOIVGZQv4?kb_|NlE*SL+&X^WmzQI8$zvbeZq56PkBAm`@j) z5f=)zH}k48$qhw-@n+5lr~zGz{$yoR^ZZMe>?M5+1y~g@iC|NEINa{GM7vn0mDbU) z@UvN*NZ4Qa+j@-Rd<`_aS<(Ow$^a3d=~p1exqqocT3J|*ITA- z0WNThVir&zju@*_qh}0#E4kdK?eE`Gq|Ny|`oKtz#(AdFB>OSB0_Myql>d7G(J{U0 zeYhlcN_j@T z;!%YMC$v25L+fdNVUoaoRQ;5w$v889ZXqs(SC($Ab)}knPLz1YCj|9!bG!JsW;Nwz zWd%*UzotMImQz{Q;VpR*LdF7fNB1OE9os@q*HEBe%+rMW2=9#zNYD3n#^Vw9%b+40T5a z9A8)eekYj?w2?myXCO-^w-t(af&I$GON8CdDcw$*NfT`GKR z`19;`A5;||aTj;K7u3ly;In-^u1zxRaNVnToY9l>=Z!W{@%fuS%-$$seL20`L|4(= zss;M)2zA*y{?3!;jsD(&ufYk9KQI;1=3d_YyjLY9Qa^UTjS^cQ&-2>H?`G$8@vwaQ91{ux?F_{+mJ!~reA9HWg^SYL0iG_zQ_(vpMlQIaH zo!9}k0ji;>j-)L%>NXLX85xzCm04w7M7;aHcdr_X z=bXLQUJYTKi}bXJYI)0`#?S7jCzAS>^ZRl2^Lf59*||G|*S;&>ul>vi)HILY>DbTR z#&DNcPrEdbx)1IXsO#EJ_sB|YFWz>S$lxw*4<>A?9`m;Z!~5F_StBhx-jfY=x)ig& z)j`+%6()Vos~r&}u|l|TzPwT8y(Bj4lNX!D7Rr_4U||DNe}3NwPgzc{WP(g|yT-i( zjXWIG7mw_<)?f2Aei8Ws9OFlFI-T!WjzZ5ZUz^X0*0<*+v(~Zkzd13vytYUp{`l!Obd=`t6x@;sj~#R6CM$$-m;~>X z3Ru#-~?mw{sJJ3nr4SoK4p4|#G1$Z<>Hde51i%s`8~mV3d+>zRVMGkkBtl7&E?$>4beLiEu)z0(mHq|?h?y#303m>-N>Bg)1 zVD)QXeWsH`xXIDH<#FLz8!#Oiu^Trla^{PDmf)QH+&3@W5&d;MUC#cj=fa+QS}%(a zR`_!Jq>JRE$0jlq;_r)>1k9!Ck&mjID1%-t%e?yxzucFpNneW`dDwGJeN$9aUd}=+ z?+<>>*vPu6la`-7+H%+xQoi%dZ0&n3)kH*C>f_#U*&e;%htRAztgM$t!JS;8s9)Fo zBFnR%+Xy#2j6l`!C_+x6wcXXeP_6UxbuSx;V{B4Mi|IOUh%r!jBus+OS zsIKNua!BkO2qBcRdOVREx|XsL_Fk)Lm7QkaYa8Xk&-;i2zC;sB7B<&clU z-2i&YnD5=}zYEuxAA3Q)pXwuH_aIlR+gNlE{8lw(C>Z_cnQO;>1Q-0C{W^0t zm%2cZW8IMM)27`LE|vj>_{b^Tln0wUE&&ycdyop(VqtLw`L# zzpoGIee-=G+|JkJI_~DPSORZ|@RIH6(ij-$D_~Ck*h@Dyv$X3%Mr#eXEJ9Ys6Z~g0ImXv3NP)XiopVgQ`T8=Zp#H0yGcVecJ6nJLVoc<$}{{4A{a} zwA(d==WN#ZZ=3U>_`o~GZ5DQo&n^}_qo3R0J*zr|@6nohW zz>YUBy$L(Zn=$1t{9Ss%AGWHkC>Jo)`+g)N0-bI(I7IJ^+uoUYFY}di3ZBPjKVu_k zGSuR;2TqTEBW*ui1s_f-=n;2}I&}fM%Ou=ga3ln}mAy$Ez`*KUz5A; zM7Nt^M2`m%NK@cCR^XasEit1&00~q_+9YHfFC)0D z5-^by+B-sK;k)bHL$(#LHj~}!z^hdE-Tnaou66*xuHMNVie%UvcDrI7UlU$F-+9YO zsvWPv%oBMKupk}=!@JPIc_T||Da{$8p>HCRmb`v4u;uZK+aVte4FP{IZ63zcD8=L6 zSG=)I9|bJh9lFOk`Z^>f2={cTh>}&lVhQjB_7T zyh9i!m@cV-@KM~>J$r&h%Z2eIPl3hTC;V0`yJU&YlRKs@^BOmCgu3@w3Y`Wx8%N9Y zrVY#zypf;leTlq*1=nnuY0o3THb^7nBoW(ZWg@oWR~?(|)YWrh5YN@z@&ck-@X)T- z?0;ik2q*l#*IW73+8~P~CoV{!5wFT4_gjfF7mvqlOxCT%VcF#H<$Ka}S?3|Lpaz zjX-!!`g!O1xlTR>$JsL{m?)KdQY$0cvmf>0E_s}l{$;Fh{nT6u1M}P^`hg~1Zn-1$ z=0DD&i;j2NtvB$e&7YA#zEEK>U{gR^y8qKzt>551Mqu`7wZ6QSnS@M}!NzngM1B;& z=l1lo-h!Lm+*`od?s%r&s}1VhYVr~!RUWQzZLVbY6#uDT{G6UOtX#41-}-VFEac}w zV9p3Du9x#-em;;y^1JThcBZJUpc{5i`t0X}x+ZWg@O@6P5FE3ALI?T1KW|&WGYuFN zxG%C@pQEl?WeW=FksD}V!=Jeh1sRRw=cUzuyT5{de8G z{~%8L1~oaEUER@y{OGLWC_8eC<}JUnxrW{Usm|ZUpkE{%4DKm7U^|vPE)nT6j5Itr zaGzclZSq)wqx4M7Hr{@&Rr+)cK-~n@Y7IABh)+xg*NXs;W?%hY6U9DW zmG%Si6j2P1`un(q;G=y81@~BQ)RA2l-tX_|p214@ad4#1RTf+i!^U&SUym{CN9a&$ z{?G0?S=`1bjwsfib|Ci2XF~w4YX*LGUL0PppSV9(PprL9uoQa08eEFDbAH^})=v4= z1}AI(>-t=_Eo{86hM(y8&JXetb#}Z8#PB)Krj5^JWu~Ih*Uf7pGA0}K zoExLBlWhBT(J=bvZm(ap9|tOPx z$GSP8K1~UX5~>0{E|z-wr?rDWl%K}K(fI^7)IgzA)(n>}aI5byp8turExD&q?Kahk zI4tGW)skU_NY}gPeGwxc?X$-1_+uYBhiA-eW{1%2HPKh&S@vc4wj%jscOAExXkM5L zxU@gdhcyp9L8DegmF-pY+tE3%Ts8Ye($Zr{`qLK({{QQyUBVe^BKXjNHuyKuh#X528R(ZZ zx<*<5*j85jYW5+=`Z~)C5A(@`tu)qC+ABQ1M_vaV?vYn_%$R7NzCVy~t4YJ6{PC9$ z{BjTbbth&W{|?*C?NB4zmrghUO@Ru6s=*xh`HB@5<6BIOi6qNzF(7#RnTrTZ)jaBO zu|6l1hRg*?SurlV9qMm&OfT?(fldl$ixwZk(s?*OeEGO;viEVbe_F`gkdD|NUsKIN zKMw-|+&72gl9$5zcr5ntuo>fw$|K}wIvZW?f^(W(;YRcICx&hSA#L!RLUKlWBd(w8 zmKdp0*Rowky=2=&4Btud_-6H*w)68tpTFk_;#FYsOOnIEj3~*5nfy7`&zmLNrwt_( zl(jT69U9z!_!`6qGHytcp zl1xPIyvedf+056w8rsWsOyEeNj71JJ7ia|cF*u<5dGYsaaa>31Ae|dD zujmAlwN~xB2ft2uUKNQ7M)H$T8-2X}_})Nrd$VGCnP;0jYkEXfJ-o=#%g3R*-3GCNEmHj<-U+G>yQp)$}zib5sxs&=_mkUkP9AP3i zo+B4gm?TUQI`aPW&*^3?968-8;)SjZ_Ky-Z-CcvE2f6P@nOizP*D|of3t(6js`D>S z`GrLRlXbfI*E&X|h4Kq4+}tzg_r9WE+~4QaO}YO*Tk*87iI!wfHC%ucVjXm(r12LiT&sBYfb@fWb^yX+a(=5;eO)cJ68Tvo&8AK zpYtH#eurN(Z#-KQAsFxnewS$D;df10?3!OVVRA~k!@<@^#cP@42HeVNgdL*44lf`5 z#))F~$|~R4s~4ZEAGk*`QtTLB0V@@wZCSp)_82)PY5Lg_MlRG)Z)-y&@Vo?3%DKY~ zlA__+hy$k@KRb%n(tEOI$%QP`eBZ4D7ELii>Qt4bY?HgTnFX76OvIm~LW2JlJwj`@ zLH|Qg`aBC;F5iL-;^)0O^O~>&C)d#ttxFh}wd=dj z2efhRItiFI%?WFm$$@%qZsPsfb+RgO(=4vAX%t1vH1_chcjA4A-gIO|826InQvyjq zxAW?DK*Hy3bM2up$Iy%9m$Z+lnJMG8$c+vo?e#j*b9!YQOYk&!aSeN~Xcs`p^1PIf zRg~gb>(3inTVLx#GK-G2>6Rx@2w_^Xrk=sFc*!}<@yJ-=dM80|Ir}vD79a+EAdf2bLI(Ylf1D=!Ih#B)q(BlScb6i9Kso%Hl#lXGtT7ZUo76GW=H^`BMKR4!) z`{yDwR&ng;mX3|m8y4^<)V+LpMyzfLqtnePyc$u@Ix_O)67ErZc6pOkIAa};f-){1 zyEnt^VYMb>L=wzJy?$(l_7nqT0Uz&Z!VN=Z5xlx>>amYYe`8)C41KHJ=UA_V{*BmA zsE8nhzPCtmKFXuAJliIgdB`CigqmxI(@MnHI$M|CBU5}BhG+n)Xl(3ybXq=7;0+@H zfp2L4KnyVto-0P_B#`*Hy)QzxOC5M$oLz7&cn)(6Y)uV8^nDE6;w=RZzOzxp+2+te zLiwP|PY(v3052KuHO9QmZ!Y}g`Lv7No3u^B^>~HD2wodLdG9x`n6-b_V`)00=Z{E2 zbzMeMz$XZBqFxZym4Tg&0qXx-pX7pAW_bwaTLA5b`0ynq&{!R0#KU z5ifD;M?W~dYWoyx^yT60KH{*R9zoZ2xMnLg+r1DaEz>nTVz~@5!5zF(w5j#f^z!u) zY0XP^@6F{!A@v+4AiM9nfvesIJ>Tn=AjS*1GXu$XcUbsc$opn9Tp>W!*Nc!G(gm-l zDcF%y$?JP+U;-*45w8c#=l~x^6(AFjrqc?^J5nS0;XQ$dFnfQ-1)RKwJNwzqF5&m? z^WZ7d2CV?RD8br*NgGrAez~!4TY>MJ`kVvhXZGn;4BB+jA4l@u zT~-kaCOo1XhOvsxz|jr%mIs6)-nI*@VF!Di3=YcJcPZsuE~@fAL92ID0uo;~pv?4| ztVxNtiSx&`Ty;OvK6i(C-$uX}c)5_6y)NLBJv}IJ_ap*_9?U}`AS3&HR=jN{9(WZm z)QETH6tse#VitU%I;6`YbpT>pUU6$N%3Zb-u-Hr$#;JD&?p4o67|Kgud#7q|;NI9q_(I%G;>GY`}kL%HJ|x+zxi>H|ATY>zj5sWp`($Uii*}KBg)Yj;9)`* zdO1MUzT5rp|Kb0&$^YRG^W%^GfBs`C{>MN3;s0*$fA-sy#E<@$|Lc$cDW&n7zx$tp zq^)xQ^S8i_{XeDy@rOVB`V)`|{oxOIjtT$xyZ@84+puZLSDc1L8y2J$`j0pP&;RuO zXLt<#rkMXdC?Di8d~|&iYGoI;6qafBf<9{>S9yKSOeUUhn_${_uyyS?ZUOLS`r6Ovto>M{nlQp-f%zf2yzo zRE>VK6c&V!-^jXmQNXgW?#vAqTnfsZrFQi$BZRipeVYT&yye&mt2?uRRhHeK%<<`C zy*|_vkD%1<&;RW4C4Fum@_LO9()`>jZ@F_l)K#(+$?OPoC zRqig(kY8F;ZJH-+39<`b0SpECxmuOkY6lAHXJF(Pw`mWSwyzS)<4A-+Bivbp4aes( zvNW9a$L`&C7-P(j>w0-?VO23do?AJE@A&*%&_g!po$)hcv%_EukJafi0cXMEx(3kt z-CNh`<_fJ69^0Mt1g0d*r{!Yt*!1HP`A(sewVSrgTHCBjbF}Gd04GAbuyVe3--3TYOW}f$Ppf-gj>2qQ_rc%5eqp&CXV0gtmU{mN__ZuC=s) zG}}8ePy`~rF@emo(sqKY*N^9RP7aab4bJZ$9o94cel~gW-p*eZzMjOm%JTvfY|!`CC2moMpm+TM!z>dc5fA24g3b7OuHSX)_CYhWB&d|zQUFd zG+_M8iyU*sQ>LmV8?ZYLIKp@9^LTy!`@Hu>B!tg^?rHIZum%T~FK-w@t7@GSWYnm~ zVPF}CE2rCngHICT@QmisdnX6FpGu&2@Z&u8?&LCWFbnni3ag#@Gal7O+?1zDL$eSX zPL0`Sc%k4-@86_kyxyB>LW+@lnb$4SU-YAA6FCyJ;5!3A`}thprz<(r@7~5oIljE? zDM)-TIJ&64ucJgD~kML>)H&1l( z3dl%OP#cMl@0gd~AXQ_9(NYJfS^uefD!!Sc4+;CNs*%YYT%*w**VNvd3 z3cW7y09!N{*v)IzqSf3xfd3y}gX0Q(#~j5;%6sqFR4H*YLv7+QvX-{%`#q z4kj$>`Taw0?E2AQ#zj#=n1aGs9bo)x&U72BgVRPkBXUGd-$-$L=*jKI@xaB|poh{1 zt&&BOg0#2uC^EK6hi{-9iisP;3FN5x;XtnHf@k<(5dpXxw>75*#M-_wJx#CU=X6Of zCpfLl?AYX);ia{$v0ZO+GeSbWiqidt1Mv(p*j#E+VhsVgL+QQ4z{+movJZVnv0X6s zcON|M-_es(wOfaV>f9d-OX`^BvP#hxES^91R1iyI=|cTn($DR``yR9sMFhKJzgw+X z`=lyq3nxVvMUo8^`04My=c6pAo5#O!U%aP}4m3-QmGsMW6>~V}Tuw#yT4yyTZ!!pW3Vc(tFyye;tX!@_TW1Ys}?->~H^y>pz9l>cbF> zuNQ9pZMk30>sI+ZdAMmW$!wetN%K+wYr%%5YW$3R2fOdtyj}^lLb%>zFk~U0rJ)G`L5nW9aK$V+4EtnKYNrz(Mew1eB9<>0J%g!A-R{k@B*4 zm$?3`6;Gyd62OVc_uahe2Se8d+K+k&NZrP)c7@-)sK&fEuh(m^!EjBFPECFGQvN-g z*Q=zI>uv}9C~;fzYZx?}gLw=&Icw>OM0A zg+(ac9Qzf6sH6>XHy_P`kIZx#=Qjd{)xtdA1jkpF53j;r_UIj|$C+Z)jy!$JZ6gv% z5wu>~xQFqsxO$1)MRMaKjP)Q?S=A8IS*Ut;XZG1MS3Y=lU9~!3OEt`KLc7>cEd|s5 zV-q}%MWE#Um4&<5ekt{)X$8k1x)cj_{Tg?5GfH+y?MbNxMyj!=*-yRHshpzs?fHrvgCWOqU$-NOCRB$B1k#6 zaw^0tHXY%@3V{EJ(DkxbcnGQo_Yt|$lkFK1jBfwZWM;oct{o(R1HnPT&eJ0I3KLvdCBR-a<7o@)+SCQtP+!VjZm zig+yNI!t8lLB&y0P5C~|?zxC>KR=3)!^+H#4gTI{Rc_OF?p$#`Ey!kjSRPPk5*uab zdEY;Z4rOjqewJ|m(%SIfWWQP1YLEBwjMP=G!LTwGIbo<}dfr}_tHXD%IxNivl3me! zn3aUu4v*9FzhXbrCUxEmolT~~x55>Ugh_ScmyAD*ZQ=NCYWk-j(cZ|hNIns)`c z#saVv*8XAH{prvAg>&Gq?zyKZ@1M$Uk@?uj*YbTg{`bB9f4vVmuEh%=TF~l<9x^WO zJns+3rGMOKj9+c<+6xXku);?1MsW|qFCA2#u@BQYtMf8oz)$n(%@^~zRYkS@;GXSY zt!t=!&9H2Ku<$%oWaOXqCHq`fo!cnO_P+gG_;I{z88R!2S20lS_xzJL*;3y-y#U)2 zGov-M`Y+bXa%Lr>`Utlt4H zYdC@ocGPog(Ydo2uuP_M`=eJ_Jp&a7LWg42syMMe!)p9=gum*^($2sP5pt&1jl+|! z*n3CES)NM`AD$Ut^n?^UAk~8gc=57BujsaLLRW2!o0xOdeHyKAjoj=dr#?hytMZ4k z$i3HN~Ada1@=RC-&8n*bcen_OaEhpWBy6WCM zyw{6onME&+`MdUO(Ddk&lnNay@Pd-OgNAM1+zZO#I*)`2n}2Pc)A>5$>Fkf>XVWF4~Qf_wrzWLx@J*ZgFN-p)tIMV3E%6k#4?y#2~|}{Yh`Z zhBzxvn2T;a;VwV=sUjBpOToI( zDHF6_@ttYFu>ymW5SF&A%}K0(ZEBPrhFB{s+Z%)1=Zn<2o^WA4_2Vsx}9@7F={4!t|D<+0oRthb}n(_$dm1P%q(eRYl_E3qF^(G z$90vE9;Q-B()W~x(FV`rwI9tX-J43;DI#%)Ydq}bERpm^UL6HPq~Kg41y$29MgFOr zk6zKAUP3MC=`dRe1_6Vx?^&`;4}u8sExuzOZvTZvM07=s2rpUwZ)7|LEK@i9dbj)WP zX{OgMcJ7jL;>-68z=3l^JYwpm71N({*Rg?KilbF!^cGQe#?_1)uQj}s^q?fv+SUc5DyVjy04DTkVL;* zEE{}zX6x_%8m<Th}0b~r@30R;fsYz?&+ z7?k$&N9p&`$)yk3?EO*-K|OuPBbr;7aMALPQh46kI(2PraLQ&Sek7K*u4?MEc$^r3 zQC-IfcV`|^YaV`f#1T{U;N279ApTmjs4c0IMYR2+^8Htxy{vt8$c{O~fApO4hvLK8 zHh9JB_?$X86fbyA%-|KMf-R?JkI7p}FhSD)-a0Mk5iu5L@(x9zQ(|haZ~b}C`aTzG z@{THhQ@u~ zq8MuHhIGpIdf#6W-!SNMx*Ekkv>yF*ub-F^<+lb_g$95chPL;H-z_Q_{ZhrMb?(#0 z=*a_sQtB-s<3o!NDO<7E z?sk2d97|q0uYZdZ5Bu2F9<(R7$Pk|<-41+}Wkv-25iWeg&Ek-5loeTS(rX)Ga|lw~ znFPXgVA;qTXj&FfPTkK#a6K9s?GAUOlU`_#H*&NmE;V#Z$D@X9)8P3;_EPx^;i(^^ z5BCO-YK{qAob~VDjB4iUew4f#LbEg}s)xSW> z$#z|U6Hs9M=#})!yYHQM&yH;`Lw7EXP;1xGktaPQ_OX@|xM7ai8vK*UmEaOZYOaM% z(`8CD^T-YNhYF5iZa`oJghj_-^WI0~f ze|6^ak8!mnXq|$$KCj`aBt72mW5Hr)yj|sbt`qW0_E*oXzcknR$FCuWf30sFjO7zl z$pzv^7W_gBzBR<5ykeO1{Y-y8Gy0c4x1n>XvPH51oZKLm`8>*dA(6{5A&%*lgEh1X zqsNc4Ctkq8d)ZzBiqoxS@@q!*FMhqhFrP`^@U4d;@CF9Oc3iD~_N@QiW=KD-HwGaf zwf1^&uWT;aUzxA*n!-i8B(F$)QV74#a078 zc${-xxb*BdpU~|!0YCWjS5wlhy)EccS8B@)&p z{kP7AF?sYvRSWwRJ5!cV;EBEt)_=WMR&U!@wyA~42ZBj{tRLd^#+kVb3>ENxsK#a% z!zl56G&jkacAGDbXwnm#m?8WgaBi9*%fO_uaga;0dtQ=~(Il^ll&rrV_x^b5qCUa> zCJ1^?yJaOUH3+T;L$qLe(p)BjK3*CcCe}Ic7F7X;%5jWx`K<%#X#UauEP>`95G%^( z`a5_pyCd>uzi9H;}zKHOUk= z`CKApVwBAkvBB3u3iruk_?$o9>i`4Yb3hzy`Vt-blJ;}N+J>*2=~2}>6jkn_zZFDj zb#VNROrop{s_S2W&jITM zt>8BDDPL1D6PvRJ-<>lrL|B4@M>@RV9sVYOlhR-lz?V!4!jNadygN`=7=Ph34crN+ z(SAw6nz)+jiIfgli1j2Xn1^(Mw*?B#1M_K##%n%AK|v1OGIKPSwgrkq40ohIJ1b=h zG3f2Tdx^Xc*WOn6{J@p57w{O=RUZ?B;?Din?%R;~5RU2Pd^peCu*YDXm3YPk!#_+m zZ-Z8KxX1Z!`5b-UU1p3S9312%#j%Zr046`{H? z*21GSQf|ux8MpIbu7}4{f$7t)p5}Mo_-EV#^#|rDuhQj5RlTLqZtw>F_2+M|ec1VH zKESQ5a84`sy2qjZ??0dVi5Elal2peJ{8lr`&_2n$&)i&K>be4flvrZs=Kelw{>5{~ zzuw0xZ=s#*2gPe8W%#Hwtw~d|F`6JJdVsC@`fF(c%Z2e>Tp19y~DGAKHi5^x$9NgxUD@}AWtXqPU z>$uH((cXYF@!-+nzE-WlP3d2AX&T3{B1Xs21u9#{2xbWzSEPsnCfNBz!^P91uy@-p zGk%V@a$bHfl!qPQ#pPUku3$g>=5h4T7^3kZfFIc%gv~h)*#k0T8=S3_qh6SIva(gQ$C@q=~%9?{sGYzuROlG+uFUurzAa1L%{1_oL2M% ziiGPX$hZWxVQa}TA$&Q!t1x__6P>Zn{w2QDL;LW*9E~^P#QR-Q+FP>bZ|@rsJt*Ui zc#P`0H}9`cx1aF>;NkZ>T!@UVh+&K`V-P=8|*g3vBG4@Pm$;8>q#oQxmW>lFNaIG4>;>VRjtph-|blKZBbg z!Q5BvqsFZ8PEFQO46uW7;XjMKYy^B{oW`xytbxK>N9-xC@73^dYV>x20Ny^&_ZZ}T z+Qh1I8qrD49;4OsloKMFoX>*q!a*P%##;e*f$#6jZN~gz#KbskdCzow3z5+wA--qK zJN6un)A(ur{a|*td6ro%l$=8AMyHd2%WmtRZUVPwpWL=53uH}v8bUuD^V{lN>T_S$ zAvu0lAVlDI6U`a(^{C5=dGCY}igqb!zPApR%{n~B5ID=+n)bfE-2x+EBLJA5LiYXQ z7mG8x3FHTV=Qre=7q1Su9QkHGr)C_qK$xj0xn#sjG#~6012gS>Wpbd;0<`z`3#3HQ zumgV_mY+v)v@1$P0(7X>437J(f~z^uZ`kkp&zs?1yfgWrB3jxOoCiw$LBQD;l)QAu zWCN_{%g*$x1sh2&Y{N>C*b2cz6wj4^=R0JJc3v^8S)uk zxA|ouPAMAqEg$1;#UFtok9kc{w@2sAJx;yZuXxz@;(8%4JbkC=k@p!1bXo@FxY``R zDU=WE1Bh0O{ls$IK|5F+k-kCsf;}P8+scfMd9nO5_`V-dgt(0!;@%G!>n4a6^gU1v zRTwZxa7)u|8@?F{4Zco_gs?!Jf9A~fxm7XbUYq4#_V#fZRRhSCr+YdQ^oaRR4S`^dAevs*bm)aWM z#|3>fikzXY56w4_0o#h0*GtAi^hj^2SYUjp-Pxvq7vCYhvQ9dYJuM~g2)6iM!NOea zL2{u5=c67)hZ!14u&~CbY#(MWB&4P7UzD}WcO<@8da_WV2HFYwhV9lMU#SO3BazcOnt7PUs?ZChq8Ba?2Sk(5nGrjC z9rirOz#%#07ClmbXDrp=eTh^Zp_E|b@I2rYGQn3g1a-a^Ij9ws?s}RO<@pqOunJ+Q zQeFm3VH5ypKBns_;A|05!L@mfl|UB93+A`!kNGXJ6osJpq)SMNvS8@)!g|I1nR%fX zG(9^rfrq~JA^Y5Pv)X51q7UhNaF+GkmhCk}XYZ?=E8De$OmXlXLBbH+Ci}Qjc)(3}DEL!f%CF^RiBJLZ7T{V>Y_>-<3I5{Zp}zSl#0b=h|k z?48Z50z;$Q1GnKG2h2rQPwEM(q_oCN4S`|+cqjKhS$o`T3$>7d9}be*GT^Fn-f=H@ zK4Ln%hQ5=Ci9>G;4A|*zZ|g=9sl0d)b96wkMZXoFi%F1Z$RxVkA$yJwgFhBUkvmLK zT_5T5K;O08mn4?eW;g&=#nFH!Fly>ZXK$EDVV%=~AYh9x1o3f4t@L6BlI%&AlGcXG z@||`xiw=NPK$NpQrym5kTNw84Aq%+4C9tK<=V2MoLpB?50`DDEWalnm4~L*peHhZp zfZS{`NNZ8h#}*B&WDba_k`BkCD0q_onmX_S4he{_+G?71=l%74zPmItfC$ZdMI0A7 zzTDctT%)^VWXUq5_k6OXqd38%j*3ZPig4I?e#a0)Q4GCn7v|cEL=#X5tkN>XTo925 z;x+({#ns_LX!N+C7Q>rXsK5j$)+-TwjYGQpOt+S`1VA;}em>U;0o>hj2|_t+1S&5Q z1c#z;kIF`w=F-{+37)ViWxEnw+9f<+nj^jk%*_!LUeia?JdxQ&y`Z_R=z*HjZUCVS zSTLVKs&}gjj4lSH6&DL3tYfF<+swp!m8denH9La1Rv)MtQ;E+F_(M{-RfSL4@N5#h zn34AR%T$(Fmm9R7_+kV$*k@SzxS$6mBpTSJhDrH3KNXMBvuG4T|B!Lu6gHR_M9MnEHN-Bp?MMV>^>)k2qg9uLL_xKl*sY48iy&&(VD># z)LFx->tM5fa9rzH6YfC)GY|dvTpu_WyW5w{U_UXU;*m(=eWnwsPu+p?ZD|gn!+!(5 zLW6urlG!UFA)g~_;T{>qw(+V{4)jYvBe0(y6FWgMG~Mow59SNf`J}p09)k;fZ-Gkl zpq-}UJVTd?+jn-l*)3j%TrOBu($?9pOWtjwfX>k0eI#qfLH1kVDmC)uJ6N&onq}>sfh!y}z5>a|u*S!3 zNEVE47dI-6Ma|*Xhb4u8Bi5-oG1E15onOJlSC9h!JR?ehwj5N_<@dY?(k~GSPRDhQ zeM%6Zc1cUt%5$Hlrioq~S_f;2jQZ}4I?q)L0$-IzQ?^~fMM-1a5w%{+*pgltJV1T2 zugcS^hgI)D)$wJGnv$-isD1~LwSVB{O4{0pl5_>jnB5M*L|@DB+QCk>h)y)2pbWLVM>Qd_asRupP9j2?#y z0V!*II>=D^b_=HO_P>Vb_F?UrFLk~Cnt8;l&tKw3|2X@W_RKX#5G!HlXbllwqox#+ zre0N=ApWxn5M3&u)=LuQ+DbflXMH{ zKyWx#H>7zurbO%G!q+6IZre{^A5p*V?bEl8@@JwHto=YfVT0i=7FEvKTdXVpitsD3 z-uc@XAU_IhVa4-h9Pdd1{!ZcRgFD>>O9=0$m(Sp=gkcfzP#VofPQzCH{B z7%iF_XicP`r0c10?r;&;(bM?BYjTcofrnOG6*ga<%;}DGF&Gl0@)iSq^~UDI=&Q!Fw$KkT1*yr)vW*E z5!?eAEn`rW{fL(RshZ2F)d{+*?dp`qu-&po(F5P_f;0Nzy9CK{DDDTlR+KeO3&PAr zI^Ls41KgHBsMty*^PKH9&qFPC<8nLK25~!d=>3o}f0nj5UR5M^N_ls{iE)-sySri!afiVMLkN>j(vpQ3;#hIcvvZ~W zFgBQRE=BqPMheS{q8%z#4410pvZI|#*iQ9pd2cr|wvR8FXerTxR~Ie#$1qo{E!5%`@{GbMbspod9-6^orxZuFE? zuxWIssJL*E9>02{1c&@5x6ZMgS;mo6-|el2hr$~~9|n6VL3&jX5)AFVsPyn@iz|3^ zOz_It$&-$U!6_SHjC*R4nSF5*V(-pZr(b$23j7@Eu766GAWG)pj9g?O)F^R{K7bWZ zu^jeSY3*;|rU};rF>Ge|gbP0+8%Lm#w96;R1p6HdrSutC%TFJ?KRrmMFE&>Wry=)m zwC+gq;M47mf@JznuNCBHNfIw2d59&BOBq8OIQSU|NK=1?>1+SIz8}`&Mx@ZqsVV$} zkwcF)goNy10CCdJ%>vapdE6VgVem-nj^jUA%Rh8#u*EPr*i*28es0@%I=xTk?P&3V z1(E}#f55yuKy*f#I6pW6>48+6cWlGi!v@Xy0=)=BMB;$X3WBSMMU9ZzQb_tX?uf?` z@;dHO#E(s4&V}t)r}n~VN(1M@h2wBkQ}YbDG{C**JN%JZyYdUyp20UB`_nxLbzOTu zP+e^%d&bC(9npZD{MtYj=ATShm@Hb@0Y1xj_-bUKBr{*{CH|j*1v*#EM!r1?I%r5H zZg-vSUnwLE8Uy$upFhj}lC4nU-m%SY)?BPj;HGb6ly<;e94{Qohvx;s7FqS0usrM7 z(&pMgfxaI`i6CKvvli_gWVIEQ#T58@wuHw<5QctM0?rPsTgbn*b0{JMZavOqSomL| z6mA-Dq$5c;t@Kk$(UtpZEOFl#z8LEFb7`#Mi*bFi5=$bR>joaq9rw9YONi_Sk$1W^ z50PZCUx$5NeWIpqDN(9KCBMajJ!3CEf%DjF%JN$H?RMK?j=Hd++>4YmxV}gxf$d8< z@1aGtO%Bj6`btfkyIF(>h_+$iUs=yxT@W>tAF-**8O~5J_tLNxS(S--PB+SN<=Z4U zG%xUfNK(7HYD@t6W^UEAPf%NGX&a#K`>gJhk4}*!C7oYjKYs5GkhMOBzK8}on~0pp zSYBrYj`( z5Ykl^@8e1iM(BHYa(xT7$@)NF#rq=rxIFt|{AngN0g@^Vk~HVOU0HHS=U@g4(iXQl zmuIx7U0wz{2@h13rugAf%ypa@?9l6vxF)Z8uSdx9b~jtc@DZr|WKS7*UGh77Tsog7 z^Smh*4jp9}#K+pN>_Q`D1WE_&GdzM0V6E3|3sX-2G_M~aFg}{wzQTex(vrJa?V>8+ zYyr+lq|hRI+{-P%$QLHMMc3p<&q8^le?%WyptTt1$IBhx5+}|4M7RQfwRUwg%~G!V zxO$)FxEmYk0wCd=2_y~t!`yF}xvzX>h&AQVukzA1N3u754nPWdQ@O!r9x+aUe!3JXucn|TZS}eJbhnA>z$qRWaKD*myB&Bj4{#XV4Lqh zv*k{=1F>`L%x_Jo5v1iEd6u>7|BR^?Kw1+=3mKtyyYE(c&qm%h-!=!2*@rVt_XTE0w*>zu&;uYlf2e3 z(^JASj!2~?(IWpvoN~HUQ~|Q=UAteP$_Lx_l&*M!KwtS zb$&Seof;_((Y8HeA89GAQI?e~FJ1c5WiTq*#u&OF5g>sJ1-(X7vJH997d%IQ8HZa@ za7;YovplqHc|TxOL{gumn^vBL1`zOZjkxl1UI&sCq zecc#4$icC6kTw*U4zN}-L~oCb&uS6D@@27_{R&p_pVZYzI#|}k*OGAt-Z#B3Brt>b zV0SM-MX)ffj{*@Q_IQHdJ(~Vm5{~nCs;S;V?guZP_dDvxYzO$QZ`gy8ImL_WJ$S8d zC%?s9Esa^vFl>$bd>D2~zBw@Ul-C+n8<8Ei*UK!R9Y>6k5IBbQvj?&21P#Qe!t5Zj zTWI&J3n%7SOTMu|n0El^4|eU+5lM<(@Gj>A^NVXi`u_GNmgi5HZ{6u76>Q#5ILYGP zBX=;2d|GEs-g{uF+c&pSlgQ4CA=5HwKI^;I!mkkKl=VEj$~qS@xAzy{V}1=?p_gaR z<_;a~QKum*^RS<0q2t!iA9=m(6> zgH4xcb33#PI|b%SGtLK2JR`84!U>eWJN>|C_@b5M59U2Y0{Fu{JhZKZz~{=chKjqSJc{1gtNUScOk+@GR6Vl zIG*HQD**$Q*#f5PYE zS>yxGF9!`bclmeHsQeeaPX6(XW*NRT%>ADsJZQd;-5!&~@9+K-X!+dCrg0$n49bJkuqDy)|=7GH#FFmK5A5qc=G-`G|+5W?R^yGiPR)uzW4))n_2+U8c?ycc3bi!HQ=<6O(`uFbZbv9^BjTo+tl zr;VXc!t`q#tWJ`Mn!-KdSeQTb!)%#|=XXDO%$GCz$(j2 z1BSHl#()jaS{U$3us4=$(hKio7?2DY@WLwt2K=48H}B2+>ed^oA0>it0abM~h7;%f z=bZoi^LIX<+zIrX6=xU*AdZ*EC_;(3P9;We7Mv3m0MDTWm1>yznlTaYR;*yj&=T51 ztjDU`fDf3F*T}(WBVT@tr1YjcKOmr#f$xs1G?h-&A2xRf_}G)RW;n;W1Q+?FiFmhs zl?3vL#7T}t$G5dJz9X6%d?9I$yfZb|s~yykFw*s$-HeGXTpl$5@d76XG)WRel6Ggk zRvZFm6Xmr!@L5u3p7tmWzVJo@YyAc_I7TE-h31I;Z$bLy52n(d=ZuWx93IyKs^m%? zTMejLQ5(tV-?UtVe}p%Xps_qbt)AHE+YxrMVon6(JQvvFEFRco37OtB3^dC$0U7Gb zBNzdqC{;2EK+Aac|Uv)kz)TLUlw(u(K+YvjZEiaIMF8k@MY5uvO1 zY5+^^B1kx=+xw!QfcI|zL7K@V?%i1P#ySzjNuxzRH{42mox34F$AzF(30H6Fquq87;$gxv+MGZ(cT7qFY7S}xWPkC;SP z7@m`}3_nk+QzV;iMlY@Q=8TxsL|8)xt!=a=jSPUa>q`@?A(@myOi>wrLu_lDp(uHp zB8`6x%uK{SPP)>h3tJ@ubc!)U+&f3Vt3*4fA6pnI`7{G{*W7?6@B(}}%X1L{O#+Xq z4`or#dUifRHEYlvZbOn=%d`Dd|&%85W8VFcK?$KI9X95>(ZQ#NnKhe1sM$ zQ=||CEDcau&?(xe2TUV4TGc1DL4gu9`a;UJkU@lX5blwGI9#rR#Q+7i#_jslrePUA zEB($!U6UG)3zTRjkQ+TGrd{>`Jt-OtXkPcm^FroEC=04naHYd{0XqWfb0Fbm+d%SrVewjY&EZ$ z(>J4Z-#xMPYfk8F1==ZmcbKqrQZfZ|m<0sm1!=%hM4vt=cL zzOoe}1Oyo@ic*g)WOoX+c)0Y5Pe3F8=9Iu63dC?9mVlDC2ff}H#d&h$fT&;_{wGM< zs2!9eFdLj70EGM$F1{36giu4h+3|?lqPQ`94l?!6V zNTe>#mO-JM{ehGpMTyCzQqxVb38X1ci)*z6`W5JviW&q;*n2osEl0Ea9a$$0#QJqk z87d+ox&S$=6~x6cHM&W4D{~P;H(Fu!d~!?;Hh>)R+p_F2t07_$oN#c|i7bKOZ4@C< zBb-q{YHQdOsiJxppOJKO_#b1~01lOy=x=HUW{EUfl$avxi#g6$$^rb;g>4itu`NaB z`cOO4AlG09pX#Wrj3Q;Hb2&T2G4K^TWqoU}a}#+FI5Ung?8U30C2a&ngHfVPF*v@e zjj882hgrA6vvZx;j7=T^6^N6NhoMve5W*+$L$r!I6Z1e`kQz;~kGSg5}? zO$oMlgqXml?OXlxC=+*9iZE_@OnkNoLD|7xpaR!th2Zdu4FY~(RQ^03MNZ)RYq)3* z##vXURXvlY{SJ!BoOp8IXN0JQs2JMubHT z+QZj9v#U7!hHDZZ+B#FU{c!0aZxMFt^$^B_v(hKoJ*pMCKNJelQALA9y8wxO|n(29ZAS^(~3H zmN&p6IY=xhFjMIuq!DImbOUx3GYbt4<#;P8+Uy!}Y7jn6Hc&GJx`S%K(~B6Xet$IS zStvtRmisQY=Ma@=t#3J*2-p>QYrd( zDW8mD0G2{@ta;?1T1#Bf20o9>RhyvZ!J1+oLfF#B6U{iWD;>$@#*Canknqbc;Yq~d zPp3>3Mk8XsLRc!?bCBUkx#bl(J2f8}9=&O31Kg}iiAw>#Gnt%eM~_#*htpvj0GgMa zR~UneV%5#YC=7vIyOE4SpB=NOEn%mi;?T5EahAv|G%%bH@L<5ZTh%x=0fYhyfbHhw zOk6T3yf+}Fr9EgEI3~bY7*eN8TqkR31_u=Z@jXBiM|>8q2(58nkdG~0ACD&}Dh_>! z_(=lf=h%yWSqbm#vQY!5vowjT;>jKqHI!1TWia5Pd_#jJx$tDZiGA)sW_j4cJmoRT z=a_~gZ+VfOJY|SN7DhImOllCeE^*OJBX)S2oU>~e*d|76+kk32kx2~= z@VO=6auB=d+N^j$D7JgB!6mn_uWv?K6}>JK0uw_~@hw z+;0-Xi4nC0nXEz&ys{ZjaTL*PC~@2hX?x5L(P`(ZTSZDfj-EJzD;`mCi2p&lZ`L^uvl?gCJo2aygRAt!3P)tw|H-pW(7NDfPqN1Fm5)l~(sN5rBhtDzD z3?K^yvSX9HnuTS$YjPmfw7KzETf7|T4jVjzd7Ae@0f;VA?EUw(j} zSp>X~Kbt6Q)r9{O;UM&>H%597X2|i%GX$AZUVGMnzv$lAH5hQ>s*2i3jJiO9=QV0y zC2Jsz;{YmC$e-No7qnpPmcn_^qN_cWkRY}6P`$^joT0Zse$Ydm=65R;P-5U#F=x=x z+0|&)ob+JwO}MJDLY6yN8HS>n$zX9p{Y;FB(Hi!gq^y;5Y;oxS>AAaCpq^x?@hHux zuvzUAUrE}9jFnI>IWJZ-*s)~9aJcKH%*?m81`xDkWkX7tBVYl(_ZbmoF7j_xfrzHX zr;|g@<1K8bF@2bjJi#gw1{saS30*d$$)n;sUxu34>@3qYPjm@pflzjdsJSaLf5e9b zU5!T#dE_Hh(4No_U`WqG%-~^#xS?V>qt3@2LIbOqF(Hmd7JLd#WgGQ@Q^4s}1-fo5 z1FJ%Y?*Y4RzTeEs<*`obX&JiK8U4&jJfAj-1$IL=GgBs_;1X3!b$I=aY_}HDK@$!s zqHx)VRPs4;sTBsFrupy%y)N!}lRGO^PE1$#! z*dU9zS`>DU$^%HkLrz1?E&&Ouq!nMrWKAtzt`)u`TCpRK`~ z=Q++k0J?0bPkWsv;;+hx6R>6n1~er>kOziEQR-vr;iN>_H%^;uvx&KuS=KNQt7-@q z1EiO2Ia84wZ?BtK3w7gXSZR|WFN~T9r_DBnZOx~Pg{(A2ust{{=p^vIpkx=3i_8tf zvqU#XTQGM{of$(f7ks_?q9mJ@!9YjnPxtH`TzcnH(q}IZQb3hTcg0%NP;yJYU_fhnjbmFl+dyY!gWMn`6`g?1$q^n^miuXpQn6x4O_%G4dk5BB+CYk2`Z1{!BqTP!NXKsJCbSm%FkR!_VA6z zOV4!rDgEF0mOXp=p)0vD)_;TvxKjFVLA`YU_)FK^zy22=AA3E}O21bMqgQIIhCA~m z!+-iJHjzLTnJ-_mpR=ES`XY**Fny_g`ogdnJR2FWP+S+nts0o*>fLSJ`}9iFkJ;n{ z!7h{XyO@+W*Y+)W?BBHilKW3FV)vinKV(h*HH;M$u06~q_Q_?9Kb=qa|HSiO+8}>~ ziy-#D`^Ceoa`)|k*x$qMc}i^e<-;Ynj6Z+)#pAxeRB8WKlE$VQeBoAc_ceYQ*wtUS z9QpO{fB*X>x`fi_rAnz({o#w!_e*T4#Fa{XsU%!})=DK2h0;q5Lw+)q5>qWPY>DB1 zh~M}JZ@q8N8+Z{br9i~?WNvb1yrZOj)(vt+BeS#q-k?*afI-1#^a z_I?&z-Xj4?9wgdHqX3-5GHJC|y znZZ{tI{Ua3WPa)xIB&xQvxFC~UB+9niSgi$Q39XZV^U6~?-!T*KYVGYY4kf3b@%gC zbh^yZKHi}&HkG8R2F#>SUtu!t9>8?ln42KeZ9*ERL|WqGJR zD3O#v@XISei@57188Mj+Y^;Z_CwK&MM0(u6Dy?s*X%_HZ_*ff0j7ZpscVEvE;7E`k z8dmzvcd_)1n_k~-Rw8=*lmrUjNE>&*u`~b8H+H7Iy&eLuzje2>t{eXP^I>y8KJZ=l z58r6!?Hhc9Q|RwKZt!a#F0#dMUpMP}?c^#mfFU>W z>*0O3?D1cacydF`e_>ttap+e48sgQ>n7`KR&-B;*Z+_xYFMa*3J9+GSvRr>Y9xK-? zXz5p&_tiPv{~q6X{k`{)2`_e`{~AxfMS%HjU3)|MEv0YxvJs9KRvjKkqMpJnBmyjru=* zQA%!za)03+ef@x4VSlu~e)H91Ki$9mGtZvwPt7jR*)DEWn;YB3?_L$FdsC+Um#-II zb4!>iU^9Rw4!lArt~;o?M4>GOP?P;qMr;bu>k{4Zob$Uncm3^? z@vnhR-!%1C?*{W|z5jQhX+ztOVaXvljWCYDP@I(ZSqv-AMeR;Y(H=7adqC-46aV~S zHkNm@ae;c8c~oJWKq)J<9l{bqtPQZ0pz%Amfl-2mCcC$-ks1BnAERS8 z^Za8qgX#MFZ)~jFb#OcShjsAe;>|jEavcr-$8R+J?iKN$J+6pruP)|-)<>cHV^=>; zd(Y7B?JJcyL2iafR_dMO>&E2yEyx{fTTNk5x-%>hQ|J9cEurMOjkR?#t6j{J#3px9 zRSnYuTY;(`g;{_aeXzlzZT!5mNR~eNfwuRJ1D7-$5jo^2(rO6B%@v|bWd8i(js|rF|5J-fdRl6iOS>|vBaVU3?am> zL|3r2@91zt#L_3Hm&K=_ym$#~A}F$fU4 z6!{rG_Tox}y@M{OD8*uwa!0)oF(l?%|H<$l&3f&RvL0?Y#z$%p>;E6VvBkmFzGPf` z^O>jK&2OHrdttR2emY(M&&M3%w;!0! z%{Xo@KGWH^ue}>{%&&em7rXh`+n##n>ra-#>p%CD8(yjQ=H;gZ{BGs$Jh)n+f2{TX z;*kO04C?mkGlO_<61F}DPrvEc$7%0tkLc;-{rr=o<~nz8Tzxuq{KuR+>H`nA8#;dQ zt6!UBSSq!zd#K)Y`0KavOwXTA<4=*J`#O7fKb}t7;$u$QTaLXO*7G}GUy!&V+KORq%0B*#0E6;FNk+1)iFT;p+gsKe;9&t5>|O;H2~VG+bz3@P`i3|_hy z!Qr*Odaj?7!Cyf8fTx}W0^FG=NJCet8lOfC)*ZDk|9koAXP;pu z@Onf0hQphLY24`)_mgy0NT6JlRu==v-Kg>KoiP4F&C>mSzyN&v@Sb0s`*$)o>COFD z(8D{$gs3`*Tp&Y7P{ZBF5DG%neO*mVd}mF}0OZxD1jcYkr$RCc(peJmcNZ&hrQM6W zFWf6fq=jeXb}(-fZ(dXkXhlH={ppZFxQBjmHe*CAz@R9%WW37tUzx*6CyBYBF_n#Z*`_lU7pK!{XPrY&DnQuH5e%_(0 z{m&mlqpNED6`SYvr`Mz9oy`1ZKp*hZGlPBmhP+w4zyISMA2%KTaa($36dyEY|JTQy zvY&)C@MUR^feT|o&>FXP-6_U-B_CHDfexUos-u3B+(ED8s z<`=&#K|Ox%^>cTh>Hd4KMEb4w6d-Q~ z_1>?%Ysl{ZLTrEcrq4IcxXQ;Ib9bNV`+F~iU;S#i`sS-|d+M35KUpkq zRkxFmM!9z@r8kHBqf+Yr-A`}ga(>~(38^%dqC9yCAIztpzqqdNFW-2&&i@9+@wN<@ z*WU{F$5@TN_(S}4@h*KaAz3jbvEtuD58yWY@4o%#uK8k*-Mev{{Ty`WJCE2WcfI=r z4)zERcA^Ge<&!H9eRMy;jQOa>y}0!qK~L<&a9JVs;Hd`~n&G_~gEucOFr*obEK8NZ@sv9oj-Z`)c*YW?a$L6eEP{n=_hA)8pE4K zRLM(hJaLa*os5yHdx$^B;F2c+373$fOT7Q1ukmh-N83T@JObcR^ASWRgoS@DH(#L3 zsDHg9=>^DyfI~9D$S2LdfX(&bBz}-kLOOxol-@02BH{11k1zYp)zOlicY|Yo-x?4%tRet^XbZ!3!Z#3~% z9rSw8{bTxJHD9}MaZ*)2e9_(v?CZS!3=y8=j@Mq^D8#?~+-|w)&GYx4``)8a_j9*< z?MnBj7ayqq`ssBv{0}~4DlRhKSIxsO5gG4|U4ApPXKz0@vS*~bM@l|wDQ}t|9-p~S zM|h;rFTBZVu2lN+jYmp-`5rp^#~+PazxiIg?uLHv`NA_4eY!#a{M*>)l`wCA=z&NV zsrCQ(5SqMeGyn2){QHIq&)*yXYJdEV^<25w)|9;4j-nuD%`MLAv zc0AACe{MYQxh-D*^yM(S`NaK&XWD$aF#f_j&c7@5-uc`EQ7_BmSJ@9xBHy(biN1pUX0N|Y(@buyX9bZ4ajxMqvLe#e|(8`C4CO4G* zI&c3N6n_0n&qzf#-Ff!*Gv9l<@V|8@R$bTk_2&cmFUxrTArybtGXCj{;C0jS`wQ>v z>Q7&UzH3kKe?9G}@i9C4Yi|)RZ@T~1$DSF~`|haMG?AA+NQ8YL#@BfFnV!B|2fPu7 zo>j8`<~P6b4U#Yt`TIM89#RtgkQYh+J~Z*FHGCf~JMx(CXm@$I2@-K1ue~B!xCorP zXbks`uA~%YnvcZbOY|afnGwawa&{zZ`cmPeT`j{|%ZCDoi|Rd7piDy8i{W@a3b<_sOTkzd&>*HHEJ>g3d(ZKJS$! zoD6r$+$9VWA)rLv+1csE<@XpI_?O;qLT*<0`@VrC^f%tvfX^-VH=lW$%k$!K&aYej z_R`}LzrBht|D89!{MBmJU0%&;RBWb=_e>9;NCR)RE&fmMJr%?XcOP8)RXgY{Lcn#WT`dPs zd-yiH@&EoYd-yLCzI?-#{-ZSgv)9Aox~Cp5 zJZ`GTYiR5n+y|&77YX{S=J6*=FK;?cem&FICo}l^+YdVUwZAu49y)xHJAe5@`uVOU zS9;gsc{7lAKl{uu-+Rh`^`fOde@Z_P;Pu;{dgklz7S?X&@Ht`a^{?O9axV!Z*Mt&M zE9Opqz!2Z@)I8kxNBy8MUC;ibmU|Mqz5KKAqLRQ*!^A;iz`mHL-@Kg66&{B(O{ zR@}>+$%`+C7hNwsN+Bo%bCKx44kAB&VG$dD;$4#MUfm^d{OB$n@4m4&?QJh!6nC}9 zZsXn%7!QZYoqOc70vN-$QE}nR``c(9Nq>0z%{E_t`r_{1tE+kEMG>M((S40Gl;?T* z{qNI1{QUdO4}XZS+_M15< z(&xVw?|vaVZA$GX5zFytCpM$<0`ReIMguQ>^<|q((t&9iQhHrf15592E_}}Y)gIXohr`#vX|8sNV&BcCv@7iu#e3!Tm?MS#*L^>_=u{>RD=1 zaBg{3_D5p%NA#;G+evKXU51? zP&wEXkJDwAFV{|1iIjs^i4RPpko_rN?M2(VJSiG0WwKK$^Pt-p+sV0Ho-J3~)*$h` zz?3;I4{FLuv_@R02U@T+%i3UK8M;_qp@8@paPj!3HWCCHSK?>PKFO zfR%2yy)gliJdOtaEdUhj-Ntq>BI9l1(qKzoTQ{`9i2Tt!jkXO!`{sHf^+(6v(lpJ@ zR@$OA@`eBnY*FC0vgSkbh!#j8fD;EJpd8dFhMCM}>YCJbAMy(DVCsiR2l!Ff?Kt)Y z#PEQs=tj5jIRN^ifIMHe!q9abRqB=pQ;TX#sJfkHXBS0zN3Ox004zHk`b*BZROaS^ zqZL)-iZBFtU$s~e=o#PwTgGHE0lblz0dOTK%xZVgk19zmUTSM5Rr+jifNId2!KL^% zDKQ<}J{=R`wEmxbgtXu>=^%v>&jX z2v7;%vjhDgip5#tQG9*?6k%FFV6D$kp&NL%Ta@SqV6h9Zj%C|N>1yjt*ZDr@b+qPu z-UKjc8lc{JggJ}D1-cJVLA5h?I)jy4-!T0Z`Uzk>;SMcA5Ku)ukjAHI6<>;XyQqvk zn*gCL4AX9})v$x^eha{qsSXsuc*z3bDh8&E?#nVjahJ=JumIGMF`#`FkinuB13)OK znt8`>*#NxEu*wdiVOb2R7|kpfp4!!7f!jvFoO6Se+MO;%r8g`NjzA#&Q;o*U!s;M0~# zm8Zi|8AY(E)2VmlyC^};sc18&c&^v0GOLM%`s-;{jTsaaU&$c2B3I%?i#q0o7415ljdpyQMmm8@6|ZAC(Hcq^_1S?=q( z?<5jIP(tzf3Wl$ZxJ?Yw@c;^s`U-)bF$TFs0oDTloXpPg&;trB5J=+DjP^KSNle19 zRy8Oez>1ddmvg384rvj90uDD!(O*LUSp3> zXZ78DoVqcPALDF12D;N;>;S<6KnZSVPd8g`Im_}%B|50PGq4y=ppUTWwgBWi>$Tf- zZNrOdIzHmwghQ2l#-)di#v+jYtvbX0Md5R{ilXhhcA3{kduY!|J4XOCoR>h7W|I+0 zjf0P5O@P1H4l41}Cjy+s=yu8&id^{U?^@Qefc!JXHA2(Qbvr@>}<92~cpl?_#dO%a-Z*@DFRciZAYh+r$aN5+Id4F{QJ}CG}(>_fM zEaAv+HImA?2M63nchfU~(8sgxY0!4v4FT=iX2Dsc^7}aw4a0J43$aqERJ$vJ`PJnIqJxFUCG)qC}JNDPh)2e1Xy*__OSLTfIUqokZ;c52Nw2o74S>3%&isw zYyt7!k`7ydwfG)Dr>09|OfyW`-S1g3GuE>y0b!47s4QPj0jisAN%8-o=V3!1&jh4p z+N$DQnWRo!bFIAD>hU(#0l&<1nj_#CeMYphUegAQB0v@JolXaP1t1*VkOfdO&>n&% zl`q3tIRrq<6dC&k2uc%vKBm*)CpslJjU@Kd1+g_}VcnY{J z35D=$Ovd&A!6Z~tL5u+U7zM6f0414#Q32G+ZV*5w20$#j2#B<@)z8`x0OW;{h0g~{$fgV$)^q3${)IuBUv{H#<{Kwu>BL?qJIuLj&p>yK-VQ*j&@SbuG z;2@sEuGb=RHmX!#jCESvP3LYfmlKYF!7$swa^n*~-e|Y)H)Hw4t=vg?P8XnkRgPS0 z*jfegph3`VIuk7%vp~)}9EMYx0K5!R3aAevg_;q|fi=k8n(2CUt-M~~80vLBrrA~L zbg(2KPY0tNOqPM(=?dG#Sud&PLRHUVsIG%(Fb!#%7V~ysq?!uw1BOje-4*H!c)fp$ zuwbi}Xn6qmVn|!3wwe+kuSqr3382iximT0~srQE_wqwphc9O*LM$;?gM5 z^`boD=4==$izPQJ1Mvi4_S>ou9p&Hl+ zpw}>9q)Nw2MrWY6DkBDa)=~gUDwY9)O;OBgX-u;;6*Sg7dGU}L?>qgym+UIo7@iYj z3*d-94bqy#SI7E%Hkgl(`GzL@N(HDWzm3S53@E+heyYuYFNXcbZ48lNd~TB|i&(Oh zW3{^m@MMP$%?8$kjPtMuqUT$Vtv)L?S!^&H`xdre< zc|vXtD*AbqG-I6|J5~Y&*JU)KfjnCD*lM>YakX`e5k!B%7~0BH0YGVL5c}rT`PGPP{8$>!$VYA(;aeg$BS75=n}AXHtDGy$$2g2S)X zNu_LxnE-Hle-F$b$Cso8Ox5b`Yhf({MExY(R@B4ckoEyqcv^TPyEen(>3KCzUuU^6 zEs_hMxp1Cu_Me7{l%%u|h-|mjav5`7s{kQ$473=>;gSq3ta0PK`xkfYAB$6rddPUUT#GBSf-*Bz?<53JaK@GL}Ywp0t})GWX=rAjUg){ z&UUGsLl2?W%TpP=JjBAvssaQlEhTynSn4U4!j^wFrU1s|Bft;}!&5Dd0ZK%bO%+OK!vim>vQQD~$a9fnK)=z0x>NGa=Vkt@;)yQkAMoLD7N!SQMNfb-?Eb?bRqS z*_yu54Bhj6S{gPrfn|631VGt6FM~XNmhIzC5{d5U9i)EH-2q572GC;6)O6jN2(g&d zpcB@p(5cOP$4LR5u?wAXHkk+j-CXpyw%z4ob?GhY3D8oP0QDTJ3}9EC6i8MO(Pnlq zVATMM+pug|DNT@tQpy%(2E$)rRP(u5o;F*DLnZYW3$vx1cVHN+h^|?gQ4PwY22PxL z1X$Hm)PtnJh!mfbwQI@Inh)3JYMW zS#`+@LV&Ghfv(qt1^5?R(9T3}*5~vTr;Q+U0A%e0@e1;;b*QXO%higls_ukyLe2bQ zVGb0bg6j;FO7+yEb4B2E0wlQs%9Cbk1<}3C4#R=Z;CxpAvjc_93xSgk@XXL}Q(!$z z0sS5VQFt`!s_TpeitE(5)|voZ#mCjCb8t?H-v@}#X?`lpOKDY^B=PiYSOUkD^V13P zyEyA4{5a)m(?+AUII^o5^j+Zoju<{6{7)9aAw`SmH*QwLMjGmrYE%HSi1Ur7rn5Xv zh)O9zw>1y>I+nAL02Uv7Z2u6z!B~7apa7hvL+zs|OcG6dNv>LuPw&p{EHl^Z<$!`5 z0kgmXYu}H(85SL-%)x+d7jqr^;2pwRl)>b)JBw`0Kk!ejgqrE%b^&aCV97a zEQb}GsRyRx6zT2}w~o}9xl98IUNMbVm(D@s8qF4#_tk;2H*zNVvK|NE3r>R&FerRURMg!-$ z3w4R%K!c{9u1Ae}Z!h-9zkS-AGEL~`e+UJ&X5kP_vlRrS zs` zW~}$-O<@c`(YZd^GGaO1==u2EKC6q}uI^|Cprm79l4hy5q00lVIPP-cXrHM$l{n3A zyzj*w0IT|ZE{lrHpwDU~VHr%OqFL1@qGH#z`j$Q|&xAKHnq9CUZIjrqo2T}8M{KK& zO(*e)A+|mmC-io{)nL9{kgdyEmcVFO^CAHmqj`2SmsU2dl4qsyaWxbskTo>8oO{=+Qvw@=!KuawKlTBSJ z77n~O9yYvb+G3Y}CKpqHM>89ZOLuut>6A51iEIGl1skEdI~Q~>SX7&w(u%WeI(9}% z;dJJwZl~@ZH*tN$k*Kk*C~Pz#wJpE9;2h`NZLQ;uKhaWP19KKRAEjMD2-4Vf)NOy8 z0yV9uV9=U|v!5VK&WH}e=&&<#=jMS7zT2oo&h8A1_4S%sE3lD5F@h{U16(y^J{RbS zqVECtba!ObHtgN4deJ^D&TSsHHW>hS#lToxCgIraOjq4GX|$EjVaFzaolZj-BS5U~ z8kq6@dVSnHExQ}E)7Y(-%QJZz4L%w*^CT7MR(8|*W)@Pl!u*DRB$eHggt1HHIG7nrB}tIywoPm zb;p~x>l?4Ttr~Hwk>*L-ZH~rEuZF5f7hs++@*g$ z#|o*wpnLW9ZbIfbwWR}nHZvCUept0n&02GfQ~2CjoaS_SxS6;448KpFJzvy^b&;&O z%3#*oZ)0Iw^N)+THtu?SsAi(z_4qUP!HM{1v~f~ZH-`gv=JR=*%l2Z@nrguz%#)h6 z-ZYN&K0B#*>#^klEU=TaK;LVw!5i4ctlr3!g56_0yEa+J)5iJ46>AR`JdiD&Z^=6R zNIs%Jsr45u`7Kx_>~At#>(*06U~Aa-q~CR}eUi^p0p`rf1@v)M&|T7x(Ri>rnWqG2 zGEMZC=)5P)xW4Yg#+~|LCVr2|bb!h19Of+iW7z6hXKKuw0EmTNED>ASRGn7M%xZB~ zR;GHKG?-|6wyj2X$fd~*?+YrB;U_3Qx~vd4S$hN?#7M+1l=h$og_&e`@nJY{jynk4O7 zh0f0-Qyfo#g6p=d+EzQkA4Nn3IE*Sr9J@}_^OxlB!PEy2J;=n?FzM#&(G13cb2%s< z_uct%-!@ylO~aa`quQyOOD#)G2DSZZtlO}i9p-|Q30p`2`-}d5@l~DWtUB6NEXaKF zlFl1;ddxWBu_qikMw-{Lb(2bUcCu$%EzqaJCT50aPWHn9_Fnj+9M_J6X<4XtDMKJm zOSL|m?RUyB%fu{`wwW_Iu6u)~8{}ztl8k7%+q=7By;1XgH>y(H@=gXLdTtlT3a_$z zm-siEk_@}?Fv~~9qVJ@Gcu#A7OLnvPH0tn>F*;6%*Z1QoVCN(A)NoHdoL~3nABJKJ zqn;b#99yRqsaX$YW-Sj-J8oH@4u^&s3G2U?E< zknrrp=F3H|;oQHcx;veQYiOj8x3dLZgl$2t=K3xWsChrSYhwzTYCnlX#jo!Mg0(nr zx{!L6nYCrhNd-8{C-`*x%~j2-l@->S28(&QzCMR;ULS10o7RI)2PnxzzN+)Cr`IxF zj#~TSy3z0(gjf6-IL|V9Y>@q)wA_|AG*8NyH7W&^e8wu*Ea;b9?{MhWrHgNT*fR2d zq71`XFjEF;wbA58(r&jnE7dbM$mwXlTI^KeBDmI@v{SW;d7wZMVE5j3_PCbH&$dnJP{z?DPQs2CHuiJKI-nVO?$k00J>RvH0WBOIv+nAtyI5=o^YcXx ztl1z>k)~s zs5i8oqG9Psv)o(fuAP9X;;=u&KIMb^MThxuyleDF;9K=^%e$<}An%H8^^X4p+??5; zBO(U{!%@H9Y%xT>IF66wIR{G#*LFMom`9)lysyWe5Us}jTF^jv4d;!I&0XK;hDoNy zI{2PV&e83~ z=ZkanydZ$$b2?m3fqlL&(kyGY;j0r?1l)dABzk?gT*Ak#mMeOs)-?Js$m^}4A*B|) zLnCopA8mcd9k$k$vM6uIp#&GcSw3$Pgr>|>fF>*B;r`gK>r`VB97cysuf^VAreSf{ zr2~(}&CNmv@?D?~eIvNUu23nJiatY?UvDC(=`A!rV_n$$RR-Vsg%dp_VecAvozpk`yYU##q-1acR}G16LnMKtH? z!G-=H=?kZuNm1HhCk@JiodTw$SSV*XOgBQAKuufYDw7)4%57|BQkV617VI9>6)J}3!^G+s-hI?Q32~Sp( z(6KuXEJ#zQ0vAD56_@+fy26K*`Y~!Z>m7Dt5V^=ZxSSDGh3R2i6fwXvHP&ydBJ|sC znO<}Q1Z%*?$EcK<=()Sq;<(vWm0q*6Yixp(SKW;uipi_A$ILdnNuJ8RZKaV<(snJ_Xl2T= zmlZP^D~-lDY(Y}I$dl^U&CS-N6F|3#8vRg%TyJaxhcL(0o|e*Gs~!+}K&>|hm1AQ! zm49p*es+nqYxaA!alc15&zow_)(3G>o_pf1zrE8{dp-Lr`I9%0qLb#0m)?bk&-(_~ zIf*O6kKDyzyCmo4u$gp?nZMTNyZ%NTU1HPv&FOev_~*Fjxqf}mN4l}!HhF4dieAub zQ)*{$iBp@@?J$W?11CEzs>4({nX;c~VzTltx;EYHO?yV_b}1JgxDF+ZT&quX@9xoO zhDNT{;f|8D%%$mS{i2g+L-381qU>5@Zh2|PKD5V2W7sk5&B}D#%|X8C{*C-J^7(x) zn{*Ged{yHX%o(mF@~61iRp%+rbTKV&k9<$f5X?j5GHpeNZPq9sYQ45N$bIOXun%my z{Y|a2hG}e8lg(v3Y{&h$OFIZmi}rAm+TyCB?lZ8OQwFN%d(i_=JjCJ5p(t)P7CODj zxz|@0vjf#V93r(bA1&5&e`jf~Aq8+g46~KYF5u%0`z$xE#cJ1}wUr5-yk0+V0=uo8yLHx&8ugmDJQQi# z9b_pdI~RGPwRA_zbhGOy?TWXFxp1ouS1Qrx&*S5XJJGpc$+wKn#JzgAB5MwwZ7j4d z?3Pl%vrN1e_pwr2Le+=wV-p0{qMikwCoWQOa6>ETR@&+E@w(pdg27?W$bv%Un%Yv7 zT9TdBcBi)w*o%FKruvz3aK^(^SwVnUZ1mFJ8b^y}bdS4qy=)!CU40qO*LIeQ0e8@b zp)c3KN1=vsf-Z2GHV<~$dVWfjfVOGAa5kRpTh$uFjJZ~G$k_C_f|%SIUnOBTi$eyZ zuA&-p5`vu(txXW(+BU{nJZ`Qk)uf)Z<8`yKwVT!=OEX(EnQ2dKjyifO$Q4b4yYK8n z51H(i!Kt>}hU~zrEVS~-D-W4neeLaGvvrAV7bQ}+%h^-U&Kq?`l@3`_Z0DOgyh@V^ zSO|hv^|YAUlI|z4aspqwU*C;4%W7-itv4DW;t8s4UY@vB4XTW>r&Yv!ZT6jdBQB)v z!Knc5xH*NXa?Vz&kTs{P74#;x;D(J!tOcey?GGRTBuTNanWGkr`Dqp^nyPMsajY|Y zVcnjky{2+r_if20xA_PY8WjSfK)755&9gF&`d+wA~!N5F>!?k7le9Tgtyxca}I#(Ht4KZbe z`awAE7)qhH(!q=>;;yJt5QpGALhL}L)f3>0G2Cuw1;N@n@TzHPK2*ehv*K9b^_`^= zuq^EPlIxS{f#3)XV&|McBVTMMsxSe zf|+cq!GLp?w>E=8bMCj5p1juNV2h{{$T`JI>6|PlMx6GdV()jxL_MnorF7Wkp6;KL zW|a3Bq^i?=muv85^Vsa?%vNO^_B5)*xydzWfia#Rs8MA;GM9b?JX7Y7PCaDA zxwf0si$q!mTShV2V3;_<_Iz4(5T%L@G7dwyDLVN+kT}&#{OT-h?G;|E)8SY>_oD;F z%T}dRB@3reKL|{wplo^6=cgMZCUYK6Q%9YxxRqeWRt9mf?h=)eS^zf&-M}(`C8p5hM z?;Pwk9mLD7w%8!DLUIK)h2CZv%{1z+S6#K$8n$r??|CrLmbPE=ya_)654KKo)6vAC zEAu8z@4C+3L9`_Fdt(v8#1yxeVJaXupkLQkv&pD+n8KptZJbrD1zCR9kbT~jsE*y4^FgPvZ&v!8-%&bE9brGZQt{KXV%qC=LuuC=`vz;+(YP!ph#8T=g2h&?y&Md? z>$ukL9896Q;;@aajT!SAUru5Koi*vQH(9jeiBZYVjDK{>OjYPmM>odo(UqV~OL0(P zSQTM58Foa~WS8PlbWvNIFKY9@>qfDVdIHZe5-U>xDC4Ce=jsEZ$M%PO8e` zt~aZW!aza7!fak|E#>qm>^Zkq=^d8qWut@WZ&}JS-)Jo24ja$MT1BO_xHX>I`}(lq zH&?sb8L=TVtQ)UBqT9k6{AOP;V3`{Mj_*obX#R+69H;g2a^=({N!Up6c+<^OV4dBr zvzV$>WwWC#9TB~Jch}fUPJby0V~a_~-Ep%usN}Fcq6C%WC1AVNO|xo*e6jziq|NIGDvt6YkN`SZLtUzC5OO z2pg}jJiZk<9J#$#8-wnQX)WoZR@BDzi8?<_v%V(wSF*g$=owq3d`p(j<(Qi1<-HmY zJIjC?x!$;6Wh#NvtR&}(cVt5IEUhg+8Rdb8*!yTTt%-EjT3Xf0w4f(qK24FaQ$6}Y zlBIp30wTuSwf438I#rIb8@f9shs;+m>tV7RFeW3OnSOtlPL1PiuA3FnO(d>6DcU5r zV0Bpfidk82;_j%oYwk>|U2eGgcG*=M@@T~JQGzg=eSJ4|2uq5mlU6To%E##7Y!!`! z`*8?3Sao)EPi7|4zS5n}m@Q)54@bH>P z?AiUjY7^N#okXMbfCx|xuI5c}2qd{PVaLdhC?uK0pIE}l7sNe@^F1YQpKilEYtJAF zt^KJxuhWK`bXv2V-7AOjnr~-bO_Z2WQHDx}(?N2pZFkz+Z03`wU$nYxYjKvVHCN#B z^5G;@sfr;S*THfKoFK;S`)k%=8iqI|&5+(1veZlHk+o_6Jd0nQ4<~-EWs` znUR(!&cucfeY$t!oy0N|Ygnssn^P~gXH}iz!aZv;9kHh{Qey3Kq}`1?@u>YJ`-z$+Zm%??-fYndc^Cv%mKxQg&V;)G<9+>Mv9%SWn8P4Zb|J*(lb{K^@(`)NSIYG|Vhbky({KMl^S<9XeX;xNzAsi%1LNdoUMH$DYRlH-^hZQ5TD29JHdvV%2O` z5tUnQl&JyM##l%P$nm0%xqU1vB+X?q=F6Qklt{iGMw~yD_K_Ay}y+Q_OVv&8Pmw)mdM>= zKp4p4%5dCUW-~B@S=5%MNi*QP%2_#Map)9;v{hEED)5>JUCi{jxi zJNL|?xwzVKEbXkeD%ryF!fH{ri(2SgvvlN$a#}-Zmm6ot>43tr)&{}}mSTmQ*c9=z z9Q3>CTZ8RJtsl@t}^BQu(7v@<%A-ID(>2j^0Hy&?h_3A)H(#P4H zNo+Rb6Lu!2lUh-A5SyxpWqutA%XSzJ{e)z3neONehQ8NEk>2dDtiA$J0%g^dsG!$b z7IxQK9^_rmt4L>r-@(+Vra?vP!8tl6IXGeUbXuCrXxM4l{DvI|ne9lG<#5eR$Lnl* zXE#J}XuDh$W41@bjMrG#XSUN*xJvGgWLlMidc4dOJr}5G47=iP-D`2n+TM%H$K^~Y z%2NN7omN!US2z3QyyIZIWcG;9IP4IwTgmw3pE^9}@bH}9 zzI?k3@_mXB$5s7*hv{aSSUi|f$E${Z^Pb$}dQXCkk8ZaE{? zB|MzO6+_;%bP%yF8+Ug)5z_ zK%b6$3Pn z=T4_1lEnV4SWf4fZSM8PIm>ZMM9wKZHgUtI^Yt1zku6)dNa)UvRb~dBaidW2;iNlyw z9cPTQpak-fTKazE%EbDxM51IB4fmNm%;t#x4NCF09I?V;(Q;amm|WlhNO>ZXyh2H$ z{-GB-pj6@evAg06(oCAr9%1EVTF9&GV803^)^-Gb|3`nL3`$&=f9GP}f6cADRIwqK;uW|E|7v$9E(LHe)@&aEt5 zCzagK_ol$_$yIohZ{d|LPi@@os2rQ|%vjaH32N7DG&{+1W%u|s_FdDP!8pF}E65*W z^>V@>oC!V{f^aYsf#vJ@DuCl*$-_23ITw2>VXggaD2&5dJ+8u&d6Yb08i3*D=Bu%J zZO+MM+SOMysf1HGU-TB^XkX4ZZCrV+6A|C>N5##K*ew&>^n~t7%LAQrtJR3Tb^-k^ zQqkX&`E@jN9490;^DYmP^lf;L4Lhc>T3ndnF1sG=ZV~1s*G%$|w!0}OosP(XUjg+U z&dF2^O2Mz_IBYr9#b7k5I41Wgaj<-PCl&CvI;e6^=ce!E3|gYueA9+I^vozeZnA)68TjT@!umOJssg#)0y z*zD+J@YIJf`mhV<7Cb73(3O=Qg=@<0opy(f-!Zr(lFg-|8&nYPSKLC}l1`ZGmQ^^* z7`y|-Cnl=wX_$QHbX0UcQg;;NIG|+q`i|SoCYI({LIITPE#Mb@<5-E)&6O|m*fTqk zBiY7j7`8aK;h=lgOm(}Q7t74@TrST?yuBOQ%(bbb(ZI6DgEA6Soiq}*&+Mw}Fvo?IVxMWbJU2E>NZ^2jISGo?D4UexJ<1IBjB;RZhcyfMaUC~bmDOKZY9I18r^ zje6z_0!xPU%s)+T>=npr0_O5G2ro9A7^m1$(L}=y%v+K$u+uY^Q0V&)O3` zIjU1g8E{mrRZ(`nTnNEf6}xqApZYv_!)N8f+oNa!+udjqA>Fp`KK7_0Bt0aHy4@kkt|ArC=tba8 zF4RnMf~jf9=i;pQ*5qJ&o9(Z+VZ{VKrfbHvjoDAzjd$qBLgS|Fvj9QUnDsd$doFRU_M+xx|^xJ+0>-p0U_Oo9p-2pWWHuG|d86*pTO3;0kK z1)?nS-jPXm?)%a_lV^*Nn4k7X%=@*=8S+UA-BYbEv=Tgnv%al2@lsEyd=zCm?P+5Z zduw;p{?)N9)z~42cAy;GXVfV2YRz?P<-$;&%G!%9H94DSJmr+=1Dqdt~H0lXe)D z$TcBPb0k)}&aEX0zVROR$t6L^mz+vP*t;2&uv_X#o1A+bw{!JAzmh_KxH(Abo*4)3 z86p$IO^0(CsTp*X(PmmKM)1ER&s{2%p}QI)c(WVA9IL`6xRnV$vDD9j2CA5PWl#ql zv>(_#{c}ruK>upDq_@LixSc2cNxy3g7MYwqcei96Bzc1V)abDEu(#D@3FEMm9Bb@c z#oLMRk`P`Iv@8h&S{?VQd6_G=2#3~9N!byLxHNlmnIE*Zc4kOnvF(ArT90myV+}j8 z+Z^(4$VtgY&S{>phufV9*DYK+#wt)_Er3J?Z}*)xU%D?BSFjKEy>L*KR;~8y@Y1(d z^fD>!SZ*kI6E=9Dd{J)5eTScVyG`dPFc}9vzaoPE?+YgyEIHUJur<~BVqsjTyR#9= zcu&WAwh~q}y{4os;(DI3Gz_`ud&$)88{0-%dcJ*Gnrj*Ty>im}=u$dE)Q}hP&-7@2 zY>z;_Sw-?%UR?s+;Oa%6aUhvhIrWyFB}_q$r6Bz=z5?o^iiHR^mu#`unF@wH~?#9 z$jrgMBxjqY!4p@p>E}d0jrn*59Ig)DQ^Bln;5vX1sZevA5Xpr+M7$CsO1y$vj868H z6Zl2UfM3j1n#yv0dQ&v-fo?eaqJrjVKM@JYXW zR@f1pYgF%GG(Dk>k(8$EOFbXwY^?2JL8j1R z=n{**kWk5uopI%b9j+(IxIEmJLpv%x;DSI8PBsGL16%KV`i0+?w+(d}%&3&gyd}S6 z_{o`}l=CE)V53|_#Y@f&7!)U3pBi(-Te{6n7_m?{xr+I%*+s{KWc};5^eS*MN+YfQ)Xs2$jnmE#?a`OY<``e8w0o4R^JO?y&?K_WJF-i_!qTQ{!CxN9mPocz&zx8)tN&jZAhkiqg=%H0dA%Jz~%c9mfyxZ8e}PX<=2zjGX^5^Ij;?wIqNdJccO zyX5%Q0^C5*>uTe6_QwN&rY5IuP@ac-vc8OF8okmI@v>IZQ@q!y*}laPAYnWgB4-jz z`dYx)1Fo)@7Sk|`eyu`^6*Q0uk2T*hrhK(E8VG{Kqm3{sk4mF9g>_(g6H)~}bvrNz zz2du_TQTUaf?Xgi1h?=XupnJKh*?zUq)h1!5r?W4n6SZR-L*AWtH!6778HDc7FN-Tw`b|fq5*4p9qp`1M1;j?U{VZrWnX}6I3bw2CH`gv(OumvvD zui8EH)e;ecktUEuH>37gSmLRQR_Dn%*wjin?wQ<8m=!kFbs3^DvL2{6`6A*oYu6}@ zxj@js2QfGr@uV@9juLx5cIAQAxh+`IVz+fZX*xt<2V3v3+YXClG7%Tn$c+@nZm}*Kk38;XWAosUt0=!Hvc>I(#!|dZyY9%3?bX%;*M57#32YmT;+zm!KlT(~ zpW_7LlrT^|o=(C8&#V$}ci^vGVqK4s?i&(1Mo2Ft9D}# z9^xHCkHizd9W1kd>SOJzqnl-^x;Y<>8UISoT{7K+yxd(obYz2)EggX6GvPmw>?Y;( zRbtL9>ogC~WihzXRXFIS=0%tX6-6@8bJ^^Qd=_h)BrBO!xm=3rY`E@)z>4WhfJU-i zRHUicDo~tlvulCkZWz!Q>uDG0X@qvxQFYD{&2h?J8F#!G6y?ei2P!}Fh;d*x{Wzvc z%X0LEIMNL0YQVp|wg7~PoSCf;HS0i-D_98NcsLp2s~b7(RcCiJ5R{7dw4Gc@+r;XC zJ5su_$8=5wxO97dBwaRKmk{;!DD2MVO6QkQfg`3kpn-LuSohFOSDduGUJ-EF-V&x4 zAF(J_P;J0Pc_pcAybC{s{3b~=xi0Sza-!XU&ysnq=LJ)I4tTxw&M_cY~vDy7W$_)W~VB_nxT4` z5~?CCwFz|&G1cRGBl=f;sU7{%fInc(OqM(7sANuukYFb!*CB*Fs5-iCxjZ7&c?S%R z)LCgPz_K~0wXQ~EnS~W-uzY8)f}GflnvEEruc;G?>w|_M2}D2bBjDo1v)GWCUe42;1nB~oKFb6K+5~cE*=r!9a z8bt2RE&T0%-)CJG5biWgpu*JY_M-#*tL_e%FTDzH)za)wh0WpYZLe)Si!3ImAv0Vq ziQBnTB`SM5juxBoTxxCs3@Fvk#BQ9QsAYtCv1(h5h~=QvF1mTxy2CbR605oZtk!aG zH3hM)#W}Zyc$P3$O;$4(ooe451lx5p>Qt9~MP)SRV&ymI%INQc#+guVkD1HyatgR> zldt$R>5ZwA$m$$|P-oB2MJ2p7M4B^2i9K>VH$_c4A}=$xjfpxMuEy&jn|a{`j=}Am zU6CKuj=>H3P;EY}?-@694c@IdCbsk8B9%upIQAJ~2#?{#zGjH*)*oE>J~fFWN;nl8 zp1xGVq8QQAw75m_K96&Du?2T+r6l&GyWsllJ+AJq2&5;ZrPJ#-9UUj#t>d)s8wsxJ z?P4GvDqLHi-K`2}I3`W^Mwi2IEsZyw)I$h%v^m#mZ=8-6Kp;tzo6?;0QyoScAP)@T zsI@h+^X=RQT0-Q5t9OQrVC;G>W9RvWAasw{XI(=#vp~BXq{RXHR*2ztuh?GK2gL@g zx-+^C&dXKQyLr=rD)+;a<*tF>K&&TT^kAl2Rz8CY6w(*wkQDbgmRNhe8OcGPh?lGy z)}ztx8u^H&T5KW9;BXEh-*SpUn*}6HWQ;1Ai-S&gxX`M;)EBmsV!Sw;fq`kgtxsdM zo_W=r^%s(OxLx=Ci`HB<*|(^i3+FzPl`iaya=!@Sujq|%_A9x#UaAfDCtV^g$HX5B zw!9jJCo<8qW;cpKb>%suv)hX{bU-dtNIB6tws(ER2B%G#t3s}t$8DyLF8vm_snk1O zFAk+aj+AP-t!d({Bs*c=Kg3ou-#HXt|8=-^_<XX$YQ5;cfK14Qb8mu8@M=Vxb|bN*GRY06jbv;1|A)CkzgAfK3de-hhv$9QN$M63u#Y zZ&&tQ$zXW5&pSw)K+BR$I<0=4Re@hGjC+2J2l!4kYXWO7^br%lzR!+BbvQw~Wip_HUW}7rkD%mr+1WNQ1aiHETnN#D zrod=$ET+dTB7MQtP}f$fHeHRb$4!v!%gt^H0od;4bnS7-NvMW@K(TyZwo5?4sx;E0{e1&yqZ2wLqc8zT;p!iI_# z83ux6j)zpg1nXtv!4&UowdweFT}OB9(+t5c0<a8b=4y5slr%{o>t)bzGq^rERt z?8erOTg@^#^}sa%zax_P3nNU9o&xWzT$#LbQWG!o7&>VHAT2p1sYw-U95{gO9ih8xgtgaxBw_g+eC$bYksR z(8WMB+cXf=0T%KIc#aCUk>QwL*C~SWXVy?=aK{_aFKt$fAkr}1BCSr<7Q)C`#;`oMlO)!% zc2=T_Tz@{Lnl&l;M>{fkCb$B1L+>Zcs(Z@K&dRugaolW~q{EQYew29{d}rDS)rme0 zvm_Y`gxz1K$YzK(u*Anze|~Pt+b##-<8qA{gA)w7(-EkBnym(|LMfCF_^s)0*`0Ar zn2RVWG0)?qFc7LbSvl)SB)JBUqvx=FC2w9$0^F`@dH)jE~tG@uJ3%1qWH{ih+?N0rapZA%tgx)V{FSzY&TU^17AE;q- z9XPiHuGQ=`j6vG!?I5#2<`pDrxDpv84mwCB(1Kmy!L#0MZgQt#z2VN2m2PfxZEE!5 zLDO;KczDHf>gdS239`^6hIG@i{gfT%9PDv1Z0-DtX%3DBn|HMCn*H6)T1Pk}InEE! zzix{MdE_E%5LRi-c@{4tX48|6QMzU}4mr(Y0x?=q+^it13A=QE2K#r(C(U(JAa(zg zceLtYeM}bta@vwP(->6{FTvRLfc*uBx|C9IW*dZ{rKyg9ie%+Vz&pF1H(hcU9qEK5 zpTo8`Nx{&eiKfETijYsm6W(r{(ils%+-A>v1ftDxJRPdvQ1$r;j`z`c``9jK+8$%N`+T0pYxk&6smd(XM~;TYl3 zrM_9ruMEFk0Ds&hTa!wh$!OJbPE04$H*#_5;C_XHm@wXeaNiyGV%zdBMo&=t!+|-iKQD&Lo8O6cK+-$*)=#zjBKk6m`dqhmA3EQtYJne>n&1nV* z_aFQzWk^GX;e(l^-_N!Fko9BmVL&Bpv6M&mLXG2_Pz5C|s_Y<`MhY+#hs831B*LB% zSEO#hHPa*nv1KL$M&nGf%hkENl?jjcx7*EC0#FAO54{+y=id4@T0|{>QnIae7-)=N z8;vCL7ix!@A6|rWwacUdfw+jOH;_o%-{K%z5<-zH(XHXdcD;*J^TNwvm?KYLfq%C_ zXx}@Yb+aFnnxwOhX`U@YM*%wIsUS~}ZEo`JKk!w3_aBn+d@CoJHW3E$ zyaNfJ?&JuHSG+aEtB{JiW7a}sCx`^EdsYcGe$WHvm)qM+?GL5F zo}Y=ABts;6vegA z@wM0^Z#C}_ScSR>$|7IvE7BKR!wHH6yUjrqaOrq`bij!`L%m8XZ8Ggx zqctR-j=X*wRo|C;Yjwis=K!mBv!yA5AA`;~i8c09!$>y&1 zrnya>$Nc0Rd%QB-wOFkeOhth+0x=GNk)aIyl$+cHX=w3V0Ci}$*mKhK!e#?^t~Kfv zR7>#n?|ceoN3V@^wJBFd>6ND)e5edyEGbId)<)M?vu5gU1w5-pu8fY>={1fp0yl-) z`+08lLsB~T$EUs=BR)9FMlhVspw)W*{(CyWv$MDSO9=6T~h75oNpkhwegZOp#4JvY8*1S zK~$bl0n$BFm2$|g^jgaXYdMXM5TGv9BFg)yEjS$pSbGYi@(###ml zbVLvQF-UGtylQ9SvZXzjwaWsWALeQ=+s2piWX|{_k{rhQ9=_PRlBQZtS8Y7JC`CR! zXEAbk=9xJS-QgA3n9HK;j>)_k=@W0Wodd%{00i5OI}L-}lA52Z+4!)($Je)7i?!QA zW0Y)H=L$l)Y85Y%u#9`+YTByEfXk(NVoC?lIMTr&=?vRF6!w4dl&gEnqrm`@SCo;N zQ^$5$;2UahW`Tc20NhP3A52soHv z+@$KQS?3va;Y)y62Fes#rZwX5^io;%bsurSD4-$Kz0$Y3H&}-*72Sw>t@MEJ3B>?T z0;_AV46eh=eB7sng+2>?Bu174@7m!OGTCd|@i%Z!r#&`&!Um_#wL%0?3448LXtk-S zRGz68YmCDAY>u1?$BF|Ymd&9B9a7p=h6K^ZPIzeJ-f_G@rEZYxK|H_}h~8}%jlv63 zvSC&kGw4nP&hW4k2kapy9r4DgB?Dbl;SulEVr*wSv%>`~CRAwaDnMuf(hi8mYg++Q z7MJ~|KAwmnV_&Z1@C3(h&Kw6>8&|un=cqv8mAdG)_EG~JLD9Rar*uVqL+)z~beLOD zN2W~lGo;$js}UX()S0nwZwE^`K)$B$Zx_v>mvYPsT7tk#Mw_q1Ce)0-MCLF&Tu;|d z3A<%~+>eIWmO}*{a2w?gx@$eWWoMw0YdI_jLpj*3cYS{%B4ft_Yk0Qkc=5Ei0LcG{ z{c@K1MFrf4>>c^mR|tK+4oA%_bsDn|3ULi>|5G%a&(@gBJ~--7M>|4}F&^~rm__3r za|lWncIjf*%jBeswfpfA}O1$OgR(XsimcNu?e|*G6lMxFB zP@&OPmmPWHWCxNbVwI)n>~1zdoO{*4yJo1iLG`QPgv^C9o~nrL!pD&|XFx*c7ZCgV zYGPBM&aSCVaGYy%HlAk4n;jae-dIV;$m51dSRoHPkoE!a%jehJK|%clUp=p2F0JVa zV$0zK4DrKYE1>DU<&Wphnm$AW-A?`W)NI!gwY>F~X=F3e!jqRfT5HuRAve5&dL4b7 zo`}*wFpf%O!0o`d-)qy2+_mPe5NTazl(IEzZngd>D!>DZ>)N}G;fX>pGa%;a@Z}!Y z-m*YWILfNka9LolUHrP#7U%kn8kE|*q?@`7fT%~^Z-13an!MZkKdQ@81M1_wj z3ZO(IS}SSQMB1@GNp$B7ayL%0T)pfeVLIIGJKou<$ssFq%-%W))o@lG{Egk_@O10g zb@Iw>Ai69sjlAlG%u-);r>>&fbP7_f&h(A}hSZY+5)TKFjR3OlvrfM6?@6`hW(DmUW8-Gg5IW-v18QGOV|t^O5OQRJ+Ak?a z=}t**7)x_j?nA$Cc{4UwFJbn1&ThhWz!Yon1TFPEQ+Lqv3#noTxWN^8X3o&k&Q!S>kh9|hQudj}|PgpE##mCh{1DC0BBw0*`dy^ zf!l`k?rC_55LC`a(g~3~sy{JndmpcK%+2BuFHtUtm%7FH#V;8Qh3!pTlmIEScxs*A zb~1&#V04M`-IvXwsROy#xj`KrGA6fB(9p78FbU< zJ^Ku10yuGXSyHM?A(_rxszC|vO28u#Z-Xp4GV9hSg*10~Rs(})>wdfEYv;CITLjj- zFgmjhIa36cjm>7yGecLrn(39h_V!nvtt?&YUgd*Q54U3uek!%%Ve|0?0=BUE;ETPs z`8rEw1Pa$~3e2Ap^=AM!W~1dX?3}OhNyvrrq=5XAZR^|Q$`Sk&dH*1`w>HogIq7D= znv9rys;!{A9E^{&xdry)rV>coFM&l`PMc_4$3tLzmhGC(1lFbH@?pRs?`!7dYIcFEg-z#q#DjM7XfD$08&u!+ zJnwl3K32x(b zH;Fe3h_*DBW7rJD^#=1RIjg1)&mJ#Hz#jKdd*Y7gr4n_VHDkv$cbu}i@wA>3eve1~ z?HY7fBJIXXBuF<&FxTpw8<&Heo=Y+*hRgkE5*E-aYQkxm zde=5zKQ`A(mn26MV_DIgoFF%2c-gcJ!pc>p>?1LatyqQ9Ev(0UcaK|y_niJ0D`pi-ANwg~!tTvq{T5hQL642d&9@}X~o zGICDpz>Nt;9h&3)h7>sLbKPz4WUhCCqzrkW5|$Fv@lIRy(A(^ZLJ=Wm1$eEtum}dw z-KV|}Uiw~j6Q*O4oe_8y(BK}jW^%dr-8jyNmHudpW{V5-eZBSN} zyth8z*Q*?ONoaa?GkJ}28?rXOC3pnOQzi2KS

`^#q7V`6%Wxu}UCkdc#s|?Z#|D z2oo4irZNOJ*XL3je!L@vFqx++nM#4PN(L>r3gst8>v#cUal(OowD-0>Ss2#AT<$L9 zN))t?d31L+@5)nuHl1=x6t$exa*zt~tZOkb{g@wXZP8!KkA)J^Vt_~8)`fPGlq#W> zxy%->*)iDZ^+``~&ImUlW|QP9H&@D5oCC1Pz47=UnatKa1#Veo@qT__wG(6kP75kw zfG{8wS|WWRL}N26X_ly#1=mcnVK`VqGzTVU0{xuAr&W(SM;Yqp+MR~^Ygi4+?M);QYQjN zx!(@^^mYXQbCmXa!eYDu@dg@VH$}_VWf~XVCqMU+AZA3h7%eMr{c-5k8v@|g^6gP!M>NaG2?P_Yq+h%!D&D-s51gusI*m~;mSPd7)j_-9;Km%O_+n`4i zoqAA5>;wS5c*cRlwgj2x{UT^>yo~|S!1dxt@@%pp*TdsXUnl8y3+?}~pF+?=*=8A+ zRP1P`Hmq^B7DL&w&jZM)&7oO!85HMRILWh+zj`i9k92#hYNlH*%&xH=%U);g49pPP zMv=I@_51$LWTzu`P3@U;Yro7!sF%Y#bDGhKRO=Jn({VK@hwjNUZI$hzWE7Cwl6YIl zOD!U6-&-qS!CkjbxKx4|QW6+1w3l7F5K1X^E%Z5mcuIdm738X++$8 zg#EA@#);B1_E>$TFXUwEf6k5ew7@3&KqUkPPJ&a?F*-B(6wG zMK!&uS;?SSwfY?7)y&f7*jD8LJKEp4-gyW;6xriQ1rjLGbv3vxw^#&_V499)s>*GK z9&DsN)tjtm9cWz}d__wIfr#&2n_53dJJdVs_WB;6HcblAcn>waI1X|%`&8AX2s(uF zP0_o~ENrFGz<6<3*0}3b2#|Z?E{JE3lN&OY^UHpRh|Zo6=aS~9qdA}kyPh6+!viv4 zPF~#W-O%;OA3*AEO_OIZXGulY#{*Pksj51=?Jtl5sXM*~9nWfT8Xb+BV$(4Y;%9=L zpoR^`+dFN}ryd1k(sd}EYyYZj_sgz`s5@{BqCBs`6FyGIHlii~rf&4{Y1G+7mFqN##5l40Gz-%I zWm2G~W9imo)FmL(4vKR|E@})oEuiAA;jl?NP9pD6ws2a<6(Wj;aIXVHwY4tVns}lr zpVrL5arQ2>nT+jX;w}4eRxjCBhmSHSip3d>dVHg)>#b%}b${IQ?s!MIz>u)8>xs+N zMvbCfTl*8tdxRzkqADoIlO~Iag|>iI8CJ7Yi*9Un#H$)0UnDkn?Ps0B9PGIMav_D+ z3q==og35ynpmw*66y5r$Wn*(0s(f_IQ`(21hKVPJ`eSqhMM`0`lTe0iH`)8ADM=>H z=`5b+s91MgFSRRhSAY+-rP2_7jSxzL#&U~mSa$B8q2LW_3}~|x=^}1OqmGVgU;905 z8waUoNTzam8_24B#GDxk$B0!;&1{_-8nBBSCD$Q3sTyn&C%<)Hiu+1eGVE#uNA50 zP`d<0EeTRhAtTwJRjc4w7A$k@w;EO@>c?c8NwAnGl1|mtVHvky&{Ap@YN&tLJ1N(5 zAK^gHHW`$C1MX6a*~-h?Z6bt9B^1k0s!g8)%^f%Ea4!fbm9d>er#n8$39C&-ihwxj z!rtmLe-9|g=Flr$fKp7<>@%~K&ZF+>rjjGe4KC31Wu}mnBa9x;x37cH>*=fB_Kb=J zjYl#I;0v!4-DasuMODc{M#;@`j|!9lo#qFn-_}jCa7M`}J;n8BC;zfv38=-f8cq&v z|J!;YTLLv+Ty!yl#>E{IaExz zbSEQi!AMGSqAAmd&5a}<@40V6#e&M0xB)j3($HPzc>9iYjG1c_;!;qvKJUrEi9`y@*o zqI-Q4SB4uwpMvWgB7gI_xiPx z>P>xi$}Ta$$v=>K*qUY~S@JbVtPFILQK?Sz7QWE7 zS+S_i8YbAEp@;yT8htAPZj7L?%`&~R$y%LV4mG6g&Q(0jgSwvsD!S$b?Npb`3C@tA+UqGdByXYM7ohwOs-7fYO}$ zH}6=6W*3g*oo}c^R}V9AvM96HWo-8h++DYqm(!9&xu_1_Wx^pQ*L|`LbdNCW6#Tx> zQW%a`>Ld*LfE+u@8usEu{4*=Is(kVm8Tg(Yol6XkCpWx=Qq)rB3E;x?nAhvwr-8$lPIM{f`yF|T1fd3eQ9($ga+#TBr1nP z1vSaK9BmO;e`H-%3O-(w z#3^)#QJvY!bdXsg0tsS@cJKo6CSLkCT+82e{h|P2(=((LqqADhP&?bP@2w>rzf-=T!sfE#|t@8s()q#2FL)MUpaeYq( zd5#cMQkb)(l5to|dA`_OolDk3KO1n7*BA$E*dVANR=g&uscx;QI;jwWhjvyOW}@0C zcg4m-dlnoJ^+1uH_FYpH6jXiGTXx3d8{|OyRoT|pQ5%l+rK}s!r~DujZ4ExuU6mO| z7>(~Wofc;cb)W2I2$lx}nrPCUlygC0`zxf;mRm!I0K1JkTxET)sn!d_xJ^%T?LB0FX40rcJsYtxlrIL1_~;;rq^9f}y)AF3#W}S1#9Dr%M91 zlx4fH(WXRtx3ic8H3N5TaYu&aG1uz``sXrrG&j&BU0FDr=}g4~VFi(GZ7_*WARIo+ z#>WtvrIiJ3-b;KyIc%h@QD>GzsY+<~ts#zw8dK0Soga&xv>f4H*JQjwDLfH%%QBQ9 zObDo`&j;IP1^o|*(XH?~sp7i$Kor~haK>JSJrR}1YpPxUS$-~1V}xqw-5E@v3u|ZB zw(jlZZl^B+PR2QaVsC`D=#aa%<`q^8cmN%5uz0K+lr2vl=*v{=t6C%ysE3?Isbdrs zp*aq`pI%A#`tYVmm7aD`#=3G*E4cl&u%}JXKKK8J>JA~wQ*3((2{Y3U687ycb*ry7 zB0Jj8=YRC(%jzHgz@RxsZ`rG*Z@4vd<|AGIT{=fRCU;2OT7hitj|1SRWd-(rv;QxOI|Ns76Wqa>_`Ct6OSHAqkUv|7G`dNGGe)+{;a#Gj(vTg5w z#C2%z?^2JQc-x-8#@O4D{lNHh%$nv0#e;XX;w`|5mNm+dMU{_`PVeUOx_&>Ii z<_(@^6rkYk!^ICA|Nc3eeaO8mPuuknbU@3S_NhEId2gmlocqlEu=yc=^kuu4KkkKc2@(?9*4yA^Oh2~T!=`nG32_MPk7*6#bxKmN+z^zLc?)pPux zpjp@f|NK9Eu>+p>w)?Ku`L4V9+s=IP-M{;uaewp9`f4xw%ZKM<yZ zZ<>ccdZ3{E-}$^lJ^Zu`Vpto?PdcAbpMLh{e$--yf5lZVeZ9}c|Ao2us*U)%cVGM_ z-rwBHzWfLH8<1CC{^ipDcm02_wetQU`i74GEC1!oYJ1!9SBBny14Hj}H+ouo0w#&; z9biVe4_5Jj!rW;N%jseKXg9$ufoT<`X^0;|IW7O=hdBUDo(tQld6Rpvh+6BdHAwOl zHY1FbN8<#n$p>KLyreakzQ$g8xU@E-_waJgMeU_l2^;LivV$?_29E{|E{${0KuVy`pS8`lF}bc#Zd?cIc^H2Qb*0)^vP; zW>`=>ZYWJ_ceUnIl;T@?1=~t!%|!IrM}?k;G{JYC=dB&y^Ua?vUiWSz#xTI6&+c#e zoBQ|L+rVT)7OQ9{{2|6{KjcA64uw5!Wv~lj8Mh`}+vtyf?4(8c;~zgfzkQ#-m~X%N z_yk|A^yg2o*#8QLg#c&6|K!ntdiPcQrtR(D`pP%I)%Nyp{0}b+?)lF5?d@-U$l`i6 zvJY$LJ@fItZST3~&vS-rc7H+Z`_<6?!fx@MNBwHE|Jxt-{=FLBhqm+~qgZ{zHt^?v z=RbS95&Zez`Om-2R`6!AQoqXv>hEz}f8Fl+=GCtM_C~KieNTUXwORA(cJH+g zBv=OsV3$@vAX?qz1q9h+dt4^Y66ce(5ZEUSL>tugCSB z8vVLu+@2Qic?!Q2P_mP4Yka%3AwoEAzj@ z4X{uC?prJKd31Lx^Y^~t&3!ejZ@lb%<9Zzu_~JJow}9vE|M{OU`|{m)`$yktSWmHm zuN&64d!*f8YfF0R_7}h2_un&Jzxv>LHPE*Q&%6Hg^3CHk!>Ru1A-9~LQ@{9m@{2b= zEvoouKM($>^NT;1g>#m|CI|BU{{&wpA3$JqKAcI#if zb-+J%voF(kf9n4&p85xR?0@htUL2^mQ}fni|Kfjpa@XIE?(Vm@=g#9XhAJ}Z8WfL$Gc`x;K0|ft*E;D@jrZZx6dB% zlgE6}8zf*_$NK&)K=6gVi`^Znis&cHU`@R;q1t*|}C+q6hgzPzY!Tm(vz0#`*y z)qG2*_Md*`?Rv+l{Q`^R`N==Oq_01E_mTatUuo)__n)U1mHxed{W+QU+?D(O!#o=d z-{Dcb+LwNG*l!}DC-|NibI{qv7GNuPe}W%p`ozQdQ^ zGa>H|_Vr?TibKBs=-r|Jp!aV4*p=)oxQ7l+pPzu z-BH`5_&wE)J@6j0e2VPdcfU3Zo5l#GPJ_g4TMFR1AbuIdNQf6NTjBE#efN>xWEPUb z9=xpL;gcW!)QU3Sdd&F#@TUlL;_vXLS|`e}3h&bwx5+|4)T#0ucX@-oHbzAC9=t>d z7JGJpkTmw_2q}3uOGvLz5PO6x1>XkgX*gbA&BecYn~QJVy1R4sG_w1+vG+e^wcCyT zKBVXD{A2j{jpL`d#k+fco7TtFxnAZaZgKBnm18uI18S!c5H{z{eID+)aMvr~aoW!h zp+yoPlKKE>LxAJvp+I%jM@Rq$TT~gq3a#h(<{41{>$?JJ2$#G(#{y6R1XrK^zNd%r z6te#1hy3RGA)^az6dIG(cDtRjw=;xcJ#`4r%}+717u>)b4={?ffj`)S3)<76jqyJD zjrV8r;~xVAaBmnf-ly;1;SYFnPd+h@%I9b??tKpzyI2W9^a!NDD36GGfA-L%)sNoe zMEfl|;;k2*!Kf;O_IPfW*wdqXJfJN*MO^#Eocz&i59G6lEX}IO%G;oKW!vqLiibBp zf?NC?-fsiyZP$Uz2mq{j0w7oqFDv>U1aE!%PaXjs?c#&ejvNM-XzOS*;*uka*vQ{hn*a?fd(o8^x>cf1LKd z^LSnzK)?0!jInFI>e;)Wzvm9`p3uMfxZgcnB+vhR+5Ydww)mAv|5hh+@2%zb{6OxO z+Yjz5@KukWKKPYg{lR@Ve(vehzuxWWpMA`Z{{FAr_g@X-TU`8}0e~ z)!UEK+=q2y>FbWUzxFkz<*V_1?IrIU)qBvG&&TR3VDWE%G5_|aD^jrEWW2uW z>FnD#`i=H*8%TfK>#xU|zP;wv8MwR0^1rFpl#O;^2VP`^5!@~S)Uji-G7PAhl~v46Orc1D}weT-HAa(Ea{E!_K- zgMdiGbGEk4%BGJm0W+9Yi{w`g3e;nG6;+Y8YKZg{^5a=b9k3usv*W?*Bj5aV+KU^X~?6-{;}H=bFtsOY7r2{kN~*UqtKg z8SAfpZKrR2;3o2mmqq+0TE9K$@S&KYw1^zxc&-iteY#1(gp!wbKX! zRWbfRGVb9gumAm$qx;4ES|sV-^OeURef|QP{eS1?FAwS-Z1`QQpbyOK4TmpDqqo!i zHnV4VkGtc;GjinJC?vwQRfMo{8+F8gcYc&t2F@eeq)O}nabWjNi*3d8Ndb0kLu%~; zwop+7)pNa{Jf`!$zS%c-npN_X=f)mh8hiKzw{6=(u<$1@F?mE@+u-x#ovb%U4>36q zklHSRuJzt_UgE77OYpj-H}^31J>dB@6W}Av*meO%pXEXyw6s8!dfc%~0C69GSR$Z| z&A)l`lmSH%w*vJjYWDA&iPJv;JNX%Ai-Wf% zsGk1kU;6Cy_kREFWyXJjgA*=73fluIqzuU<(0{X*{o8(o)3BYSD6l`zn)`I&$5~Xq z0hKm>fb>{PJO2EUJp8lAz~j?5r8}W`^Fr!<^X3XPQj&e1W9HNN{>LZKDFMUhPX1wk z{7C-!@w=BK&)47{UEKlt>1v2_iKPyPe+^z+VV;3s|d(D|JG)5jlO zBg^)pPx2pu0raECkD8Z{zRBu)z8Xoy$iw$H&pz0F1wJ*^aFe|%w)|2KPY((}BUriY+gQ~kKy8<5^_AO_dgwA{2b^( zJgBV#DslX1mVI8>C;Zh&__dN|&^ek+=0_t+p7>*;d*;$znTL?^shC+pOkt`WJ@XMN zG8HQMWxevjhN@K|!wqU?cRTBG3RWf+t#*R?f@KWf@?Yffm~Q3q*fVdscD<2w zPd?+9aoj6V3yv%N4OPy3v3E?a)Y%w3h_(#6B`J+}*UNHpnONCratUs0$irXVeHHr9 z;u)gL>1;1-MY>HFElpMaHusl}+1fQ9FVoxhxy-@zYOQbbv375`5f?gIlha9`1xsaW z?T^-Fwzu3+7a!!hfIhjuISv-Nf!v>Dtg@;c`fV(MDP7`)gGgSJGvoOAc<*q-blr&G zf6b=uOdap# z8wh1N?!iC-v2{g}Wy{l1Q)*KbA*=)yi#{vMkS$r(vuH5u9TItKT{j!hf#p|$Ob>^n z%oP(Q+l==#+XNA~M#E5)<7)#EjNrz##8FhMSbg!wmXfT^f-vIcWO}ocY?w(?ZgrDa zS6nYOQ%C_U#6ko%zD9GK<>~wqH()7Z+L9EhN^EdWz0;K*hD^#2S>Fc+)T>7JJwEjW z8l;W4TXEQ2PBpk~UK!}g=1T|-;dr_3^Ye*4B%raX29l7Dq(F-#QANsp%Jfg}?poCZ z?0k#)z=QF7RIl7}BrZ>olQ+v)Su>3TIlK;xjOX=XAfZSp(hUnvPK<+lX0{ACpFnMw zUSE5h^Z+$O5XK&Bs8D#P(}Sox^?Zb$M@unamGaF>?oRHAT;x(D)_7M2Bm$TEz z$Q5V=k)|5~IhmvnJKUGz0v39~@>Sg1BJ0(MB*ry8vWg}_5N(?066x1WH;5feKBt=DY9v^S zk``oNPzdvLcOU5>x$Amk1Mx>c3`sW>MKBKqEX_S3`Kge>d2&58GqB+Hrz9mq-7db^bXf&iDBC;e)%} z9>?3N9Qe%Rd&TjS{$QHCd$pQ4tJQk$G(0+8Q7!6OK%-?_-i4#t{k1<-W7qjtM<>2(@}tyYHPR=DWH| z?-@w(Hk0d{wKc^vW8{Y-fMHnnvut3B!yxje<~cN2zRS24LGJp3V!$jcW6Q3#T}N<& z>VWap=vk3YSJ@fVzIk2uFz-SI7O`q^&3(%jcK5!Q_nIaTr~T~m3?;WOQSQu9XHX31m;OL`eyGH!`;p=}?v|(P3B! zbiSiVTl0{L4#cw#@!ZxW2I?BI$%9t326fRc?g2BfEi z`y>pI;jv-rG5dgIMV2Jo2cg_rYKT{l7bMDRqP?-;GJ_GI2NC<`l7~7l3%l|nC9Scs z_McqNOeRnUrGpxTzOV2mMG-`EFkaYa88N*<86wk)q;p_N0eY`45naoY$wpSvU{|ykHENZa1hTm1k0k$)X|p$!TxU{FT-Zl}}4KKhF!V8W}=CLFhQX(W(~J(!ORG zb&-HTuFmEVqMb7y)=0=2GQw0DSSYbU8CX@-A>r6@q6uKpJ*4xXvA*-;7L!C`KdS`nc@nq?9G!YjE4xX2UXB-O9Y7AwoIQSg})T?V* zseadZu$zMK5xS;vHyqbVzR*`}o;wKzMyT3uuOP_;EX-9bxv zKDr8-yx9W@FXz3j$Xd5*8!PtKG3s;TA4tp(g9de9$I@8CWecRoS-KI2Zut;a2}$q_ zCKuE@t=1!Izh}eBiHH=sb+klR^9tTsGqRAo&GdA#2sIaA^0Bu)O|9;Cr3ncw zFH52vnKGSNQPZKivj^UF@)(4@*Qp(OZ8n+F$xXY5F{o&tfz*c(ekn#)Ob{0ba;#&1 z`7mymGelHu(V+xqD@%7QB2eN8T1IFijG1AO`79wR2j3UinaPQ_3VtZnVmYK8A4 z!N0L0FIzCd^liUckk))f4iilh`ppWS#g!fSJ3DlIcE8#Y=#t^TLV-*fqIzo;6RDiUNAdE`2tRJY9z}n;Ed+4+cT}`xF*tYG(JZk23B_A%NKT~9b#_*+Fbg|lajgF zzL|je#LMgO!hytFjAj4L+PS282sB1)R|g3MclNeEZElDGMx`&h+E=!0fzednY@}nI z_Rt^}A*lZ1apR9~cF)5wJk9NnG$ymt_GqS~zpt#J={Cj$^NGTl(fEgAZ7%1b4SG+2 zDCsk*N56VxFu!B;)LC*%0{YEL3PZR|^N6!#_3}J+{Qmq7Ml|boB6`YMn&u5_=JRbw zME!Z^le82#xPDs$t{-RzF;SCV2d%QRy=EpDiO)yvIWnWqHFvS068a(~s-ldn)Z^!^ zcne>|xJur|E_o?sxg%ni*~zqvwX|%H?)C(q5b~N?7ay=L(Dr>f-8;{4Iv#Tx6ivoL zd#P9Hcxl?M>~hm_7=kx7Z0DB9!Zyl>wB(7P?wU66eO`e5BS;Rx#500G3)g8EwW~DZ^Pjj}^pN zllj7y2c2w#-Oy=t*Au9HiT(6yLI)Gqe1o13aTW~tp_U3C1@gr&sXy2@B`tfGfvPA7 zyE$^hkqc&&l5HX2TgmFdrL}%fPy74lO+L5`#Xskw^(epDOo(HeMHZ`Y^Kt6t602lO}ICH_2js?zbCE_?`NnetRv?7 zVH9c7-KrHd><0L~BS*!u2)K?AiKgoo5MTy4VF+A>gegUZx}J?M`rFu zXsxQxI%%9wgv&p`eeF~bBL*$B)XKb2(_p!`T$xP6aI`)5%TtwzP@|x*b|MUVF%D^M zvfJ$TG23mc-gw;7;}JyJ!`)fj_IQlr_ z3&->F9ruqfsN+7z=aiV>yN{D$rjGlr#}_ac)|2NSCr9w7Yx;ZroHmOG=cO{y+riG= zE}plCe-oG4lWD+kPEXY^K0h8IEojR*yhbohfhic{h0-4RQ@l^X25mdry}x7p8z0rq zxXw5*kNw4mFMWm|d#qH80d&c(x38Wa=7#ay3S%y95@<&Y+KWB%2CC*AAe8BpKGi`& znF=>xlD?6HL@ola7+SU|AQ}@Lk1eb@ep)Qj3s!%8#N+8?{d3smGh(U zu_g-jtzMkK20fjtn;Ry;Y_hyyZ8yd3OKKtipqz)@bJr82iH8{h(fdPjC9jqe=s}*1 zG&`NxHMg2BTjO=D)+q$4>K8M#hj6YPLiNiAVOM#W|G3Ss@)sXe99w@nA zr3-;tJ@=ugeciZyBX$dRTuTx6F&)ViolJKuG;+k4AhFhhMY)1YbUq_RX+XHQ$fUS= ziS@qQAbYl`jF4jFxvj{9z5Td3n=l;urI9`hU$Jh;aWUJ=y^;7f8Ck`f931CW4~~2Y zPrMG5+_p#}bVQ&nE{y~Kqag`7ARS0wK$tA;CGYIm5+c&)!De2=mrWs*Z9-n0Y5@no zfcp8FA)dUS2T7`z{*4OYyCrs_ueLD$R4N>j-n?<~D`Iz`e7DxM8ad;G9h`V3PeWsf zSO+0;=r|i|4UN$|6F|LQD9%;{>4AwkCnn_j|4gl|H0v-zAIu>3T1aG~Qavx*QZQwUWCE}Yz7 zT*a=5KiTy0*?i6W{>i5QeYEMBZgQG{m>qF=>&?&m+AXSkQ^iUHn}J+4@~IO^6f=}5 z_EZP)XZO74QuxYed$ffLBXIje8plh0=RV+@tmnmMwwi_X(!9FiuGlKhmX2y2!V(00 z;37LeZ|}G`p<2}xBtecLeZ=dBF`g-28wH7K#L`HhH1yU{mo&{VuKf((e$Onpa7ZgR zQ%-W%+Uii*OkDcc+C;b&j{fS+fcfW)2NHI#A#BPV+7cP!HgTI#vIq4Yah(&#$LMoU z?;HETsn+A}Y$PJ}UPfsPyS_JOGrjcc<&G)dc4$+m z33-y!{soD~GLQOPi$(meivq|+#xYStsYl!gu;mrZnl&0Rjf`X$7 zW*^s^Sd{`zxV3cgWz(d!Ba^cb8LKlcFQ@@t9u!$XF3SgKu2ss3-^y()QqL#W3U|y~ zEPIuKW)U&2&kDJD1xo=!0nhD$kv|3$U*iDFWhC)=U?r>Fp7uwbf})VlGKPAG<{3}% zT9NH5Jl`>9x5v5w860DM1z8Cs%KUp4%1~>m>+<|yc3YvK+%MaWer#-ORm4N&DQpBe zGGvfYdzd4Zy=UWi=OkURs?BxqnFjgnf*_7kVgxnR6Pk@oUXafet!m68uGH&mb77RX zpG>R!YT(Cb-7uknY6W8cR1o`_oAYU@NW#6YtF4CRqwkAytS30MEIVWFtMcZYkK9)r z;MhVP2m-&!xqkGOCOc;G5ZCvuENGW)`jToxVxs-z73I3`T2vJa%1z?|*7&$VQl2f% z?S&cD!X}gbjQfAf_kWmCdTP~+=t4U)i_%jR zi$TSaw+&%4_AJ1_TF)fcIvo( z78X@qkFFaEzdB-geRb$;0A7M9df$UdM>*1^47d5L2zH7x>-(nO%7{`07WuflmwIE3 za<;$D)_RZps4PQ7WFUIv5ZG4-F^L*=j9f+Di*AGISXZ8&RQfJps4+}&4fN8wt%5sZ zE%*{>%{Lkn%0{F&)+K5cHOj0HCgT`cnA#IbxlvJ#juC4L6#SX+1__al;*syEi}6B4 z=Jc^f{VJ9|a*wM^_DC{PM8SfD8c&=$)N*<{ywyKM=iJhK;;bCU$TBY)7K*od*~rPL zR?gmn6I%r5Pg~WXLp@YGs80JWgSvh~kkzP6M+L6u6^$3XpU*rrSHqN%p=%~OjObLz z?}RnH$(|0O{=`1n! z^)aDQOdDjQt(&ORC8ICd=lW#2sH?S>z;Tl`znbu=vK3`_l)EdK57<*=O9`E1m;n{nyUlovVx=O3OaI3gEpKjw^ zzKwR{NTJF;#>a<3S+OMQC`?a4+WVzX2Z8Mqn8q&2tDwLqR*bynu=C`x;7pcSweK9= zs2JtOUczj3U4n+);XW&lj!_tsBL#iAgV7A45Znrr37HI-6F9setjpy1K z*Gpg$w%uA~&!{I)6er(6WKZy+dZm}JH^^%moit(>*0aS}#b(2Ns0aZ$OcYt-*5zQK zDSn+>v7$v8{2_z{WKu}(~k8f@^%WR@BG6!)db|BDyS4e zX_7?{;l$Ojf%G(Ll@wc+r5k@e=3#*P4r)8a(?p(YnfBRy^}6wTvWZrv7OCz{X-*t7 zBkeDnMQt^?o6~ej&QWXhR6$k7Gn{A(JMJ&?s1$~W&y97i)kIi@5G3a))^D!25E!YYHmM1_^V9Yc^oo3I6fSAEoSnM&5tMII(w~BiC$B%XO zv94q_@Fh^%6ys`H((ilK%$1F*i?odzKyVmEZ9@7}BGrIi zGmE8jv&A!@xDi}%I74YT$dhpV<`c>Y-Hz->U+pN4HAP6Eb=G>UJBZJBzGojyU#{;I zCMr<4b`i8_1kd(V7S{o}8bAM&#q$BUOBVmg2yy5c*hin_SS=D0bd+8I5d z&_$6)m^9vM(8pANMHW2sYVm-Mm+s)n|70zJLP@BQ*xZKA4WDS`k zs{_o2jN_@10*@doH*TEwvx#c26F3eTa-Ab2gA+*w*CAAhMA@1GOrSw7^4j0S=^ImX zs=0F?wF-RIHjX0OiJSfV7GAdXp6JJ%A7Wv%{5#6uWgE%mCryf)1=F7&H)o4z-Wh9j zz(owPftt<(s4!u6+An~CST1u^{%eRl*=<3G2V_kTCQf^yd>bd|xgg!I zrJSO6#`I9FR04gaTth3QWTIo}5@4yQ7YGdz+$=#L6*A?qX6H$Jemz^zm_Q z>ce?*AurFUr`y?i&!g~e(HkZ6@5OvvQ`JFw7&?+wY{)X8SDTOOWddK^`G#|0^~kA4NfKyl^ESfYCe)rM+&g$Q9S%#Uvw*1>5a zf_%Xbgr`$)W@E-EZBgMJbb8D=+_d01*72$!D1-*rc4~Qo$~QBAMJ`;Ij)6+`>p8a7 z$L*FYWwHY&y#&2;eiK2PHEp#R^#X(n{xPpz)K$`Y#MiXrwM^$gsu>5DOQiYc2D9~j ze|Yq>VaOokr$%9urm>oaB9Ku#K!64ZWA;qZ$7s8_t3F8>0Nx#jf(&V_%^t(?w9SvJ}P^^a*xW~tgmCHge?u+};@0_7-ls%vk8rB@tAcgu`Ei9?4e(u6U zdl#cF<^~I^Q`bd>Wrd6gf0z3Sl*I;4^2DY%Jo)H4gmaADlO8UQPT?*SS$Qrso^V#e ze0n|UaZzRK7ej0JJ+}Mjtxa9Vrcl%Rxb9z0>7R1+qaFPHu!(=lQT-t~sx`BkAylD- z!MQvq7Y)9zf)HA#6U?7y7qbLns;dm}pCQN{)4bIw7r{F{zhCL@N|zLCxiM`xg9B>C zBc5QJ-9?u~ali+5Bmf5QOMTZsk{RHd9Lm%I0=BS5G{Ao#QtP=pJKp07bd?1lyLJcvc=D}}JRItkkvCDZ%d}CN(oA-oz+>UXE zx~FF5l89JsXuVZtR<4~dE*4!s{q?w>Ca5OoCWQZ13geovSrmqm21g2OpqwCW8qq+)ZZBPY8*4!6&8mF#$BzyQV!mh7aWxvUEu9x$uR9 zlv}xKlPmAG4>nhoB=T@4P7_Z#Tpky$A%!KojN^z{>mi)@GR)?TG@q8n0{-@NZLbfh zHNQsnS(Hxfg?+~v3+41N_x8uHsdp;<)Rq_5&W_#-DBnhW#d_Te)O5b#!YTR2e5|d7 zq}&IZZks#}_o>|kHn9!)?8;hHTo{&$>n9$edBvs97Fod=8g4`Dgw;Yy`LlE0jN|0i zTN{org+QR%_xPAir1O#x-A!n%n#!2h63b~jy`kfF1vTz_J*@LFf3%oGXjWpZIFD?d zyo+!vt%3Jmw`94nsCVovO_xNNs8KtlS0dW@?;M1_%1&7lCi9k_mjzx9kg^B@oaE$; zb^@k0ahbOnr)U^1|Q6o(@Wk|6DbD51>19a?K8J|KHe)(V|1rP#;#+ z0?6ct+PKl-Ov{`RzZ$PStoD9V8=>dbC1shqGlo@4Q%A^#`N4)+`oVsKXMIxc+dhk* z9!YIi5+Br?OCN@8r8$Th`4{6<>*<+#t(=)OqhHja6=(O4=P)r$FfT`$3$7pkB)9tq~_~=b6S%d#_0ORE??c_XTP#iwap zwCOFL(?^Zu=22C7Q{bW(Ly(Z(hTHV$wJ6rqxbNh!@c3+_TMs3%O=Y?%stKO^dU9O% zgJR9j=iT#|?bdrrKOt0EOw`BT>37aL)Xpsut#_fjJUEAcoSXM0-D>|w5t_>e#;Z*n=?2CE}qPnZyR zO~R9|c}aE!&6v4BoaQ~BIyc+cnauPSt*mWV#8S84p{P2&t$|I6WAsqyD|;spw6_=U z`m_&R?R;;MD!FO?nRuLb6FVsDogI$}yZxe_wMdQJLm={`;^6{wRB5tLFZSW-xDX$> z=+h=pq*ZHe=dpa*_$QS^$#;L+vAAl?5o(WletoK!>LhJ6=q>|j^GwA(Gnv;fy|1mPEg#KZw}pARsM<d3>Fwb33>#*% ztmE0d3AqKu7wwMgrE+s9oNK#u5Q8jumOW(v!0FU2-w*IJ0GI{bSY!!1Q1>FM?Hg9J z^R{>8xDw5?Cm2H0X? z&r7N%85ahN$7~a|%iEy0?Msga+wx}ovwJQ$N@Y(Ma_#Zl`cxPDBX^<3ggAw~myOcc zaFK)sQ7f77BG0>XI$y4&$0VlqD~rErYP*QV`+T>1&iB+aU7S{TYO+j-`BB&|_jI>R z_UltoY@X2IZb^3g3>{xpt3l3!&;*uc{&xcBVmk zVs^9hbti5T^^t{(?f%6ICPiIpgNj6w4zfZrUS0L*npZps4QrqttL^)<_NXpNcVykg zA~KZTK|1A}9+qLARgqLuIq9~>iRg5GGKs}f*z)?^a+Qqn%C;o|8Z9J$KxIW<*wyWN zSqn$GKc+R;xJ5#V#v_^rPas$Wt1K&vhaSD5A6HLdnlflXyatC`xpEEeT*^+}CN^1L zFE*>kSa6oKNweViX9RoA-tQ&Nr_yi`a^AI?eiZ@ef4Rhqiz*+b;2~D-)VM|T(9nr= z;iPF_C4%Dje01c8XiBfegwpojad~0U%QH+fiXkpYNT{_hchOu$iTeFyFin5IKE-Ht;)mV^z-i(U*K(y4$n)O> zzn9lxL{mm%-Ng;fuUBb0L`-+2Rs^D=@}ZDB4(j%=w2VkNXQ8)EX?g=JEngPR#i`gya`tFj(P@_gj8H;T-M|+Z!tG1AVyrbKKiK%S|(Ql&=H7GEXz@UN%l# z%RAnpWPoUS>fR9!9(+IJ^3Ch6HM?{kR57;~`DG+>&td@-HJ=<}xg%OeJ9nWwm$^PO zso2@v5ss|6^v=*$$qrfC986iV=~W#vmvkz29dj6-mPjQgD=Fv8UL$u#$2ZlqBd4qD zk~v*Dn)FyF$kr);Wsjs>3T;MQyspu7r)pQca5B?vG?^NAc$`Vswl=B>t~5(?B)CoC zKCh#d=E#=h3RsJIOAaRI2pPZ~IWzK!k)AcreU|70Tu8?)A^ARcbn5B-Ae{(ivDWTK zCL$`|5dCO#T!p>tu$wgTJauNpQ}PU4^HZqr%yY{H!Eq}3M>%afp*D4!mofg^=5QHO z23?W-lF$T^Ne&h5Xx?i9|2e_s5QpRJR37H>>et}DsU9^ZnLLq8^cQJLq%d=Q*=0?iYWTZ0 zXUj^x3c~U7qE(S|;w7u_;WSp_dUmDV_$r|LX1-y?J$=ff%%r0e!76sESNGQ~O?7i& zNcHq20!&UVlGE~u`m`4e+nm@3Tve9(0CWLx6|^)}p3R=7*8A?{@=biK5u+W*D`Alp z$;e(8qdE~a%Xt8}1_vDSrXfOgjg^Jj$J2+T%x8rrd3Q2V>22yN<~8Gbh`);+#~ZJW z9Vkm-c*N21^_;$Tja8)T)Eq9dN_)-nMJ;-m290^8f^3lO$}5UP);}He6z_XVb?2Af z2wrGXSl+kNrqgQFujsiDhAAe{g6jIgoIZJ%xLct@vfWH?gX_dkqTxiYmeswKiOcPR z3mp*2(@E#7e5UP@$0xm&LGmV?k=CZs(buqiPomEMK&t0D7)IaSrzb9|*XS|5o`LIB zR%71b_TuC2W>MlX6XsfJ`h#^_tY*cv;WlD4J)d!I&EO%kN|y{uNgi>T_7`&|?9*DO zPYWr)(nENvlZr#|th<;$t(>I11_S#zMAL`2?au3TI-4OoaX&xw0i?}#ru zucp%|MDs0ivz~KPe4xIa?1no-Y$8LHay0iAqG*QqL_Ael%I@nX>K_kg4*bW9)(&vR zVj8Fe=BV8(YT?S)WXS9pC(V<1P~6v1-+1|9qs|$2x(>V(UkME{puGYuz-^EFgnrbz zoSS~Wmn2r7y(F|9KYGt~YfmetDig~qGf1Oboo?6mIRfJG+{M|1MARXPM;iI&JAGXT zVu?jctyS&uAZ|nYL?MSth`6!p5J*_!d72tpOvTgmAm1fmL9PK?qgk9gaF?U^+v+nt z3xHPvd@s|V9&2E87Wb~RO&=XW=y!Vh&8Ctb_bYLuy}8MS@?|KF2DMlz>_gtC9KSnX z+{bfC$yhT(vU*B)*w4OI8o$1a>uE35^Zs0>o7;9SHpUHr{ijks1X`!2XilUS9#M$u zxucb~>gVC&qR))%zJ0-6BW-J%EOrXHJJ}@ZalP%AyM<;+n-k4b zQ$?t#+lunV9yTt&SVeOd{YfT2b=DSJv65XOT0~dE=x*}LgFGgv51FNRTdDU=;=TnDP(byQ*y}@XGa%3c4-V~db{cTve z<#TTnJF2%lR*;qaL0$XB9R8u_g&R~wYAC@Ne< zRa!~9O4cUq6YN0Z8GqkU_=awENnSG8i!+kUzl z@@a3p?^~sB2Z9fnQ`?ba&f`LwgBzQQTZN`rJ?uE0?aaG@wIewe^ts$+C^S9itdv)Y zR}0(AzCy*&6VZ^o8_->XFnZ<#G-+LR-|%+|@vz#yj7_XXX6mEi4lttEz4Y#4kxvKk zLr7YFSfj?Z_o;JClk)7z>Mmerjs6tE;3m5WJ}vc~7ePQ>3`i`l)y4~FbAPwmP^h6N z^P8t`qYcLGvKRNYgj^i z*)nnm$NwPe$%Q;?As_j~tFvZTvOm!aQydNjL#_^lbtG;7(zmzy2;4zMt#cmeeY$qJ z)U)1pn&Wbu8T6ARNSfjwUeAXro1;(L;B1Gm9&I?c4`f!OckS^ zYe^=zS-@thD`yh9k_dS(AAMiS=-2hG^^G+8r0e!9=xPyFUBrv)&0f&%j_6+_9^?yg zDbuiTB(aa-|8JL91e>HYg)`g&{X?B1>gUsnU;S(@yfDAbD9rBIjaeXkHb^1I?oEXEc) z7hr$YO&uZ9Tlne{E9aCYEKl*P>jprA;*w+$dp*p(8{8O{6NB^X7Uru>-~{o@9sT$VoS30 zPCl)51p$pf-20sz#t8}F%Pirr4iw3%V}Pf*cQ5}=ZPPHHutR!oyOj-`4S(?Fsf3a| zPtDWKo{DQEO8tCm`2igfF)KVI3f}UOq|n`KYs_UmTPg{WrM*akH)W?E{DET~Wl+WawShTj_^!$Pj@@3Y>v^-G zfHe&WlhQLsB39E4b$|J=q%u(+QgRM*2UXvKVTOKV+vV*>p*ef`j2ltAJa?`nj^|X| z2+kuX1=clq&mLM;cQl^0J;EU+^i8JogsZo8ZC{x<+P_3}AYa}+`z0_G%O$u~8el6T zS9NZ-bfUWPn8!`95RBcsm+raP%;MFE@4@T0)vxW-yPC5M`4ROCPQR7bi$(+RM#a~W zDQ7%^%LDpno)7$L%k!&1)LcWEqfwn52t~Dz*L|I{ds~_27x=&E)3kDw&ERREsWH$B zCQuvnE}bu|(k-@fs>J~S(etqn$6te2b@4iaGM{T!Vp<4zqI+Z)Q8#aZ!(jY4(UqlBE}wT7d_^A{kDBv9N;k(b zZFBh`FW?o?RP?11Y3r4MRdp{_@8|P@pvL>?vn)X4(K8r~*^;T{lxz+qeFJ;U@Gv~F zDBR?3y#)R_yER*`-6T)NxAis_2*6I|H-E{!+&rRy@mle~m=e;HG3_|{0lX>mX(K0% z(-%2_V#-2}_J@OHUAiM$EDVQ`W5;X82CuvXzL*)*UH9sH07wL6a#+zF``M1Ib2Ya! zs@2fo-lHMH1U5ZCp;sbzL0tm>&(B?yn**N({vJPubN>Rc30T6*(>>V7#eSq60q+i2gzT+GW>I1&LJa)W3n1(D zc^N&G^LYIMAx2%;JfG(_qZX*~#-`AdVGsb2UsqmlUJEPMfNmypnXeRXU$gAkGnQ}A zIC}(73b+SQJnu&O9@SvM-9s3TzcWt}#^jyxcdzu;O9%gbZ3eLAv!@LUE~jhwY&9C( zabJ(f4R2?__0WVx^oMHyJlC7+7W6^jXLvECBz9~nv~FXWn~~3ia;aN5UDmzjRJtDpOr14_uBFDU}QcCuV8m*>ONh-i_^eu0(|wV+z~3GhX`g@!X=Sy13#!F$MY|9kF9* z(7wAEkx6p^XGm9~*uV*AXn#cVq)R}1}|G|NMp@8?26@; zRx+(HX}Y;i-LVm!0DVawso6FT zJku-mM;GXk$D9L!(p>@h$iux3}at3?|&*+;D#@Bvo7vAx!}a@BXR zBOy7TU4d@{J&?Twb+Xw*|up zj51){Bo<4C&K9tpd}F%`%e{7U6}=}EF#E1UP*T5rwH7!mlF5t3+LpOj!e8J%@7{e!yKfTC3_=PxPl;k% z0G$!q2JPb1p7|bAG<+s*SvIZ0e#w+Tyr|r(Uf$-kS&9m7zC!0nP`}8u{E}-E|x(ki>ABYQ`qGfLv$E(csM$+{0x-d7m#<2oQK7c*@}OtVCe3x|6HAKs0s3VJ1sYoSCW z#=!w+#X}~I`F!L+$?w3sr*Oq7tj)#oLa!>LoS)4(>e|TL;UQQ-QICBZ{7u$j&3XfF zFZdG9liBE7Tugv_dbHW_+bc^Bg*M`WeE3=w#=)WD2Y2z5Pj7!}N4Z52QE4A!Myw1M~(~GrW*7Ldn4{VdiNJm(~#WlCbKY;@5c$Xd!ld6+>Cr< zZUYFZ3#(LuSS?K8hHb1DxRgHTE?*(=WSh+aI3Jw1sR!W460OU0k;z6FhC~MfL-8>7 zk5vx(hqUU9kV;Si-%n=8D`4U{a`haB z48V!*bsX*B-k2;;izBEYlvCk!gMXuwL|-h&b=Df!P-k31-A~8BjeN{CW`PhaFj2~- zy&aS4w1#z8U`88Vk3Id8%eg3H-$$MaZD}l@k^AoNd}Hci7M@)ZMUrTa@!E~E)uq(5 zF|wV>9{pITeUIE2tHTrT0J)nqALEQ9N>I=d#pXPpVl|-*HGdJtQ9n)B-g@}h8=`0r z9=u6-*2>>G4+i-pUxI&C;&P)c-nfg2WjngL`TTDqvF$(0KI=YwJ>FOShhP5h z^S2)EC~^yi?@{XDi^c+d^<5gk`uv_Z4b1&~pFDIe-pM z@ZpXf@(;HL;;_vL3N@eQIzGJVnW41MfLHtmCTz&Rf70+BS!N+Jo4yrvKr7yWk#;L$2wc{%{z6 z9X7x6){lJ8_b2O zpMCh(|JsT!*Gv0v-o(;Sd>#c%K>H8oc?iASMpSN$V;i=WTo{L?Q+>tFrOgXNnu`dfVHN6zerig$0l z`p*yJ$Mu{46B%pbpPm1~=M(>pcmA{D-LJd(|L8q#zqy=GFZ|Y1{T=k1|M?&DME~J< zqTih7$AACGY5tf-+1F?Njp1JYaVGXR=k&)q;vaHV-0x*)zE1St@AE%$#=m_L_-0K0 z*dp-D32KDN=$(yX7bT2 z`sW=!Td3bY6uvq7*P-xxiQ4?n|M15|Z9e=qRhtigds&+weB`5s8-(UQ$>$_N`HN2) zX}@?o4?p|RFu%A*w+4Pp@$I|$?euw%_v@d9=RUsp+thQIpHsw#j}kik-?@TL)m+N< zr+;~X*MdKiCQ7RI!+Xv8s}KK8`IkTYEV}dM&0jzn=i_H)Hx&0S(FHEuC@~s*`2Hs% zyp5gz%MatR$J_8lVD6*m_2D=EeEO2V*!{`}|LViP{5g~U#fN|S{d@jC=ucli`|w%+ z=U?60690)10>ZYjFeNA1%1@^&(R<4KKge!hAK%7ekEuKMYC z<4OG`PU`cJ_?1)n-j!g4d~!g4>lgm`4xdls$B+8)9*w6n-uN%0_RF~Z-pPF!Z`AuZ zpU3GRc52_;q#rrz4^Hj#SALut-#af1v~Sd%K0nq`%<12J5_9^j{Pg2iXt(aN+#moF0c75N^{EcE}|Ly{WWK-4`0-z(CikzpGC%q7@0g#{k|q@Kvw+84_AKb=MwwRvb*Hn zZR4uFZ@RC$GYu5OJ=^sTT^edh*^BxH4asACx=fqXV!GaM)E%oy>CxIw(O8ufUSIaP zOunK)8y1BQ)rev+VkwiN$9yr~FH|?#^6;_TIXT&AGQf)yed?c=XrBbb-m-X8v_qQB z9pw$?CwKij?4y;{YW%#hya5{RG%PuGeFFLexI$Ezn~^dnXVAk_)x|YZ)>Hae?Cm{z zdPglKlyOP-jyC65e?37DzPJDkPk_Ge0VG2Z?+W1dC3kX1k$X6Acho`X@38@9(-$-e z;@Ahuq{L2;T3Vf--7WF$w-nA6vK2yI*_~H)9yAdkPeSx~jdaoIBl%Deayo#@De*9t z$+cAoRUnpvbP2jVS*x-~h?yX@>&F4ge0%@2T2vk|qGdx6=;aw9IFS1SoV3UQr4+Vk zoY2t}EwrHc6f3BC2=&?rCA0zHB*0xYn89rT?uq)Q%%8%2guhX=J8iF^%k7#Fa9~h^ zb#C6JBp64?+SPU44p62QIF$3>C0^5v<4Ofgc%&cSX7d);rnWYuA6{lwEN0epwVh9< z$GQMrCIIs&(FYq-4;d-AGr+_)=%%-Bd6s3lV(}o$j(ZH8JykCO zGU!k^zk{!vHW+TbGpeFpPqgiNH|Zz44s_wXFiOl60B3lCY~Ac|l;!CiM!Vd$0_c7# zpdcL;G`%c+p+O4_91Qm)=oAnd@L~Zwh3}B#&4y=f&Re2h;7&sP%C?TS)S^9F(x&<|P!>Xe~d=x{##4Fd~mI(nd}ps|R^GPH&%`a!Uawdyf~wJ(}k_a7!~5 zNfCsAkkSG3TJ5L=Ix}KIv{hDo&PX;}p{_n69OA?>O`?H*5c-Zh^v8@4zE7p>QGlgp z`>qR8nP;=v%t~)(Gfr#;oD5h-K!u4lhE#bHx(aNbh`pN^g2(2=OetzAV7K z1uEJ(!_yrYv{;M}&=ZfeswD0SAabu(#kCUcNEcg`bcA;eTG%Wv#DPxm$QUrr0si`? zh|@vCN|D}nE0DL_5d-9ovc2$l%Yt3UA1>Ky&X@5U;NzMe3niLuSRiSuS;!YslaLnZ-h-Euk1o-tAH{p8>KW@G=ku z6u8!cAc?CBpl?!V^}2hU)sfNpT%f{Ja2O zrKmXhG&V1j{myt2J-a#_4$;W`D>m&aVA_gcgUYG7$KjA=)6M|pqTKc!iT`s9=91{U zD%9mtbWs8Yr6mGnd_++2N8!0x97lT09k4Y3_6BcBn{`??C^PNALkuRmopLagddsxh zewgnzJUFq->U?bxNA5*{{DDK+0ATjAuf^LWRABBjgGW6AW@=mkH2!#bK9D4Vf zdHF|dsm4`_TR=n0|KGpkyJKj4=1q)4wt9D>x&6Sc;B!VK!rht%Q4}d+zs8e*)jbV( z?8~NzIhu1fJ$#imLprTAFTUP1Gol6I#Q}*R)qP-A42`{)EJ)BJ7$U)&-+eaOT@$Yv zw#E6x4RqR3A17L4o~F`vUFW%l{jtA3<5jw+pBbF9cO#-AuZ?&fX>DQwr_=%zso4PK z7ZSZFVvzZ2B9=h-{2b4@>%XmDgf*U4|AuAed$q{)V3Ap81I`B%2hR|4o6nl9_1s0h z>;BPe;B_4?iK!PMiB!4uQ36z-2 z#dNsur9-}8jT|r#4XyQlkhY(?sdu2=Qo8(Rqxyui|DD$k?GQUzhF*S$@I5#rH^u7+ zI_*NF;D$LW3Cf_o&Wb`oPSnU}oZ(y@8O`eA+q#!2fp`G< ziZV4oLbt|gIt^^r5yA#epu2Q@@PauWvN9qs)EykfQno2qTU?@7K`y+1-888q%`mpJ zXUhYcVA};DNX@4K}ZZBSq#KVvtx$=%V&+;`1j1$h`?UmZW-r&C_i6 zkXIYU!ue_=#Ip^z%&!-$aZgH|-c>4BoqkG{udU-XOqn8{9c%%dTt4pytDB8*21{=WR^9QBp% zf)bRxeZnr)i~DvqHZ%-=Y52Ym$$j~=P`(N5tx_RA6%fyQzO3wMaK0}hxTooDTXvm) zhi%s5ZfVE~+0Bu7qxYy=Tp~$8dc_XA=(tNEBSj@Qlq@#sD&<(C|270ELT)4o_hrZM z9jv{PQ45HXNc3wIE8=5B zk$3+rxpBpPYp^Nlp;`mkHf(@Y$_hd=Uc~WF3@g3Tm0%kT(G{#lKjR{7cUAdiTZZE( zNPFN>vN^;Xai<3N&1x0izS4XzF3a{XNxu=1y4&n292(prE7jCWswq9O>1v$|8FQ!K z*4)tjE#bF_5S{aw2gMpbV-{@R+WLF1ZT!2RjrTOg(X~M(ki>i)@}uou4FWY2kNfqT zp3*NlUci1tbY&cNpfuBI7%gj)DxNfXUY7K-}HKpr8sj^6n6%S%?k9l zQbg($J~am!3=dn=YFkUVqkrZE&!+%H_mt z*kpkk=WEg@?P!usav0pOdaf~U=DzdR9H6yECReLKH?*;FDp<2`Qn&rB_OwNU4MFXA zmP8cs%Y>D;`NM$+g(M>1xW{TADsO8{PHOymHR`nb1+Jd993QO|&@G3(_3C6}P$h|W zNfHLX$jkf}H#3{05Fl2Qp3DYoWc3woO;~n+5NW1eSWXp1WG`6qyEH?Gf}I7{2Ab3& zhrK*}A=dMK?YqcWMQ+ZgE3P8?n>Vy&uR=z!rgYkXx_o2T{)9?*Bsjabl!wMxxg+ia z{(^t&;HD!w4S|m-EUKY*=|y}zmL3MD3pvL}tpAJP!OY6ffe4U^Wo zzUFtF7s!2sGJ+1`{a%E*ByG*mb%qTzCqqE&$0Y9Z{LexihXe6}&go+Pt*xz$`{^%M z;P9_^=nY-x5HoOB{ut?a<0NFSSb9)-5fu6EDo2G55zD(;V?3TKr5WV@h>G=3jM)=kS;zO!A7r(J_S8M*sS*#7oI$>)%X_o6R%%kSS91D9zN{O_mSO!nAM6T zh#vNdvVQxMLL8XQL}CruHfmVg$+_!lV~^g2h69z0&w{R35e{n%OW&h?S{^;E78FN7 zYdLM~HL`>(S+oX@U73h@y-4f%IaR7}^B!ow#9F}cF0zXaFx1yz(YvJoMKr9*VQ#*# z3lwaWjkJ!}x`^}>5p4=Rq}hV)t2%~NICh%vq~HY88;|}#B2o@=@j)aa{TRP|i7%q6 zSpAsyFdL+3ONR96&tjkRkmL<1A0UDncH@Z9Y;kwg;Jdf90wZx}(moK>Y?Av)!rJx^ z>d`Mro^aN<8`=1y&YwW8R&*G@--VFjH0OZ!JAejI3{`8hN|8v|MprO%hm^7^&q5|@ zz#Gn4;YnoOp7pRokx+f0(v)TR>THIA9`Ra-MD>0eSBzuU+0Lc#m=FidlEBvjt`Zw= z`SRLU0r{IQ06-{6&_MmLIx4?WV{6%T`mQs>X^Xh6-lzxfnoz`HX_y;(f`HIA zhMT61pLzYgmzFtR>SxI9?6xn|HnQ6PSwd>34?Isk5Q6R__>Ez0BNm=gS)yy7$2HOc zF>6|Hr0=puYV+0<(CUj~)YBO+{`SFj2o)4<=V5voWrbMk2#s@EQpY3=T|XaU>Ppb& zlcA|#vHb$OkE$DGd5*uQibI6WE>)OVS4M!rv)aydznoI8mG{#0p%j?Ouq;KB5Ix=| z66c@@%8?%fnexJGUOm@dd6gM!I@~cq*U!#3Pi_|$bwdaL z^W%C)vkyjDUud~X9y`>_cdj>3&X*^OXihL|0}dyNDl*M?J!_Lw&ANZ!;t5h!tu6YE_#)*eEE-m9?|d?imI&*;1YD)U;kq?YV(f!T ztkLC5icytws+<0b=NZB^udmUkcV_@p@4K-lg8aR=Z!*JtS+7AY^=z-c@1=+#ibawg zs>>$vwOEmk=%(Wb3fw_wW1)XP<(;pToow$#Bui2-$vwe5Ag8445rsygcuE=lSPh zXV7E=n|u-~CGA7CZ6JKsXVagn-VUD140OWVJVqj5!=J!HI^So_w_2=NBlLJET*l!v!F zw7xH<__3}XI;0frj`rX4)efasvNo*slV5%<-CBUF#cSA;GQS!#qWE1I3G*E*0wb}i=r?Wa13Xv51wh5bhn?k5nio;qhg_$_!`ym>Nyr?%J z+F&idT$o^wkW;m3GE}*=X(U%jNU%*m`HJfv4hnN;VWAGU^(GaxM*HfgkCv=*j>UJ2 z(_?>Z&&p*(AOert{*-g}nup&T+P?Js-d?>mvb4#CsPv|b+@&}Dw{PwJEfOFdQh&H8 z5`AOYX-sC`#cK-ce1Q^PBaeUU+hFb^+q^SR;fO+i?Dsh;(xio}(z0aG#l*BD^fb># zdmwK+FQ_d2F8{@w_Gf#xc_DHp_rW(J3Z&1|o00tpt9JEE+6Eu0RG9=3`+3%!ZpUAH z$M5{ZK&3%SI@?mf0S+O&@waC+EQxpG1awRc-zvHt)w&DuqH*=;k9sq*l%o%v%PSAZ z%Lu2g_ObWC_n_v+yc6=Lo08oWxV2)p=Al~k%~HNK&R**)oOXG!K5TQ^A2!+XKy1zG zyC7b&u=URb4f>Z1U)NiKL=D%ON4c%kS5sb$%k{ikKQhlIk6|Zwkz+3k@TH`k-JrptHlS7LFIjm4FB9W^;WF#$>@!JsOBkGwPJZ#E6x z>IXM`$@#UH6#hpv3r`pPvnuw)zt8syCtTxFq_xQ=m-wJ3UJvI3V^jA6BdMVX$`5RS z#O7(-o;PJ#Zv4-DFY95_p6e%3I?L&kN=^MSl>O^U>$CGKB&p@j^J+ip$rq0;lT*GJ zTX?nvTk&>2Q%bp&hwxdxub$;QxLffUrk9_R^?Fu>@)7FII&(e1!9l2p(RqGJ`YGlP zn$#q<*0WRfkeU{+e-Hw>Ga0Bsaaze=E7tSVFm^lm^9AL-exBsR6Nd`5fh!*Eup8O= z`H!0+JrS$)+`=9WYr8#XtuUtN^z=HR@jl*$GUrClSC>%RHU8bdySaR6@m^iv`S_(w zuWf(U@+X|H&Q~h@ASL+_rTMXKUdM+WhNFps;ern}K2M6MZLHRwOhSbxaWUjKj%kWm zJ)El0)a~8))7H?^OQI*kmJm}hw`QarTZoIkCNyDfPxfT{<`}f^R zkI(f==K;>94lPciFe zYTr7&;{?^p>Sc?NrHgr+E)O6P8y%-4cQg6-#la6h!p`TQMty2NCd6MX1Igcg=wNCaaKFcJNx-H)$*#4RP~2D^}_1*#uhjAj zI#Xn61P<2LUbW8!xhGGC{Zt{ko}ztS(UniDz*drZ?&x9J{PasB4RQNxvtUw>tH%V3 z5y|PPd^%lM^s^N{4bB{(#col)_W{#Qrv{F52c@7c?^*Q^X6llbQ{e(9TF=UFF`Ofb zwA?LftZMqEAHPdz-A-ecWnnIkNmvvuw)isCM)#(sSn<9313B3c{9?H-Mvr4?L&p!w7}<5GRGG-o0R>?jEt->jg@qR z7;fPEuhx7pzLy8PhR0oQ_qQ$O?Ka+YHvBv8oa6E3oZ_+bCjfSSeAB~+!XWm{+5!?4 zy02*8*yAzWU9N|Yg2c4dIG+nv5q@sX6}?^(Vsm~ZYJ;Ns$QoUYN(ott-T zdlWmLA9dOA+5I?I?aq$2#iRIANXTItZQZ9Q=_W8FeEmCVJ4za2?H&{iwO`Ae->Esa z+YqId-{WxQR(>1o7MZNq80im!6uV~&^5qNW-^}-aj*9Zd!5L_{!-c5Ie*rBQ#gBJnw7uh zC{;XL=Hp}E?BJU86kEPi0`=G%zO2>P*2uATu?Aml*^TDbI`*&&kawqnc>OJ*#&z*)&axsKg5)skGpKapQl-M7dXM*DI;gmBf z>bI&j;Z0R2xZDbd$F=NGV4VXyaNtldrnCFhx$|px48~ZY&VO~(26vIO-Lx}IY6$7@ zqO!j;T2dwGk>hj1y*}(<_RP8Z(d}H#^)BAxsHJ|JnRkn)_IY?o#c(HUzB2x?%a`k~ zp?EW(hTkRsN}~St-QNA~7BN-Z%f26nCB(SG_MMM_?1245x6I?$-=vhK)I}tmsKTq; zlFv8fUip$K9A@3?O*2D*bPTtV?jNNnAnCtp;}@P`MAR>8!fE>6@UHTBT!s`q%CYT~ zpMR%G-5kPi*N(RSY7#zg7JJqoiP9b$<9Wrt!|hWt$6K3RYJm@n-`Jeo^Hq(rzwZoj zud(O2#P!Ij;f>r6u%#g?RA~RLwN9?uC;TMD{=VbEoe(mG6Son})^AEoOX1D)KQ^`E zngrmFw0;6>4SqwjOIB{cU)*7ac;faTX2x{P<#5=j-zBHv9%1gW6v^(De*k){)I(v<8%6XK0w_@TBX6x)v zdfoyn^|4FM4fQ5_oXKSD%{Ina+S_-`c=411`%DlJ6l^VV@SAE{ARylL;ujzG3 zgOQjM#1?JW`Tf-ka==}!&%-*)GcYCnypTu2+d`aN)gA2~bKik?+~;??Wa#^`5Kah^ z{QKT8%}j~A9gigE?r*8PJ8e2%dmLZX`Pb7B>B}uOCx!K3$y1J8ZhKdGd{`ODu7GDj zPouM*i|@Gq?hv&@gRR5C zs=o~Ue!Pfmr~euh({32}a1)}^Ig%tmAN z{bu(2r-?Zshwl|{t%T7{eaRcR*s+)_xY&dCvP%(W5X%2y4?pHb&HZb%qg=uoB8>@5 znQd&{e*_>(3C-u0-*iHMR62K_GTH)x=TcWCd49?_%QMgP#D8YG^THteosYn4G4257 zlJl$W2l8kUE7l8~3Jhc{o8-THc;=2MifVnXP2@CcfqT*dOh;N9)}Fg?F>mS*#)a}E zHeQk1CZf2W1uD_0n60!Q+AGE1U>#{768(p8jY55p)H!b9&;c+NRz1LqpKHXP>o92y z>q_{`$eZ(D!on>4Pp`iG&X;?sI9=DcqovT6w_M*3ZclVT(#lA?)x5D-TPtRKq`qvd zai>S zK7tJQ>0O0b_^(rk4FIt(7U87?zQX^_pw)&07IMFx?#DwXRb?5A7L_;Cra#oaHe{Gf z8>$&R?2p>OvaLq+^uUCT7abJd+jMW8$A4T2>_G#_b+ZjXm$+F?48p@8Eor%!%SF_DhMjk!z~j$rivhQjYVIB=e($esaIjTb|HtQw=3w(2 z{h+_+wRm(p`b1ZhnT-I(3)}e4bUI1(efLPNBn!_^oMRjLnTMD2vi4N>3~_I7|K*?K z;=I--S8qNU z?o*XxdOv%YWQeO8o)CWa9P<)~KHxm$77*YE1ojJMY-EEC#aqCt5&O|^O7x#)-R&kM z7!&bp?wu_$|AJAfy8I6hk&APU818kRHpt~)X;}?4MfQ@ZKXG0pc$-ec^B=4|Els~vH$2wMp+SyGVo6&Vg2c0Uo%QB{+}SS;a|wSbD!Thj_3sk96pzDMFlxE>ne zF8-dD6c#}vWBa0&mOo>9jU(u3k7h_8!0spTGMeGmRR`_&J>G#odc~s26L)LW6Ypn3 zM_<}h@2*E!XfYI%2C7U4o@a(q{8|Qio@uUu7kNC_x@^3X9 zOSsjn`sE`gH#_FXG#MlI^7-9h!BKIb0KGTi%f2BA1~fJvnrE;2?LG+0iCkcOu!tsc z#a{f+N_v!p%sY+Dj{cA&e%=rYPNZD~U!Z38)mpen_my*g!rlvmwNR<%^TgC{Fn~m+&NrTGN z1Eb2lfnU(SfmNI*B0sl}>3(qAznbCv zg1qgEsu-AKjR}MEWS-o(&d(BbhS}!?m@Qj&#qeaE<-a?M4Hx@Iop;mFPL~g3y$s^2 ziLmHRxn8N^9j{${(|*8#`FSB06yRfsF#GGyYEvbBE%%e=k&ahIIDe$Wwl#id^k(0f zA~(s-?y}EfKEKCE-Oa+H};}jybjx$Sn?FeZYg9!_Un(=QocULvS`f3C46G^_{dq3@?WNR@lz)(ZV}_M&czPq zT@9Y>LB38>uFT*ZBM;}zyh&BA5T$xqYjow{r2b$^ z<#tef`-;;5#dv!d4XSkbK>CFq^+pQA@Fnx#$=XIJanJK}Hn^>?!;i`?%qAG%s4vI- zI&UF|Eor}ws&Fvr`;r+F=W+MN0*Qj~_!qbt;qp=0h0ZK|rDUI5Rn&^9naVx+;yGQu z+s5a!e=ls+3%`eN9KZ?C=L_L+TG`WwHzs^V1-DzB-S(&NSt!bWD7Esi?-npC7A(eN zd(}AY+yypY+PFl1u;wNL@!NF2V2?;tS}m>9*br!JRqb7ixtd!c2&3V3#d?KBEAVqT z;Wa)Hct>0%kdAi!@tfxJ*TCR~M}F*<6P@+XjxCqahkGLws*a$=mYYeJ8fW`0EKWpqMIeWec*k_5B@z zfCCD2Pfd_L?f1wWV>o;tN#@_&Dcviw@_0QX4yO!UW8X`$VxQT8FE9b9vhfrY7*AM} zpb+3C8RuVNgo$Wy$#XX&e&+B#ep-AuUi4Jz5~$S^Bi_?DpQ)-p(lOPDJbnDT8%_@O zAV%=5_WI`*Dv7NR*4h**C;!z~EO5H@mMmVop=Dzyq^p zi?K>=_mXq4CFEgeo=DSHkeQPZS=ls|7zWbuG=C@2`Mt?OXPIJi0p+PcOw3ed)3zM_ zTZxM2YLriADZFbuY)XOFn4Pze7Y%(hGsL^4GhbIK(%&5yZsy9^>u&Z@5^gq)lN;!= z#6GOtG1~EYkS_E-&0T7~^Bm9M41*c!-rSa<8tk^-E{eTnd%wSM-<`h4uMUSXIOqNp zXD>aU5!nLk*1-8WABxnm1qd>$0o8gF$bYL!fUES}jfNP=n)n($z{@9$g>f7lIaqx_SXV{u|h>uu?Px)0B2 zLER#c<69`h_#N#FZu3kjk^zz%d0X8)MReS0fA2xmoEEOyC=cF;uO^hVb^pUe9%I=f z?GUDYT=d9e{0zZ|e<=7+xtir&EOqXMD?tu>pPlGfV11tsQKI^+p;R@~vsa`!P1VOK z9?qdN@uRT*pRX}=&8)&mH(fDx<}&4aQ=L!hHH^1aZX(mtI1q!M35TgD+M(%-W)Cb- zx@sc;Ta0^(=S4d7f{wfWlm)oAN1bLG+yTsaoutrRpPWD0Yst>_A0`(~mhhw$U}S15 zx89LEW5fonhhW~3evUg8wHkGqzvmDf$0rI|{>kOR0a1OHv#yVD9=PudWTf@W8+Ed6 zEku++6Y$r7rRr`N*b5ojE9Z2NL3bqzSbs7u#XDY^?9a~g1JX0g*%u%BhFy-=G%2UehQG)EnBhnIv}>gRJFaBvCFx&z&RgadLW!cNl-mYTWxP4K9Knb$t$8u_M@P!L z2K&wvX8jT2?A!AruHO8^?x_Hj@cm@Efk7FR^Y&xJ5#%-{m9@JQfAc-u1lJ?$=6ug| z@Qog)lV)I8EEaoXUKX~VS{wFt4B@W4F|Z0Z$sQUQk_Qt4wx{5%^a!v62<$Q~!ivO; zDT$|FYVQUXyu0!>KW4;GT&Mm}ToqLl+i?JE7;}pKleodP(|7|Mej&l+nm6=J%(s2^ zeq52aGS7Y5jNh3PZny3ZDmTr(zp?##6gE!Yg4Q}_7>RG-t{D5`2Pyq|AZ7X^Nep*i zbui5EV~2&v00ShaD<=TEQBku+M(nCUs^<~yDtDuV9G;c9$U>MAUXaDJT3+9J2h6@2 zG)=#|QaA@dnu24rLVyo`iVL4PJYR|$sqM=5sY*0^q03{%zzGvzBoq>>m2l-=uy};C zAU55oa} zSl^!cN5;E%`%^`ATW=R?+ZzwZjsUX;>PcKkc`mC^;Wy`F8}m2XzdMGhD$$F&RLO7c zt;fvPCT9p}XYoI;?!l+F&Po0@OQyD2Heu}|<5zhVM6c`JrfY`UvefwredmH&K@78fMpAk2e|`V44^20!x)onX)WS}$XX<@HAnST|J!D%>=jjzb zTAQWq{*680k{66`x8;M4$H(D_4&S1e z>lyjWYVgTl=QiE9DTWXyP^hyBryAfmiLRB4EZ_fl0}&2-oI%4A^N42Oh)P{ zs&YyF{Js}}?p*u^S7x2jJ9VJ9{d}7OXZfZh<~ts>4R#N!IyIonegoEtelPtJ&~$R9 z!;*Bm9`|}AEXPoWrb^mh_B&8@%a@FlFM_qOF1VSXy(iI*r*=$?964n0XnnlEMW^8| z&f+*w9+Z|4zx|PWA5rfdX;kH==ZXyIj7I%yb=lX0IXg@1XFeQv?|dg6YsyBP>-)ew zaHoX%tu=P^o(-tJ+X#1(w6y6xO$ryQ0_1bxGF{CG&l7#%Jc*7WAIaW%qEf}WP3B&r7}|33h#dDyxqm1`#BJiW`KC8{X4s{#oY+N z{)nQS^m-77f>iBC%hxQHk~%Qvt&}$%LZo}Ou|9|2;oU2x{$%fWnCG%B0;D(6{#mr% zCpf-E|G4087hI&phB872rwsP4xL^;LG2im-)^}zpc11utM>IYlrAnB06?gv)#J`!*e5Tm zT>SB4I6@lg{`b4ZdJCW1SpY$+uH8Dmz5JbR4bigwk|^%CT|~K15nv;J?k*x~!#tmd zK;Rz+b%5NYl8zw8Xs4%+JCAtClfWV@@_U9&Bx1vA5f{%JX{qaME;0xP0sM?;iL6#rquBx^$|0VlRggp`xaPKvc;2$$U4r5=ePQJt z@3x0?c3bD3{O13+Z}}W%H5F8H;pHk`gU?$!7WRj|;6M(iZ-TU4aL2It7i!lU2ctyj zmYXT4UwW>MxUut`GYRQ)3^4r5}|ZT{PSCb8uoT_>%ze z`{}j1gFC6btTXO&I(y>Mq%?FDwoV&$n{B+4?gvootvIc@slC$UX{+^n9X3(O7{hS*VZ3f{U(w&-I`X&C0*u<SgD!JG#H5dKC*`h5KaDvRd-eJxJXuiVsdkpCF|9$183<&|L6pD5p8k9v z+D49Vi`Z@+WDp{w?6;xL@}>Tq@6Ldp3W<2ab1I*m$LYdHy&!CX1yoY# zdf1U5d+m457zl-Wq(n{8O50MtG5_HYi0rMvb8DHC>cl^elt$h6PXQIkTpzUyEtWyVMSw>07Ta+u$dhMKZvRFnTfr1Q8?gGf|Vy^ok?4<43yyRbD;lT=TG zz^dCGX0P*0aKgvoyhQ%kiBejei0wf4`t&2|AxV9o^f&wvY`zq$6t-gIl_s!*%5`(_ zE1F%2ydSYPE**N~tEbkC?ptD8>fw8Sz@fe0BTqV~_wnnAn)6FO!l)>3Kk%&Eb(WAE z(?L32+n?dDydP{#qqubzgcIlzc-OkJr$8OBUGhMm1(HkkPdZme^b0By z^A=SSPOCt_%kP)IHCQn-tnDlKU!EG}m6w{kD6R>-PJ*7yRN{NQ1()nAADg@{3!^zC zn1yA{VsMC~BoLM1%mL15_xNI-m=ZiUK~a!l!zfo1jMBP3XL;PU;&!IyQPS{gy|0 zEG^8Y%W{nzDQ%~?e-4OW50n^E2ll-6e`C3jYyEJe>zTPFHm|w&AJb0EVf*5GWPpb{ zwd%V!JCma^mv4hLav*G=^%~H}MF?F2yh{709-GHK3I401ACy&v)t(}qfzLf;%eU@L zHAyz{B~u2q4qPU+hg_@Cc#!e!KRuP21KZE$TKd%~6eG_lF5fuZhRUibrDLlrcq z5Gl3;X+o=XRVm|eWn2yWg?!xY`j?>TQDgqUrsrVe*f}Gr`N@K0RkKTdXOCXD)T5Z) zOtq#$3@s4G3}wT>!W*9tW$mz#9HQw~JA9-6yrFS}UL|U73t~2wWy~(HbZ}jy^Cp04 z6M|p1%|tU&fVhV2@oVD!*Wh2xrtssk5L%byzT^8Yx8STJpK~x8F>2gy<*CgmJfCDs zUoRBU`iW@X?=AlqA!5F9HF^+g9dD2l;}=J)s;q5nl;(Ey0)Uqq*B*u&oyyA~{g5AP zLJCp0db0i3V+G~zCQ2>sG)cy=M$dhX^e{1%9ooTX|FI3*4g1eXXyi(Sd`nvkTpHR7 zxp;!mLWd*y%5A8SVpRVSP1+W1&al!2KX$LR!(z+(EdmD9dFOWI*?lx*zuPG;X&5h7 z2VIbv&;7_Ee>vch*22bl`k_c_<-$1PHRe`57+ZlW(ul9${rY~QQ|kx0ebrm*u4b9T zC6!xIjw**kpOp`BHmW8wBImw81pZ~z8C^2IZe5+QZgoS+M-WAOq}Z;}-gJh*^twCQ zvU4f7A~dq%ItCCHBFp1f56l5}9G`m40ntj{*o}IGdEhOgq(vH%M+X z&yo!WwfW^NdloU@lE?+w%4DyYrzW8@Rs;sX+j5?JFmUYwhH zNtf3JMGf@k=J)zNM~8l$JO8rA`JTO6cV2^GxzmY4r8VwN!`(S&eD>umk8Ncm;q)hP znquCXTXnCHFT@@O&uz|8r`Yo4M*JXsZ(~S>pNNrtr)SMoyd`R~$!NLs%@DcA?czLX zh;|8@k53`D?qUp~=W#ZqG}ZPrx`#uc+TU}fAmGR43*N=erOVOlV{=u|9k0mD_5#Rw zj`rtybvT5)IpEOv;i|upBnnf{!2IJqrVn8oHh5H_1+q7f$}| zvIzFA+Fj2U7oEUdMv#WF+H*Loj{QUIa!QJGMTub)^6S#l+mTVn_#pmJC~f#mD9ToZ;lMEliHj1})<0G0YpKDfgcx@D1(urNI*p zUq4c?5{dg=H;8A7%VhV6w&y`ISbS(Fg-kjm#ke@c-CHS$^AKhLb zC=*y8q9Cu}X3!}(OP!}j1cfyCaam|gs{izoFTA_$8fn9>lDJNTrf<%X(l*B!8u53W z@xAA*XV6}Uz%aD7Y!IDWH*4M35~p7PIt0??Vr58Qj(+tGB61wXxj|=}F3T+Kg@m}# z?`7M$uUWs&V99-4DygCCBV1ZM$pKZa`mAMyR2sVT9?^UZ_lANx{{{RWBB0@UekYqU zWb?2(G0!)QM0S{EbY&RW52TbMuv6I~zj;qDm#*|%(lW)NpcBFiNQ7KY6cTl{E?J`Y=2U3>UN&wWH&-7T?0X_% z0Tg@}p5N=K_7*rOG##gf>aZ(ZoBK^P!C)H`OJ&g@zybK+L8pSDLErysGgjb}(7sI< zizSTrw~GH@!E`@{VU;=*_d|OsdfqBIJNLGWdwsN&G3=1L;%jgS^Y*}R8s-J-1q4g7 z3V&gyw``Vj^GJ;M7wn+k)xS#99tD0U*2u>$g6n1%Elldfm`3Xn)j^<8!{$X@3pHd< zHbSZ;Up|r_)OQeOpFZED`s>pYKCEMt+|8li+X>i< z{>Hj(S>~%5TZC$wX|l{A8jd*6=mRvmYxOb>4BHIiL74JdIM-^SmCsb`7D``VWc0Z)HV>%?>7j%-(26<2FfmuV1 z^!Gd{S2#j#KcVmOc7KRqyZ{HEf)td{I1;gV3n^8?OkR_RZq8=&jpg(tfSGy$B82tb zm;W%g3Y?<9zS)o2fE}FVqv>NtZCv)1ciDS?y1xHM$JdWtDZChqM{O&Q_uk+?n{B^U zw_)r?-43~Ml+iTzc>?%yK6GLE_-4}aYMh?#e%12)$G>$-+GK_h`h^3okR`9Hxb3mg zABbU%4dNmK5r+n_9<>ZSKH0RxcUNb|4F>;0z0Qa%U`LU(*C>(1Lr&C0{k_g`#X5-o z`EwUD`RG$U^CFe4@_zr+E@bzwc_lCW;6bpwG{tfc;Jo%4>z{AJr(^j$_rsiA_;Izk z-(c!;7)9p^ny6MfqZXMYR0l_BAvtj3QA20)7F_l#@`)vIH7>otU=Sr8;#KR@Uk-cv z4u-4?zWe$VS{HPiu|Ztm^JAJ$lR>nu#;B<~LIw}jpH*$F+*g1M?zc53ue0c^xVR=J zg7Flc$=L^(_c*`(2u+vXm~5q8g0EeEtbTT0;01$5f9&jn7r{pC`s`b4V$H61khYoL z7z#npa_Hz+a4zkb(`sMND>2WM6tdC@v?0xi{+h}sS{<^ioJt_~yY_lG3-$Br9c<=P z$~H6xkEU#tyeF1ye`l>duP`3H+_%M)UuN)OC5(^JN7)Ejmdj_+C`GT>7x~^n!{$R@ z!F<2g%CTN?zX*TKAt96E@W;M#niL`%e3tJ5s_MWG35KBfuA=$Cg zaA^1!J+d0``bQIh?UmyTXe;v4mg47O)OldU;Otm=n(LKWG2e1`oa9xvR(T^EBGM^4 zV@TOwehl1Xu55aX`$tdU_Y4yt*$1RDV^(hs7PLINWt-6;INZx<{$CAgSYm0u1IK{z z6?9|l(j#q|z(6Rdp%1-dhC{d8s71*PetPxR?j-feVH?Na|I;~G ziiR!KYW~IGmfEAA8{_!dWixxHS3WxNw|~A#deMDhv0kScnRh>OwoCIHoR9L$OwRyK z8N(bDp$NW11e{gRFgp&7PP`PBocc&N*>;xwdg(Mt$on{nr(P8X&k}pfdlX7-uO(QY z+E0$uGHE`j+4H(TWtR94M-`n;^%e>U3&?J)T^~rmQ|Sl}tqHvm6iDq3D965k*?H}$ zX|RQ+rTYd}DW~rK>51QJwAo!R`ugtuI91QFNTv~n+Q_3kL@TCpP3J^?U_kb7jUN%! z-4z!1b@{8yYT)1H<5;QN3I_h?#3juW~A@;?DV8c5vZ}UtF&l{l9Q2AWi46 zetv|{OLkjwteeC!cfPZqwZ6JzlIn*pBaByOpI2u9bl^TOa7DKq3Wh^DZuNiq2fl=! zu8X7HlT}IZ4;qFiHzbr*xBi$?mzU2vK32MN?DhTvqc`X|h^n@0lh_@_-OY4%nL!_k zn}bjGfIfMf@}J+h(weqoM!*@;{$ z>fXMM2{S(rmziDv&S^v&afr?c+dM_iv)G;Jlo%5>he1)Q5plmcWL~UHtIOqxI$*^# zrCgZQ##6)AM0Z@0?^NoBjrVoJ4NRu^^-X`eYj9*(Yc9S1oAPuP0uHDQ*y{jn)g6Oh zcE&~%by$JCpRcynX4k4skfREkqg;V#BH>TB!kMru%4BmvPncCFZIczgpA($0w4`{O zd1&~^R{4PhJm~6p!ddscj^~S+xD`ZE>^VFuUjgT!*NTd&3h5YqQ)tYt>h_k9o74D1 z6e*u27?|!qTswa$YY*PebA2Y+Pg9dyq$XkFue@ZR$2GMe5gr5eZE^$w={$)a3GlBC z?j5O2@u!hKn_a4XB@VaBG zGR1JrGWx{=N-^c&pLZY{81z71VaS9H=~tF@U^*%!Pn62luXPEo;?eMS3ymGsr&Rdcp z=oN$lo$#mK25&pP#0J?qOn?bVc8AA?sQu)L!L8X;TEQ=3r?=g0e23luUnN09$M8IA z!J?VX1^18!UW~BSwy3W1ar4RLVW>YXgl7+RgS7MfJI9dJP=@fRs+R6jCl{^dsk!7Y z+YQ1m2gjY1C|PofLngeBrH`Oa_lhUqg;wMAODQ z0)sz{^{7sQdHssvKF~Mbt*KF_(3@_rL}$6b?z2uDV2g@s?8ZSqU&*wn8MA2tum!>e z3x)9RMKidn=bwGfz#ro-d3HSfan7~nkT;H$l#V(xF4%0tThngfI?exolwu}xy3r&mlaUG$zq=1 z?nx?~yAa{d--p7!p+|%#->Iqa_JgT^AwHr@0L1%jgyu@7?;nJg#YH76|2yk%xJ;)v z|7MLpYNmH!!2b_l3pYp+yaySaTZ7hB**+j@Q@ex2{pV$ZHf_a&U~TvCUu5>$2+NEf z>NYMH2eYlB2o!-~~Q#ZHi-`yog2(VeX4dTnwCwS>CEG=1cfOl4`!Q5mn&Zdqz z&J#4dZ1S&4kucrm*ofh6AdipVW)IA);{AL5R5$qd~t$ zKGt~aei-tX8Ou$>QJ?Q-dCM?A^rRy^ib9V+FugTgw-JaOrj4}tj$-a!;-FKf)7- z#F;)nb9hszeu>FFhaBPz#d^Q`x`IRtaKw5%pkaI(a7R^ww)FMWVV87wc4@(OzoGU& z`%F7t7epFo6Y|@`npKP&+0DBM{(%>gj~@^@5=F0Hog|W69t0H$(ct-36F2fffD_c~ z&5h~NU$tJb|0b$PUkIsFJ2Ab6s-yLp6T?dmgCEdq+gbR=E zn@TV<{)bZ?LY@OhsRf7VDeL3TL*>KuuM=z*TyX3v^&CtOlE+ z*!_R(y$g^lSz0F6J^h*y!_4*!b7x>?81?Pyx%JkoN-Eu++k@);RK2TI>f2MKq?D8@ zNvX6UE6`d4@~SD*6Ua>fms3v%xtg*9E8I|c-On^vSExP zY1qooo*+%|Q19KRZ;+bmxC=yR? ztZNI0dSzqc6_kxdvg#_r&+6x@8rzKqm0BA7R{;LT#?F!2 zSlu{=?xwjF$*Q&H!D{7bVPbP-g<{znNvT>nMgecFHnYEQ7Mmz*i4ZW-YP2rm=d(M3 ziqYDfSlPQ+yV$92hqXkjhA^tdJ;T5{o`mD2T6`C|F>_>tXqHwoCu{mF(%7nHl+|5X z#*|p@xH7yZ_H)~72P^CCjW#qUHrce&n&rmE9ss#6sD%E0Q_t1=IsYJEiaCi~dIeZ0X5ZDxl z>Q4@aYT2MBMOW2)_WVpmp1R}n)aja!bY-JYD^TpC^J7(>T=zcJpM-n?}Wlh>H z?5=`j9U|TWSrWCSP(Bl?ZEUP2;H}nPT%2gl>{>Pi5Q8IRtCcBB?+Pxh&;mR`&kojR zl5)%lAeMe9mRt+MX1ZBY(sHJTR29b+ls{kII0&sFa3EQarC|z`p)yD}izk`p{Cs+A zr;uwTH&RBba#7bWW(ybaBMRwKdb2SR+sGj$*2Q9VD_@GlTjiB)HF^@%)(#8h^*JlH zK6Eg%x&ZQNl}HPgo?ywhAxApJP* zYw!zNf%y_Q>T0OAhsdqi?84^kEWDVcS!H&YT-xAW2YrH}w&ABmvbem5V534Non8tc zkWD|(fS=uY=R$o8QAcEuwIbNbO&Dv`2d}|`Fr_Lj* z(af@8oK$xF(As0GZ=dE@n0er_c43G1Q{9(}PX7cBeZRG6+KXh_&ymeB^Z3k+IS49E{L(S=s z6IN0?7tu4#fS#iY=^Hu?%nCtiD{EEUl1T457g%f|owH7{s~4%{jDeh0YDCJzMOFvP zup3RMTkHE<<+mLFq6ALHJMP!gWJ6bKQ*21Cldb7Qe+eSju%vwH=IE$4EB9Aj?vpa{`nZ5GL z7W^vC)7#2y9FCZW?NSMf4+Ll*}3m3Di+h^+fVL3R5pBGOG$ME?% z#~1BjwY{@*5U!|NSw4im8K%1WhP1WPSYDDA8N;QuKM~wVIeIc_| zsU9v#Yk`YKal2JpSzO(Y*7sK{>YTRF&Jzs!Vj}{%zYNc6^`hKB64>B}Zf>bmp^U*U~-vLojv zwvoGbEie~q)Kck0_C$%9I7uuntz;q#vFtt)?v-cYqFtJqKt{3Mbwz_B5@#G%cO-Of zg(u>xNa%Mlry?dF)^(QOF|mTgrlo~UJUzE_w7R$*Twjc!mcW^IaxoW#8iMe7G9Rdi zoAZ^G#q+t@HDwpEvyq)ld_R&qS%DuJ2AE$2sO4lMzq}kjh)0VPD~+{@eU^h0K6HD? zU72YhRA_cCTQ4T(BT_mFmsG8_zeRgCv2w5$yV%j<$bK45ZB{oU7uuOt3FfuhDy)=Q zX}*AcpIEp^uON@}@kORqi{)X+IE&XJ$P+qotj|HEjjTC1;{`*#$R(1Q*?cP(oY;|} zD4At&vRpw`_`gT12Qw9!sg}3YCKjA#xVYXnl%sGeRB9<&4C1011w(4=Y&SN8wOKiP zZk(47QY%N>hejTLVBu0caMC#1UJZu#z-FKWB(9VY>9(*J$Q;CG4w2`177<4K)qE>a zo25L{@E}KdjHAYhdI-K^rVyn4yja|7)lSm6mBb0fZ?u)=`J`SrSXd}FX3KjUi)>04 zE6c}A;aWI7Ta0fkAQ^!cY8R_}spAcJqY&_fmvn*dvxC&;Yz|>_N0ofOxOq^H6e5|} z`V3(&u{eLQr)JWJ$jfw6Tvd@cD@|qeJKy zt7qX>MxIBFgV6TGZdN`IvK&z5dFcYl#9-c7USBC5RZ0ipCbFDp+dIp;QJBd|JM)yQ zbp8aH_BR%n&tb~nrW{RYQoN*6KB}|DWu3)!sd^)7l&jU*R33?-ODEBtDCHJur?;xn zxk9NlbI~XqG-Wu;=T|dQrm%mQX~6uLZZ$x|b4DvIx2Zp=1!nBSuK zCz49Gx8@p~?c7!{6bhH}2*qvfgfdd`)n6hNZ`wL2yGl4_0F?aKvQq!)hB!tmurYqU1-Zb<&7bci^10A<5aLM!rOP z=p!=^l2$A0$jY?2w-IUv!ci-(U}JufaJie%Ae1$cyfuMPR+L#m{`UQ&_09cYLs^{A zTIVaNHGO4&W%($2nAapLhEYD3v%4y6HneDZ8Xd^7c{odiQhSF-%UKB#(G_GduBZz6 zNgU=I>9ir&59M4$GWO1or1`KeC+6U5I&pTyWfRk9@F1*on>6P^J!;P%A z-abJI5+v{6grh)IS%^i$+fgmqsLRS(;9>%q!9irCRM) zi1zu?ULHx1PxkfYnaD!zXbyftTnA~znI+G%=h^MlcJL&-wQ`YGL;Ee9n-=N?tu0MZ zjUV{HfbEFHGl@_mIzN$L&xa7u4;xgXn878^@}g6nqhK2KM$QrJxSb8ijX)e#0T7-T z(1O9az(GF0+dQk4ckn9C<5IbC(k{$p$h{-GeY8ElyR?9W$uI*1P{Ria{2|<(C(o;D z8ukg~Flz>jvlYs*%>4C~Qm%?f>X#{`lNLjv`|_kLb;odbYh=mi8^XY_l22!|q9TXYe=k`p)i2W+IN8VSOeV z+^MX=6j0i(AcMnN5rKJ2&4LP#`NQ_X;+A~C&LM*Jnw5lQQ zzEH@=XL8COOjPmh*%E>l@$UTgUagosTxjOwt-w}kZ4neyQ3E^hX+bO<+55`cT5dBtc-OB)LXq&`AWX-HZK){&Jgcq}2VWSjEgsPdMi z8HKeI19?}srG-MKshYWnm$S&G(ntp)+Y_1N`Q`S~Ec8E8sdk}MC3(&;LWoB!1+xZn z|EK5Tbh#$BNq#{Bu8GJ_;AFQR&gB+$t%f}e(Njso{*+HPp-V+rQZcqTlWZ{_7ZxAn z#6X+XgwApp;>bcrOX6mt1^YPB;^yWODCyvBP`m*VB&w!CXH{&)k7wa-ISY@cWGWIx zlA4PP_=F<+7V_02y&eABILJnh<|0s6sdI7$Ze;ls!mMXb+9$aS!_A7XZ&>M~?tLnQ6}u>p59N!dM*7`aFgUbKy947zS)SU@at2~MoD>uZOl`Mrv= zDo>!&%!!%{AY3yi!M`L&^_oISczVJE3vr0j*75q%!Wjajw+>OoCtZwguOOdW8b0vB zv+CSLm2%xjlk2FvbzCkZQ``(P2OMSrdrE6IceA0~#4NO-7wvXHTS>-d_U2NxgSi@t z7j0ABwprv3tetDi(RjF&*lLE9IeBvxF;Ve3(%#H$K!=lTWfQuES`AwEBx-UlBD;JNzTeV%3x4|5m1q=N&vUi7)~srw z=p3rTEG(|J6r@U;3(ps3veHH?-?)gO=3)kce<;6}CpW;r{_!#DkEJ*Emr#?2>g}y> zqS`|YrGKQkq86GUm)ni~v)bWN`J}RcfLv0^B=SDA_2RCcs=)@kzXYB1MMw=UEF@PD zg&xSwA)$Uss;9QAWhI`ks#C%m-g=2ZA`H!hmkYWr!V!){Qavb`GOHFKMY z0|;axW$xq~`=!P6_-uGtMoBrS#^9u0+t^z|af{HIauAB4m)_ z)z+cH3oy(~)V5G5aa&s3LWQZ66hz%Kr5MP?Wi5p4`H}cscoTx&PPM&g=WS zGue|O@QgZ36Nq6CAe;Q$ImL18uWud)jpah@aJ`^J8kOT(0`+uY8G@g>q6V|8SVVY= zQsICQYGIw_G74fkl3?j~xAfKh{MHio8M2Ebn(Z6_98!(1gZ*Qqu+zXAmUmF!263Y0 znX~i+Sg-@sT-Zn|4I`>IRx(I?ZRAtY04jN8^K+$mo7E4Z+IqG5ydJ?{O{o_n!Lu!6 zZWSpJQMct7Y1?HAG*2(9O$21t&Kia7`NB$OySh;ZmNV-!ft9lpL|E+aZ(ams)Q^kS zP`4;5?I22LQ)}pm@;W@)E+LoV1j=P95Rb9^TgAD>Rxoxjp|x8q;rL8P=DS&Fsa76gz>@5H-bM z7S+k!yGE^Y3E^-VOnc! zogi<79*Asc3jhMOaVB&q&(BeJv0PgZ*4E1Vtp22tG4e=PfO=sl2$7G#EeFn>@;Op) z9!)e11LadfP~GMNh_QmP8KpQkS~Ia3ut$s**Q>FMN<5&pkPAsaz}8sYMq#N~3=6JT zAhlu}mrnx9+96W@FCNRAxmXm;x_S;G722oTu(0yPa_!keP}4x8Il8m*xY(8dwY&34{^ znSZOAucMfbfrL~0#!4W+1BMuiihO=<7d1f+4uug$ljySS|7xmGEN zJn&3!wM{-+>_26hbw6n_W=E?-}QiWGn;A^s&ZlmZ{N`k>UE}iSz@*b?2bsVA^ zk}IGN6VcO3GjM@ezq!qn4%rd)i;)sCJTY;cw(PL=vodxZ_-x8^i@cOr>!p?5`L#p~ zF`BE1G^MrK$NoTlOL$Oh1RH?~l=?a+7!UIDZenX8%<=`T>j*VbO0e@cV7@=uT3bR@ zp7}k>Wzk%cPzI)=v|5NA*g_q&So0XQZO%~U=om1q!)Gx`IKupQP#O3j7Y~KXkr?(! zK%;QEP!zIt-~hT0=qN)4)ZIb89q7jk=2==<2?q{Q^Jr!_ME97Ysxz_u^W~LNd}ncS zF31v&)|Qry!E*ALA7qw>c(kS ztyhi58}F+Z@0ISMm*V)kP}lyi*>&wDS68JCG?l~->HWL({+-gf^-^)PI943rc&46_ z-7o*y-L98AAIKuc{nTjJG22cu}dMK^CsDc)P90NN?@;jXadfwW2gK?O&8@ay6$^3x143eP-3Fjs*OMU#a@@ znu-tTlYRipT#?HDoT3|=lBp9|lm4dcFUW?!A(iU@t13&HU&ZkFgLf+ci;d%d5FYVM z)tm*U|G_LqKG52JL-p6nQrjKPC z1dD0DS7>L+dDbVoG`~1KSwG$>{aRNC?w#MY=ZhyeFkgD%uROyxr!zXneYx|AtM;HU zfveki%%$|zJM7rAt>1Zp?PNY5>zh3x5P&(@%=UJcAmZ7(MuesOfMe|7W$?>oD z%vcxC-RH#3+^#)$TfY_h)+bG+ocVX2{8%1~;70HVY?{e&EIe_e`OYxW`~suU0xqlv zeF1(yJFy_@9A4*v8{DyBf-61o8$>|GJ203Bngy!zSGq4Vj_b=m*f{T8{tbhC?mCOm zlhpis?u+1EyH1@6zgHmgSW7pv*nQvbxGyNq8`d+WzcD7JKbM+n3d9kWCHL3Ra`|$}eakBO}S$wVdS>mR@4Ez9j zKRszPD-QA;hPv}R#A@l_U&~;YiH&0%>m>UK&Y4(fdJ4XI+P{{Q(XLIrZBs?{WPe#X zlfkO`iYOumBI(B>_+>D9Wk0ICs?&bPcuMG`YJfZUYqA7}(_fTye@#MH3}i-a$kS}8h>2Vn!psCp zHqa__&wPEc+AwgW##$!VL38w&x*V#{f*%GG5$F;By znNEfvXakT{D^M6fb{*NnX)_4|2~>F{tB?$;=6yMmuq!aOlE@440KrsK8w#qKLSDqJ zr`I)^govWr1kEUy(YZm7bR_F6QzK3G5eWeqz~Rn7Y==auFc~s~M>y;i=myHu{&`tP z-w+m+3?z33svoo>G#zBDnPT{|pg!oJDi;6@NX;lh45hjb>co5TTTS3Es~{j*gTo%E zn5N5P)Bc3d7Vg=IcuO+;Qn@L$iBvJ5OvS=wAr}e;@zxoc^g$7TbV`!Olm|4vQUxac zd0C>_>(C7_;z{ZO$;7TTC7@~~I2{~uXxQ^a!-V~Be#d6`iLR|Fa6~Uzi^=!4{oUDT z_9OK4GhIFP+~kby`8&=I=U}iM{!x$3Bf#$Yv&U522lYAZK0Xun*`5WvA^+}mpyx;Q zstJ69*De=k;QhPD6uf9UM^$muQdHoE8o~h*z#B?;*Pe|)&zFaDn1v|~tH;~S>hrgx zEy8qe-PLX8asTEtPME@NdhvkXeb_(brltJccRzNMfIk@~33ztWnQQ4Ee-tdV2d(ws z*qqZXoQJ7v-2~3Sool}60izqQeTe7>UcZ4->MA1Yb1pMsM&i@{;NlYGYyTX5b}#s+ zh6(Po(@yWDA9%@g!TWPs0y4Zft(qC_A9mps!#~ka7WeB#G|%Ox+9ZYAK_uWhj#@%m3(BVrlFaJ_TLfF)spm6EsO_ z1E{l5QnRr~%At`d^3Gt622fCSgAc%RN(3>+3^TxONFL7P|hnj*6 zm*3`GH1E>rHrCl~^}^?F%|5;F<1cg9DhJWdZPl4Qwn;T3)IRQBR#u^Khm%|d-f5Ry z&DfzMli-KkuL~Qz`xk7Jk6}8dAEK?Wy=18KX!F`&Alazv4ETZvMEy7JN8a$jr+edm zk#F4n&Ef1^OICP@B{_v5Cgh)>9Xv*YgYd{c2K1B5#{S*`$X+A1y{B%!z3<8%%OPy< z-!u985ss+V7}P%SIfZ9x9t;lMwlKEUH5l0K4bNnS7P*a*`J6E%HjUz>U$1A2KC;rQwK{P^;Ov;=Fl<+l zT2LUFY&y19jZbJ>d1azd7y%OV{EscBBB;aQeIUd<1?ZaqHtDOlJt4-DWQ0LA8%j zM||qhd;iNvI8L6I`N1+=2jEWxqGAc4h|z$ZoV)gecoD^0ML>ZDA|w6;VwNLm6kbGi zJa;5elhR;UKAFyK?3DgG9ltxJ&UH>~Q3P0P4?rgZ%7h3YaRv$2=$KcX*7^EyDsBI4I6_@cCO# zRyaeq?h32<6J4w6Mcq8wym32}3zN8QPuHG7{aX}6HD;D4L|xUEDIz; z6RZ2^+uXl@8@fFAo8vuvXnpXHcDI@mMlc)u$0k2(cFPD;6GgoqO+Q3rmj#w7N;{JD zU%GkO+ktV7U&UUyrHNzyzqYw#&kO90RkCon%ooz52jX(JACWL9tiLS`J`#N2USRzQ z^u&6AllZl+9qkqrqGH#B56y*&o4)Qw^dEPD*z?eFllI;p*r?@GWCDnnlLBwRx$OSb zNa)>j;WE&UuzTFn?-R#>Opj!xG9-nm)FDF>8(^9XIHSf{3m;0VC@LZ_%ZLLcPe<0a zEM>rMnAQT};=kos!nXMQZM(vTdhS7gU5`hby93PF-WHufT6@6Ke_;=tZR0Gg^I9ha z7kW6{PJt7HL_Xzk%TvtTb*ip|MVr#I%9&zJmC>a7SmF`H{AKw1vmjRIWz#tqdIjdB z?|#_|JBC|y@jiIoU-;VBbG6>zo-i?uef>I*)C6T>fuG;PA9PL*h;h}xeL^ciJm!6W zwVox={28#rOm9(@QR_?A;T^0gGFhSEiA>=IP*zkV9b1lVzwYMzls~}{oE3k9gJ7e4 z7Mr#gEhp@M9@x)3`w@=)SJ^9$#wqm=VgB2MeGjJ=p(L>E!5oG-P1SGXyHlbSke6qc0{)E-iHaa3KJ6q&l+Lrft z&enTDHHAQEAGBfQlMcy(?FWT4TqM;)nuNRR`aBkfoj zh^RpshgfF-n`42|8c9_U(%^B9G_`P%rzmtx7xFLAPEqTs?J>_}JsU0fp$Kh1Stzp# zGqR2%x*c8Bu-6e8Qzq>k#;uMy)V1`^EvZ0=-^o$2{bvWD*&c1a<+giJnDOoVy7oK- zfo)`v!yrIASbGnM-0am$*74$$)c$VMJ;rJ7bS=+73{cJ?Reb3f1N4no9%W&EJsNpT z%1gHAV+yG!UhaIO+x4GL(w6Kyshle?xFR|&ow=M~L^*ZlG z=scFr;bb{e%G25qheJA5fou?mLsnam4LfU0XFcJt2gDa*LFzifR4MRG%8-f6R3}q7 z8)GhrYhb*D+{!|Bv6V5IHs3eH?9yJ5($5j}cS=8doju4fpg4rn!GK$1^NR1`BB^Wq z7X2n?HvVzSUr#t5ngg`7L1t&>zX80xIMD&(l(kV8lX__hU$IQ1b=&XVTGK6J`3|d! z!B&Q>{M@2w!)QMNYJZ~(wJx&UdA82i7oY6+*&M&p_2NMH*>`W-v*H7@`|J~(Z@n}L zdr$;}TD=(k0N^RKP+9+b{^3QL-_rBDySII$* zKtAr6R}$n;LAKOnzE!UJ*@m<9Z)oq_DgF6%wh%kkKm8wiK;}NypY&BuYjlGZiI0rN$Lh|3?%Bo$lcDqdZ`}GQ|H7|%04<>NIjE30?$~D=g3>o(bRg;z^oOMY z8z4&#L)5SSvB#KhM6Cc(Rkn}g)3edpNOW{=tj``sYr^VWG>YQ5v(eG$Sl!vyL+j6; zK3jm#-hKhT`jL-{0{pJ?;b16g-E-u4Zd>&beSJK=KD`Qm;KHPTWjayM zDU<%K>0|bNd3t4f1G&HC#-x9LIw>JP17k8%)2aGfqea80-46sTb&aWsnO3zz0181U z4se~G!f_UgngI273J_wa3$Ss*|MqkTx`xw*)0xm|FoMc|k?C44KZd*zq)?_yn^_x$ zGYC~=4sqTH(uSlTnTfMK_>n4x&pgCOS%9J>zS`!gVz=% zP^_8<2>K|d1SGct7B#G(mXw)y5KnN{p#CzoYNQS$ivZMRJ|urI@SQmpxHxIa#eF;K zIif)>{VV~EtP*BEK#SKVj&5lv(-dgHkUGSgexd73c&QaU+Jvr+asm1qyzBv~i}Nl0 z2=OMCLZVwFu651v>T@sTLHXaD^U3GW&gX~R8%LN{_b+{6{!RhrZmq<34-@7~G=W7>I_vR!qmh^Fo3)J+*VZX2ao;@wL zX1|OzyQOsfr{8K{md@k^miz}T=^Asg z-Wz4Yxft{f>|{O?XJ5LeiDj^-NuS-*{3pYl=BK=-$(P$1xYN6x$CSK|e)qJ|=Kd{q zl#P~K_2L10f5IZ{8t3e}3CjZ)VKLQo4t24zHi}F=z~^sGbUyN5 z*vG}0xIlE~Eu!O^?*o5qIy@H-j`*QSGAWH~{tx_f)4^CM!hUz;yq=We6PSyrPH-Qg zk9m&I57iaFx{>ut*8l@ileYmWdFemf^O=wz}jP<}{BK=R3fHlue_4^4cY6XEPkV{PO1*YkTL{lffXe+CnWq`Mm zJ28iHnA)wx@fs3R`bqI181G@%cdE1(xT90viq|IBxz`EXj2r{WQd(& zUAMaSj~9jaXcM|N#)TQ&;AIa;U9_R6nfA16jsqR~5$DkJ`Ljd+k<0r=n3~tG9od$s@xaF+cJR zoKN5*Gm2^ufG%2aKjBdv^gZE$umu0hb+#?DKQ1|kR9J$4%^jLSLz##3 z&$}Nu)GxJd=3)lP0xu}}8_1UfXIfIA5RD6gC_=kfL%nbonPBGU7&_ zw#*11m31sXd_-Hw!>-dN;J&5Ee0)q1;gq2V4$~FN#||!*neK?lTFM`!Yeu{`al`-( zCjT?qAJ7J27|pa3(12zRc?4scpilZ`3I46D=8J-W9;;K9{2d|VvWAcXvNllCdK$=k*6z#AT=%g3Pq zy(Dbn|HC>2y?99S=KDEVc7@me#D1-KHR(KXlTKuv&%-atZ&^Dz3FUmX<<{@qV17;< z%%tx&rLwL{-W_O^Wd3GaN8#w7C5L{ZPIQXT<`i(pru&J^w&tVa#X5elPG`V3HA0Nl_41 zZhRV$D7gpaT|mUjHMYC(+92CK^*$HlKtfQvLKRLC z6=2C-msM?SuhYFsEl8r!bv3K700PDe;J)?mNcFM;c`I8+a4aG}pQ{UtwJ-ap#Al%1?^@9KV_leOiX zEDjz!SzoajtG`66X=AJNmvC|(+p9k{5f)#&1~n00bErx0(f-^my$G5hCl=|qk*)w7 zI;!WHikTcfAYhmq3+a+1yC3aHRElDXY$7vw2Q`;XTEKQI5e**_E>NI5D=AG!9~zz< zLl&(GneoGdGKK<-6=>R7s52uw(8yEiEk8$+C!KyHTGE%wjPbkPql#jZ!22e(4CFrQm?bnO{9X1~tTa2DL5FhlClV@|!# z8O8iu(0A!u`C8?XeN-DQma#D z*uoTU&P%O@)EFlaj0Vn9!3G&PONAtcI8q7?A&J9aDKtL1tra8`)+Ay=KoSc0J?r_K zy{x6L&=I*)`fdE_CuLB6iM=ufSqVh1^{(v#`#^QxJs3PN=g`Qj!l@732f90cEcI!~ zP43+SJ0f@nIs|5Hmzl3vH&pvzz!xfOQ!2bo^4nt#a=vG&9eultTkJ zR7_&_N=3&Ta2JvO6h`QsSO$~Al)iG;3BH7^4~u!}A76g-y{|Z}Qv1(6^USOM1-ztW`SVzfagBOZ14u`ms4cMvcxnC4Crf=>M62T8K-&;Id!{Ug1(8;Ygvtv_o{ko zRS-C3$=oOXTKJR|@SCIzQPopkYQ_ASL#gUM->YQK2j>Dxr&Dz)lV2&=srYPitNV8L zw%3;y4o>&Bw+~Ji*Y@Clc6U$K>t#c~>swx1Iow-1U0W2M29_1-Xc_DeS}Z0G64vjF z;49hFz+InDf%gCqsOr--WDl5@TgZZ?k9H3>cHig6e`r-X%PHFB->>S(22AJy<}dW2 zk2=3y^|l4H$GiL&A4L&iDUjg)lqN8mR|=;}&U!*$?%3o#?cSJ0V4fbYr^oB*@#05th{wyms+mgr zv$&mm3EIWiik~h0@3;v|{|x_7jP3W}4#c(os(r0G=bj%QCuzTe#;8t$jE#`+k)F}tM$c7B2fZ%; z)fO2#7w;a91kwH7gaR9B-rm?&?Inj1_Brdw*oiv zp2}#K>;l$ zikl!mjyzfh!`2$JUM04fUME^O+ImxYYtr9l|C&Rxk8CUeXpV_TX1DxR6P8g9)0@Gr z-w8)@+4((W{Z7M{WyFBzM$2mULD1n5pQ4v=i~k5k!$Yrqq#2KJ2J+Q?J&Q!^N*=~Ac zOwpQb85fuwjHCt5Wovu3lfsdkJz4A~NzArSI!|ar$!*x}Z5dEm_GhRUi7^=;H&qH0 z&*^t=?i@uJ0d%kn$Y>*#Ss4;bAvs#Ff;#Ao-C#waC31L@z=fEaQ0_=dJQjnLj2{%I zf!aPa{V_raTtLkHl_bzvf<8A)fWWcDmWbMrP$y13qo@zXc!g^vJIljCvA`}2t4CZE z-)HaFtIIBWvp|E;pB+fuld(HN&2q2rmusrGILlL_T}QjC^>Kt-if! zg=t;-$pdOH>o@OT;NoBIyx(*u(EY*y&NGjm-T3I0s^=GniH{GtPpSZI_b+{6?v)bh z*M|x7OBOK^fVCfaKq-XhzGzY(&It3b_SMnhPteWpT?b^w0AID|zUgXOCrqJNE3a9h z#-Y!7on3mKJ8^AjOXk)nP=AE z4auB~U<-t)o$gx@mI7F8e8x3}G7_~q_l>0TfVhs$x*wKt5SMBbg?{1aL@e?$?15mD z%_s~iJc=3#c`84_)St*iXE2?)V;(f+dzF;6;J;}&^YUcv8R@XfP*D6GLf}{s8U`Xj z8Seqr1vv+>U=FDCp)F9V)A0CYb_LwjxU!iaC_;^S9YZ!T@+4qZQHZbfmN`*4Ux_{n zQm$eq=EG>j40!t@n4SlO4+-V93dCfvAw^!~!0G|@1JTBUNvzLdP8fP&_5mKIwgw{g z@aa04Sb0M0CVr% z{c9Xghq&Lc0}ovEC+fl^9HQ=_g&FN@sx3bM%rkfI-Ze$#ci~?j3!x)<63UvY<1@?1 zx+!{@<${=0$Ws5xc2;C?Jh%GKmw)*&^K0Y0xU!*X5>^wjh?1eCY!4m=+9QQD4=9wUvh_}K&7~^2_KPK;hqEJRr zXZ8x_Y7Hfv+Jr7qHtqPh^bMKx9iBs7en^9xT|&7y8JV1A|3{xN*~t(lfPX`iaeN8$ z4|q5kV&6hzlb5?3{0R+iu|!33w_q6voT(MrJ_K{4Aa)k|%>w>J;>6D}rAZc&gdG z*pJ(c<_#3~KmDPP@nli=tbSlLd-YYvWe&iRW8J9zs5F6(coHz5jFv+9(D{(ulC$uK zKZWKQUSW?*4HWL=Z5}-?&7sj3^-SrJ@VMin`ptn_84hocQx))g_gq+dC%-(bzMJgc z!|J=q?>($qn0{9nJ~sNh#__Sy-!+zxjsC9jd~Ec0jp<{fziV9I!|J`qxc`sa1}$0mLc@a_9cZ{XiA3hj$8m}Il~@kftGi{A); zs!!s7f?rkVgQh}{5B4m1Y28CFaI_L=a>Ay`ZuLQF15GbEYTc#x@02(pckD54v3;l( z*l&UIkCca3DmyLpC&_e7LmHm+M<)FEv_i~%XfJhWCOY#KHI1`_ANb{6HjI5+(WP3y*AyW4u|4-SJ zm#Ro#W@=p6oJ{R4Wbu}qn_`E>63H%& zgWi*V*HX!%tyQlbiGuIFdXLmmkYtdS&C1t?N^=OgD4P$Vm{ZXi(ZVHuYneA9O1xL^ z8L+z%k%W&@b{gSHGV0!u+fE)4$nX`$RmW1n<8^2Uq14k5xjbzx$eZ_gkIc^yi({y# zR$rs$nu$%2Xj9WvO@DAN%Z7pDqP?)@hiMwCDvgc-;01DC#(l>8A)~opl{d`V+h#Y$|K)0==1Ok78@k_{-!(P5o$U({{sR zZLk_n-%K+|%_>q$snjcO&Y*pH7^w!Ir_r)h$>b!zao=yeX-so8K6v1Ko<@hGV{h5c zi%$=XaGw3x^G^>9PU6!8V;>&-z}QzG({BAdZrU4Zw|>^XqB#%dzNCAGto-Uj`?mg1 ztbT5$-Fku1%*UMr|YSXk$FzKR9f(Syv>qv179}eU; zwk#L;m0g67J0GCbTyh2(k{`eIc(il$$v1mYl4IpDzg zM=L5Ofowu#PGYerWl1;29;bfUM;G?&PN)6 zdd3SDgJeBm69gbM6och;-i~)Hw!kbfIo^WzFSdyEZ86igMYL~=5Y|Tk1Pkbf$-S6> zi7h6k=-yt=YwRFg4<)dA6VmQEkT5)0kCWbyh1PhNBai?S2k+f=`pI>jcye7^%=Hr- zz+z_y8KKXH!4~beVN5YVT(p4?k$dm*J#f84H0#V#*hgX?!)P%KNc$i6{-cEYQrNRS_Bb5-W(R494E1)T?b1?MUdOBjYmSwHp!Q37Q9(_&dNsIk9xSMElCw2lT_J`>hy(Ycquvy>RIz(~+0!^z)=A6*s+ePBaJ-id6s5mC8C77f^|< zPcCYdZRb?gPBZmxx8b(hXy!}Of(u32cMutuXqFYiY?Gyf{ddEA>qPi zwD8OUV}D)HAq$l*vu!4j3+ot(iUjivVUx0f>GHVC;1nKQOD)p5O|ew;OP zx_FIU8kkSryHBrcjKAStL{l*7JDjlP1|Ly)$@jRk$d)Kn5iuOB+(*Vi=Jh(DW=R%Y z5#CYP1wrBVE4IR|OJUc`b%AT?-VxV@-+}AGu}t}1o#d^gv?4}~j9>6gAY%j!4S0fp za1_eXSSDJ_s!yq6$n_f0Ql3FxEH5qGv5^o41ub+G=b`9sGdx|4H8*`HF2{$j<`*g3pw}~)qru!5%t$u$PIoWE0xSFHM8;dAk|#wSnRWcD5_8>fBB@PNjfp$pglZ*I zR$LZCX*%$a9U4v7uQi8Eqf%%=klf^5Mfw|Rr{ z!QmG+*>&Cm+ACcv?Ugyjqs_;=rp*y10CHs$k5S!ORetvlE>_bo?%9OXBeT$dM;E+3 zFI}Ex<`0~Em(PXWyPxRFGGi~UGic}PHht&OD?7{&4|5)$VjHV#N#wer5%(?3rFXl& z^XZj2=4XaEpV&>0jzeE=+=T}g`jaAIgE^$gBb>ZnGRC;`zhi#Dwa*M33po)9qf<>436KH=N)$(d-uk-SB{Y`57xcKf`{Rn1K}PVBl)$okpz*I zbTP$rD(KyhZ|;#61hRxV*MIL$kK%(T?-^=M?;hg>rh?3|;lbX4}w%dq~M;k!EfLqC88_K2jZFdIQ&eNJ6&1hLqwE zplNO+HRPNlO@Rwo7#rZF%Ip#THRP8?syvF8HmlU4DU#vh9fV*Y!2$d;>1F52h0|-h zs#X*jBuEIym2F~Bhn%kB3Cb1b>}ti3zCd!H+!!EKe`xRaOZ?y0 zyOy6bfL%}RL)Yp00e%0IuD*L-w9M)JPrqUx9}XD1Wj{OE))SC6=o=m=aB&iy54pE9 zM??2x8xM^5{P}fe&RLABCht{Pf@8-{-QvOid$hUZeG~<}UDcv5o<3k7XO1G_i5oi~ zTfBzWqaSwO&xwaH3FlcquuF?04x?Cgx?rm~I!rk4_NQ#l@SQ&8)r`L9^GD5ur2escsfhR19C8-q_xZH{fI6_y5=zv269z4a6|{ z*v`%|gvGyiHy!?OZq{{=^BX+f_?m3HqxyF|8;g1xtBhd(folo zeEhK6-`(r|yInn8eYiU>nvPwMiz7s$DXiA#0)ekPMOwe>=(K0~9)QJ#C3B(r6P$3> zJ`%SgIH|^S6evt$J=K)+TNUTbP*Rr<(d93eKJ4z+r5Eq}Uz-l+@A_Xe3)Q*G*o|`P zMTOFgHaH#%#bv32SR5`~FRF-ytv*#nCJx!hs2{-6HNpT;l8V$-(|>5^&`=GqAREHjl)Nle>f}Gj8g

%C(~OAUF*jcDeZ=><<5Cbxa=I zOH^daw){}J_`|g-oWCl0JD8Htk zo;%ie`<3{oXK(XNj*cE z=f7;;?x*?Ae;a+O7Sln zh?_XmKhN3H0bASaSo-EAhQLqZPdD!kukmQ}p{~vBpsf2B zk1-t5#%Y*HtSG&bu>I(?-@MrPzk$%?s`zc5%2SJe@ydc*n>&{JM%8*2`-TTPPIb0z zmiOkRhON^`1xLZY(Tok58_tS)>y; zzq|^UZ51j;R-v4gTrdseAZD^pMx|=-_|z7f>9ShTLGk3#so64d}Son>D_|& zPr`%e7>mgyCP_}-<|u;xn-7nS;|S3iwFeu+gB;aBd!S}Y5Cg}23=O9oY1iloTzOR| z83n=$IlC(eK7_R$IF(@?XX4SQ4Tf2poSKyY5vJ~L=fGz+BqlUhrPUFaebibLM%YiO z#3S-#b*~gAh*bYdgSodpxChLmTDp?$k~F%xTvO&DFbH90r)St4@+Ua}Kzxz|8!GR;KyQERsfJzpg_ zTD31R09C(4swwCMXq^z1Y*eJ$sCy{js*2L`4-T|C%Hl|^2T}`?Fxox1+fXzmr|6Kf zELqb~jIum>_kfNK<)Tbj=getgNuzg19s*w@&T~tK#d9;26#c}cU6<(glnz`$DW1}S zr*z=|4|Kq0P?!K&15?pU>bTN&S+V)Q8wik@KeA<*lFik)dGYJYjhs7jLSr1}sxtzw zv?-F{rkDgb#U!{XlLNvZQ-Tu&$=8^NUSEmtvS{fD3yx?h!13XVEE`kcG!U{+Mc0HO z*-y;r&HkxNZvE9C`#Q@8S^Qjoq4mQFXi+t!%F^qo7O%CrfOP4(y&pbYx11g&1?dmvB6OuRHHeYW{c38iPHup0WRyUd9OO+Vl|oal zQ%)}ks8r0JeE1O_=-Xkb4q#8tXm|qa6)g?=QXphnMe3~JJeK6kDbOdpQdZ7nMC!5F zknwSfz&3Gger`#cCRy=(_zC)5(zIegGsMr2SU?E4OcrJ_gJj_btF=QQB9FG zANa8%AVUu30telcuEDvPFmtJ!T9Dqtk$aa8z+{<4;uNmk z<~3CZM4pe>3d7)EZHgeKC;o#c17+#=sb6uU*~k{)G><4JVJ})=1vq(wJEea)G&o;i zq(<0*)&m!`m)uf(4*lTpijN$a#i!`<*=4tm>*l*)jKMt!)e`1*jhDRU@1ZNq zo6IT7s)`P0b@E%MwEwIyT1~A{Rauxrv$!o~RcEACCfJSnz$`6V-7q+kaIaYk6>o|Euyj^b7P2MlHaA;CjG^M3I zb;V;k`?jm`ozhGfc>Pa%+oAa!Dn|38f^PpB8%@C7O?$XDmTk)Za%Xi|`&^jE@cQ(c znX~df=VaLwUU=lJx~t)x((cgK9NFmyI?n=f27bkB?jF>g??D%J|50+1dM}a>XHDL- zAbQ}v|1BjD1?QG>Uh8~VnqorW6c+*=Tj_^Uy(0s(}njw;CkU{qs6}hE1s`FIALsn>&_tZrO{y&ZY07!*-|i9r)8ry}@PK z?+)3r?6ZP*-t_LhrPFEN<~%b8eBl6#TsVfw0FS zTa1BXwV;Zs0%tk6A6oGBV=0Ji39k)Aww4lzx^J`C ze{yZ+VN-Mu5a+<}0)c~Yh4EVqgiS(1*X~Z(4L_jo@Tm3yEUgkM@06CWvyf~# z{+hFj3)ff<;kAL5BPuM1CL{F*aSDtHzyf5Xj8bL`VxktOlU4{GOB}B^eceIOsfnPs zu8p7{bP$xj27-jw213veFHzc1@J^hUlKmOjzOuCM$o13Oof$>ar&t0m+K6DXa>b%v&E`%&MMr|9+s+Y%M!P>2`Fkx z*2evN5HfpNJMTf@a14!W@3J3qcG>wgcA4qk9JC9iu ze9{%fD4J)Kjwh7praFQ9p>+oyo2Gu-S(~5QcW5+cH@rgwVud@`Uf~~hR`~m`vBJV@ z1Fi6K!ZeH!V>3Pug%FwE$#$+!f#N!X1JMO&O$wkO6pRWJfzE1vSyGnG=zN{fpTNSn z9s?hLs|z0jQ~Nvt!MB9@#aS`yM_4GkI_SxQuv9re;vkfd&SZ_qO{VKWbfecjFwmb9 z5+?9d*VqxlYXj|wJB&#o0?+JBTm-u=UOX7tlpQM75X2qF+t%q zXM(-a`R5o-(0>S8BHO0e`KQY`5)1|K2+h`6+61OK+*5J|Sjj*OH$nPTfw)cjk@OjX zLS-<5RsS)VL>wErdQX3Av}hPL{eB=IX)UEOjlh`z;zI+W*$4_|uOfd`sWLI`^ zr}TezjeRv8Dg7p(>iGLDVZEm>nb1L z)*`&z`8r`s2_w25d-U#JosW6$MZP#67k*`NCDvXQhP$eRJEfOb-2&`Qm&V@zgDlO* zyw8YD-8xq^@IX{=5<~<~*dGoFa$f7a-YmAKz;MA}KLv)|QbJ6@h%?3N4Zw&)0bsf@ zWd_LWQ=m#yrV&SQW_*Bdh;Qq%`!iz@?}ywD$UV`+7<9=L<(Ov%F-5&5*qZdC5^l*V98M`w)8MyYkNIPaoQz!N-_NDrx&1jy)UI~)6xuk#ZmjG{uo{)0!T?(CX#|BV)`IkcEVkO+>aU5O#u6ZUnMJrqK1`0_6{bBk?MM+ zjiijYLMZhIY_#;8wB%4YSZt>~Iz&G9>{{@xqRF(+Y2%mSwcfD_99xEO9p*B48Cm%{3GBhMJ_o@J2#+?e^ra$BADsajd*EG= z3_oC#298qUiH)$1@DTmIeI3*>LGh~X`=Ut%g&FZDy}^FTBAOg%`?~|O{Rln%(9i&W zKc|iY$j-CgK);`pbq=-jzyaHNhR%LwXmG#gkoO)Ji7=@FU+|c*ixRq1*R~$6w z;<1eYb}atxb>`iVL_(kIx`dzfb?D5c^NtVgu^hr-f6=_B`5N%AHtsw83A*{s>nuUf zGv1pva+gVmMZd}gMW7+XW?phw{#LM>2|;rJMMDm11LB1J$ZUahZmgV`#B5AurKqYo zhcg_hjf&S#`_V%%(}rJs)_OEHLZ_cRp1_4MH3n(_Q8$ayiGwWz#|B=~<~Ci{%l0GJ z0Mr@LJ^+T@N&UgdV0Y!>cu9@e)fG~M!(rNHgk8V3;A6#7hsiLn!ByX4nvt^H1;CXv z1UMC#G^`vTa9pO;=frZ-kZkLa^H%)Em==!5Ejqyt{iCiO>ZUud4K!VT6b5WM(MA0*sNHy4Kbbka*1(`y%d}#XWy{*K&;YYvxAPza1v3Ca!UC1Vmly zwZ3R_U0`kRdkv4>X!yvQNw+vlc&i=rAvY0lhU#?xa=pn6dP%=F%*lMrlV%BXx}rhf zsSS9tEn#iI*6vlE8zv&YneKxg1lS(B)32omYWiQY0(lZH*=zWDzDo5osOkmO5(?t& zDR;!AKRoHTG$v4&jF`8Gi(dE4f&b=B5d32fp*^z*gRIwPDK05{cM(Enz3+sD>m$4ImplgxbJK;G%&sUv&cCYY{biLdY z;ar0G31`dl!P%C5(WLakjQEqjU|zW)ErNID@9ydR$3vS={|zf(;rr~J*Asti!wRnr zv|;~ms^Q}l8WOdgqoRn2z3b>m@x?OUIve?t{`FKLMe*B8@DWoGDZxt>WV)CZDXBhs)%zDRaq zKG$w8eU5S)TKdZFG9CZJP1gQzI$Qa_w-*uTcx6Z^7tX@ay2N4PuzFMfu*e#>9u2b2 zoJ8E@RreA6{b4QxryuqZ9sZ;*60W2e*P|%^Vwlqz9rU;e+h*{0`k|y08gWif&>4bV z_5XI5^L(Dpx-&q>{N4|~&U2s+d?T{Y_Ode&UtYg~RC3h;zc(lqvSH{9{EEj!`(Gyl zp#Q+8ME$Q5;Wg(v={*{GzzA%Uz(3m_FsX00(#8~2XpCTSB+ZXwhQ>BFjk!!Iq=IHc z!IU)uM8s9laPGFN7nK?az0s6WNzSGHX+JV;pst5d36a(cc4xW>^fwvdLJtC-hKw15e2O3txY; zk>K@;`hC@c5DCA1fN_V2C*j--y=&~uXxAxnU9fI7_@L^7&v$j%bJubv@;`EJUJept zBCXsW&atx|To3d?l>>Xj1pR0K9ooGPe*ay9+G7*+rgh|MBkVMX);X_rMjXNhCmNHm zO$kmaQ_M-l3D|zkX>Suk53iO>zB7*h$soQnqk?AwhwAbu{=nm@rt^;$KsXWg0yhnw$3 zQ%sE>;2x%Q#gIaV`O=Uxj3)aKT|-rJa;CUIW! z;De#jC5mHd6?nOjdnKqnB{_q`3o4vSp!Za^6~%f+SQ&{-o>gOIs3kLkzu)8pqTRnO zF#bJVbjL-7dx|+f*!AW>#P)AF3&%%h#P-uYS45appN4&B)T?ylPY!cN9~AeHFah^- z4|uW4m)ssWuDFQgz}gIkAPXzutNTEdC6_{b*1OuG4>2Zly)x4Pj%V1TT`W?may9d%jyP;?*s^WlkV-UpC0aMzG9KuItGgYDdCYk~nb z&6e^*s(GIW6JSe2tRR)gkWLN~+%xXV-x%}%WA9zS)4Gp2WnN$#cH@ueIN&cGWr6bp*l@_NT6M&VH`F_F8MN zwf0(^CfqX2M@}nzwt@5L40QXMg{;9XsPIOQ!-E&5dW8L2!cnVQgD$WHws;chAmP-B zI}310rCCYnWd_l}9DGCJDqsT{kvrMYPgew}20;T6YcOf!L|oyeqN82Ld$dCub3Ru& zGsFTB^r$atPZ#5kn0Des4bUOIa6CPU@VW&D`<@+Mq7l*c@6UOH<9hy$Ikz!NK=V8`i~fn zu7<)K{Y{abXHB8)lW!WoezA4`LZ&H~+g3|MTSY!vt7-KHExzg9u9n8fS3PC63Xt1v z$~T@ijqvVq!YTh6rU0?O6NK*TgZfy*>#G?LM&R!=bnJ!2?4d;ePH z1*-Xdrur*`%y=Q`nfxJDQxm*1vDeW0e3LEnqdIGC- z(eFeTucV{8ijihb&wcNiNHVZ?neMdTEeWqRUa^S$u{5WfO)7DjL4?Wcc);WW;O1^} zoZ4yK3%S7g@l=76RFJfRlemrtI2*yC%ajbOz-3;zpm`}x(A?LDz-&RQuVyw3K!Eu& zCA3?EG4DJryLjse5J@-ge@k-#F5%OKHG27?x1lGsZG0uoq2Cg72C^mQ}yrzKr{8 z`ri578$4zzKy0|a1{CwVTYCaX412#^AAv29STB7*aRkB}ETPE6K2uZf@KsMJ>`(yR z?(AurG3)Nrd_|8@U5}3(>dKOl8XnD8_~gO1E7d}o0TYD2rOQ?u8Ac9wt4YBnf{M}Z zGY}-D27*}4ln2eV)PrfZZ>@K_?PBd7jjWx^EH0HZQ%ItpHCiy9KN)GH`^J}e{?kSQ zq6WbV5FY}l0_q$cj{sKd1_4a9aL82?P+r5IB$mGyKcOvur?`q``TOz#%V)!c4Pteha<>Qy&?hePEA6G& zkO^vKoL6>Wlo9qC!(_7)moDWsay9^xjgWmdu!DsAF4_?H(qS8IHF3S0!cC&KO~Ijj z9L?a~k}yxw?h{w8R&k+U#CL(_7kU+Wy&nnX7lUdG0a=hNfwIffGpULf;do;9=o9B; z+SX=0M;&0*zd5quoL$@1`uO2U%|u)ElLmQI&%9N)wJ)$Wec#vEv3IjGzL4hF??{G+ z)_juIh@0cad~|*~&B<>XHS;Qe{yL^4YGoCT^$DY%GLNLTFEwoo(`(bS7KRCUbk8h2 zSm;i)GHNyp*I{9RH3HN}7*}9*qWhJ2o~yMg^GY;807hZ5V9F;_ol~-j%38IyT!L}P z@qnWckf37H%u_+U#?pK+kBFs$*5xz1$?gJ5#!P{H8k(=iFbv+jPW@v=KNr~wK1}tL zxlBXQV48X7X*V8Ie#Q_+gDK^Y4qHc&otZuRvb%?23~ZbJOt$-mb}Gm7zVtg@aS3uj*jw7N+M;a1of>xGrG--=(Y<=JBW>>mKAn^0vrc8yxXr@eyWC?}0(aLd3@6-s;nzT7Ok_&99M&Q_~ z`tAir)`r_yeIWKfM>dQrE_{w6zsNfoyG?hDJX;?X^iMH`MB>8BU)AWMdS-Fq2Jd2< z9uc3*T;(#@K&kgsJ8B%x48lu+3ACNmw$ab-PPj46$z1K?rZtP`m)PN7rtn&g-@I@q zaA%qm=%R+}4(+XC->Jc>M|Y_Ar#aM1g;^tS0a?HDK-b3)qEkW{cm!yIVYTeM_O$Mz z-W3+5Q4n$7xnOxFO|T63DVsHyP9@@|)%BL_a+-74-8T`orkg@z9>{b@CuQ1o(c*H= z7Zg%hZ(Nl4#x%isnc_kgRLnci4mim6(`?aIpT*)RWUJZ&0_;MGIvNy_x0<1WpUpW4TjZ)nH2@ea0WHJe zw&UIKc__e`7UMr%4gUD$_zVX3f9rV*kfM5SQ zBb`MRUiYDZ82wa)a`-5%r;N3AtO?m~J#D+8ZwGCh?zH6;=}uNyQ}r-ed5rN_BJ=Mo zU5fz5hg1OLbVz_5Q}^>mW>*krnLQA`I-SXcwF2>Fee2>F?k<%?M}b~HYv}T^wyBwP*cdC`3pY9@jDa}PVm?o_^0}az{Hx^)Tif|bPy|(u* zDE~1{P@a4(Ih&qG_?mX>S%`S4&>{dvDm2x?%T(wJ%jjpnix~BcNiB&{FZ-~#tf2hx zS+<14Dbli#r3(4Ui?j8DLYd?h^cs|JJp^^lK=~H@>8A1nQCy;c+OOPn<3*fyc%XbU zSFuFoFGHudrwGj$E4m@P%@rZeRv;w}zfB!Wd;d)I5xTYHrN{>z#wNJ{+>*$Ah5M2+!?Y{j4O7M@p0yvNlUvJ92P>>P$&B zD>WEeDZ1wxZjv?;zz5kQOHln)MoCizWg?8NHOO(S|5TOm2GMdLKq+|$f*pl@n1hUk zJGO}R2Eh-7qOo>l_y$U%%GY|497R+lEgp3b6{u9RQosfZ@hmnLWR|7E3WUiL$BRSP zD-biu3`-hPeRCBfrXfgwOLnRtYpfn6H4n?Nl-?}&3eo|R4i36R61AiZ21Bg?kEXDU zMC4HMUFRuCtu1%n4iRYnQ;`klm=?NbD*4$+%|uevpBX7i^~_S#cbs?VXRY6PIJmf# z3%6kRC_>A>HPK_9`>RMtoLNKU$UBc0R~n^Ct%@L$$KgOWbEZ5s zzJ7wUNg?Y1`>i9pROv+7Drp>$&T%gqaEwV`#VU1ETO{yQdUcPZ(TFn6aF+^0prOL? z6No<{egyOf;3mXf9M?Q^G~OzzTN_%v!7^X`ShwS7#w>)|O!g}m5?4iK0~*cM z5NU<-#iCN4f{d1?ZtT>wwR<|3!TcVGYywA~2rrcbRIt_08cs}89*H!PXeV7~ zAf^VvJ84{d4<-eM9!%JKT@MJ%X6bZtu!c^HEeN2)Ii|lJhX)Fx|gK1ac6bn2-;X^e}lRD=y_LW+C*Ne;DpDe*8F^oA5L@VK+D7Z*C&m z+ypl_>1l4#Zf?@w++?)5Ni@f^2<{NEvIB0P=iups-`o9&7hXt+PyDZe20!MhQ}{+` za7(1UuoUOPQs^zp^rFBNvNM%PAcJiGIfvegEogRbcwb#{k{rlsPI8i*<|MfSa=|Jd zC^;T|Wlmx^pJ|4va&HW&jTssyk%So@=z|%;J&24-^Pn-Miq0LIVvBk~%m11v;6VB7 zV%AhK%E>?03%ABqEJk_#-XaovDeiv?URUrJQ@qFj(TpP$F`yLvi#>_3BTC$ z5 zJnAmk1z1Wi3yCEggQZ4$rk$_$zd2v{YtJUIHs>wUd?92$*lVC%(r0Dp4AeG_EvH9b zzW8yLYBtlKo}`Hv-~85gM(b?Y@LdkgWog0|FQ(V+C+2kPPHJpLf!Fg2Z&Phj1A&+C z8bF|%0wjU#@zVVskXXxi@+cG@9FI2QU=9-%9C8qv6-u9DPl5u&1O<(Viact(^^4W~ zYUyNY1-|-)e67AXMo(i_jTu1)hmDqTFiEGD7A0AE2*N=GWn7==ZPuXVH4=rLzE65B zLf=>PHXP}#v$!j~o5z)Rb5fAm{>v{H^+KvyvdYGHpF|Z!ek&S~nu9sTMxc z$B`~a$Dy1$4w>`a0o`Wr=ex)mtL?OBPC7|Wd-qG%#nSuK`KPzTro3F*Z!;R<%8wIf z4U`YX?0o%QTA_8D{M2fr1$LDH|Gubj3ZiW#X89Lm-;|-|uvgM3;gK@UuJcr;QCq@rF0I4}&YS!0qx*7Ut; zfYqXM>9hy5IUb=)zFQ}rGJ{E0*cxxL@}BWHdhYVR-Bs{2TdJVf2^30?=WYx_{l)Ax-{8qmm3V8Z-pch`krcK4h{M$CZJ%NXf0eV%)g&Ts-N}4Unv=Phh=?@<@tk1THYqa0cHX+f{tsym z`~LojpEcorTk?P^@s_Y?ur(t4t*3Pt`L#(*(hvRY0{^$u1b&t`KoUNo2Z5-(@{knI zTiU6&k-|C&Zx+EL_U}=mPG5}sz)COSGl`M*VVq7n(lB;rd_q0JGInNs5>J%4z%`AX z8K*<4u?0mfW2iadv^wFmI?Z=j@zx0fbqkFV{EEr0kBT?fB zx3#x!0Pg(?miLd5(Yh>e3x=m84*u;<_2T#Rpfghf^X5O)OQt$!f%$=c;h@JB_T7{R zFx?sB)37dD6H_=q6YGp0N@DNcG#M_frVY)0gl;56rYIzr;&2;h-@OSCx=-y7Y#*G=vP%!D7h%6Tj?`5pI|<_l z%9HrhjdCef$+pcmDHZV;S23!Ra1sZ4L;@i@c#)m@TXo^t-y!gs(r%M7;Fg%W*1nG| zis?KU*~cys*iaOk1Cb^Zou#ic;8H_k(y)c*Znif+dWLN_L?_wR>VVe6C%e*J5Jzk8 zwXs1yf&?oJZ4>Ll&Nhi>SM*}@8%s)%%1{O3DOZ7D%2gne`l<-wN>qXD$yFeAauvv% zzABTl7q68q=*i5cg+uO_L_OKHkPv;nD|wJ)Px8xZGqzm-s*zEN{#*Vd=cT z5}0ka4?AS4O^Vu0CULrRjgZnorPg3Yt716_288pbGRPGcaepf4XqQW&B|ZCcleX|c zx!wz@VLzbB$363X16%Ii-#-fMS@nR#;#QwO3dy%pTxjN&2pK4JI4RT-QvgYZrL!`L z%O_WBtJQq)7};g{27vJGas!ZSv!;n&uH*j)-U@Uom01hP|bBEVN*fLSwO@lUXM8ZE5}N0_m621kz=CgXvB}^UgEmc-hTDJ|nxCYT=gB z{Dg@#LkF4iD5m5bMvOcz4kP6}O2i#P-trg`5n-(*;-hg~dz*AAp{HN|BEff{{0jag zIXcY#s@`jV%vCI6JO%NeH5uacKmpU0r-VyrDfcWcD)VNIB{riR4SX`2TGdRao>SI?vjqIEi3Sl!ct`OX#>^q^5a zC~4c>GZS9Gw>ovbc&Cm?jcu#o=_qqSFYdFxye4*%PRsN)jt`m~v23>OaCa7w;6zvd zONV@0_rAg?>h5zpZ)5M*`@FF3*$qIa#qcV<~U5tDIzqJvD7+Z@N) z?~KUAK6~cyI(h%9fd0ri1U4jM@St0xh)2|I*v0F`LWA;w!9R#Y!d421LP4pK7i+jC zy0nZ4u!Z^};*Y`2XPCy#(PK%6Us#}$QjeA)1}zn6xIr389?*GsD>OWwGtJTpk;}?T z_R|uMOLMhKYZ)2kIkSU^H%B#8>H|uHFk4i9U37HdbYn-ttCuD0T*fnxc#+>CDMsWs)xx*1`iW1D zIUsS)IyoB0WUCOXO`SrK7^&t^n8eX|5 z&*D##qwVweYD)YaT*Vx1r*DH`azF$PIfkfP^%_Fk<8CNB5f>Zg9sq?U6>&Tp`9%l_ zt=W93HufmOUBR4o5ba_QK7ZEkF(H9H37;N)TW0PjT>fU%|Ka>aR*r#CEARBE-8 zBn=^;7l37@6lhBzW=7(7uRl^;F69tNw8ZgJaVeqEn86|}w<-ms3OFd@zSdC)%2c36 zqgXkGfL{>9pkZH?&I=XP>C)*EY}(op^uW4;NV30O2k%$7}Hl98c5~;O5+Qb-rbSR4W`Cw9?58N z{bNn`LAhz^QMFNPWHPq}Ptg&OPK^!9mPgC6P+CO%AaOL~7}pD>O0ki7{P1J%+V{}W znZu7gc64U%fy21izx^-=zHV&K>_71E6NmTB9N25U9ox?h?Wfw?GkbR*-7UWp6XCaG z+cTNc0$*>bHS#Nk=2E_R8re!3*=T3O_h!g4k5bDgkz}U*d#8rX?hsZg%{+af5ADMY?hLl)SE`HKF&Hq{LUU>ePeT||sY=U^9G|!=kN4Y$M~CWE1IR=o0^^VgZw}ibRTjE|#~GC~)C3T03J9Ht z{$YM*VRa!ZIvk}ipwJOCrm~zZarCayiHYG+UVY=6Q7s_H;Z%0$(9mcw^bGy8oBr7+ zt}*dXnP+*ZtuB%xHJ4C%0?8P6@&D-8PNZj?*nz(2|MzgNM)?CTM<*uvOIuq@7<{ol zwd0=X03nqWPtRZ;Q~L{*MiJ1tE!cOuiL{nAQq0Ut>kgAVN(hbYLwE^UW$3mc@?{dY zw%9&QI-@zZ!Sc&dNL6TzHtIf=OIqqwR%_KEj;<=h%;!O4igmjEg;1)M8pI#6JaZ0L zR7adusdUN91vxgPS{!;-EKajFBiqV~HM5H#R}vYh8&gAb+zoI*FRb{Hr@pF%>@WuW zP_2rvUQLlg>SXar8q}l3C%t`-O$8Ik^h7{=xL#|m@ND-I)u?0P2}~@8gL(=z(rB<- zwS#fXz89(H>{=G6LH+M1L6M*aof5lW&~GjSz?-(U3TwG=y0qL{cJp-i9#B~n{&6`JFl zOd@W7REYm9f$tC$XyB|A6itHBRX|Q>rct-Ll7C;ZUTb8t73}rh!=t-~hcna1?j0Q` zJec606ZCU8|D5EXBO}?#{P@_8{J1E)L%q#S*spimuW2EU(uKWh323=e2csu>UF0Gb zd@i%M*gy=hM+@*GK8-FxBJ}EHr-z--dERC`F1r@;kCBd%2+Xme2~8b#pU5I9=P*B? z$Wo45{yi~9s~~w})^10IbH2c^FEDJ9Z7wkE3k;jiC>Izuq&Hg<2re*emS~SbAicn_ zUr&aukNIo|{wtjEyXjLYt$g`QIIWle0{@Vo-Uo0_hjQRAwQ@k2QCv-2`O7F@SbE#D zXUR=U-qt984lki6c%h)RvwtDmi4LOt)oA_j^0{KAGSggJDV_~4CXEQ|kNyiaIZ#&m z22PGO@@`DZXnezOdUxyv10 z=+9Dy`jL0N_$_*&)s$E-dMIVs>Y;{dx9nMrcwL`0ZH1SuS=agdzKpYoD*XHU6oM1l z!Po33vD79$VP9=%kTD&|VRfTh&|cF7{va>?2ffy2($*$?YjX+b%|&+b7l-Q#Vserm zP3tea7184@KR}o`P@d`qAz3~7^5;DxG^-hoz)FJHKFA22!!G7KZqk^QlSjrJWdvMA zGB@J|%C(Z3z}mw47&}!GYuivqH_=G=AbBY*ad`vWX#w0PWAH)UK6~Y#HAs(e6>Gu& z4bwMLB8X5tpr0~quq>*bPJ2M;`3OS0egy;0v@{?Taqsk5uPkr#j$!u68(tlJ zn{tdYlQyJc`esN*q!Imvwu$`K(I%nkVKL2-{>SBp^25uFm(djd+kFc@S-0xgpObO}a* zg;oW3`l&3^)Xp_!jp6p2EQrdb_1%M3Dj0~Mq{pwU@h&HS`q5sXWbMM<>S160Xk_!a z#7N)okeg9IyF2j3NF#~J&E|s|SJWVw+&uUCueAlx^I|!IH%fdEWoa?HlG5V^Uej}7t#daWqPygPD1m}GxT^~V{Iivc3occdzpu_j%302 z-iD`+FVqWjXC~IqRM$lkDx~;Sk(TL>W)`TGv!Eq%S`D9=bZSz`<7bX@1-qBE^_=~v z+gVr59PPc1sCsQ1N-F7+ZvkQ8YD5iIXnu#K4E?$P6n}@tWJ)Q zO;^AqUn+=H5#($|98Z+a$@{hl(kX;`OB%o4*SfX;{ok=0HXZGzBH7TDuuP=MMC@w& zcWdyeAu+oe?-?X|Ws!NPXK=OyB%VB2LpLk$l82v6Kpqlqa$h9s{IteNv#{2deuN`WsbI2nr@n>&F~~#TC?RDH(#=9@bdCv?nqn-(3e!nX8FwPP>LCdN z@Jm5bO`v;$ea0D~;4IjL5)V16!QtU0xmtq31*+W@sCa2#P$8uyYCFf{n9GeoVt_P& z6ZhccAO+fhy$6)3)f8lgj=zV}?cU&R>MdjctVgziyMHcW6^7bF^2uG#Z;CXKXdlhq zrB_f5fcMcgQ6O8BZ~qoM{7sW;==kLh`{QX2`zjX^bWdUAgJ;0;0+0nOLlR82@Q#j` zYdWO>7&0{?H*F#s`Zh~|(WED_1mq5o12~jYLZTL~$Q}DuOVwj74eatS(~=I9e+hrO zi6+V_P=8kA;4g3$OZK?(sKp!Az}%On-`^4a{xh3^tbsSg9`yw?trw-*+x-Z3V?s)TniF3 z1i{&2rM8OOUc>~sw)B*(lui~arKMU8_rJ)>g{x$uEUTp1xq$pNF>W&!F+#fCU=4!z z4Kc)gZ__oDy62kM{(lkKhAu;kQ8vCBsoZrBt3a;L|F~Xl)j2b>e9+U9+0*W^rf1qd zNn4FhiK8g9{F~|N7E1;FI;j#uwUD%l0?4B#Locf00ohH#_|{nUfpWNzSI147B0__arigCV z5qCsyxHw%w_9P5S*ikLM$&0#L%rGb9wJ<9I-yC}-=rR`bR=orY*5%L5J{lT)LznA}@B zZXs(ZT?%=6b{8T>syLn|ptmb&uk((#W|gG0XL{WMtDI?3YFJ}-9C5J*)k7uE*m?r} z=`=xqE$?NUHd>SNeDn-HF<|$nLv^hIrqyCUv)Y>m6R_r#PJ8{#Xn#5!$(mDIE%r03 zhe!zMlR;WjN~O7e=5vL`tgUf5uRZPeL~4F6O;VF5D&K=Np*eflYfaCz=%tFrUt>H*zJj6 zYRxlYh1kj7;9Vvqk#_j1F_%wfUrrMccRCU6Vh}`+BkZY~2PpB|$67hFeOxQw?`O^& zmzS9v%{3g(j$?1koN*kI*b54`gHRNwp`}?FYI0?LeckqqI{#cBzBVVPF^DmF>yv%! z7+wl)QFcy)CPv2)-U(W^0uMNG)`cQ?8JW7&Z7@Yw>xg$;-=9i%#w(om^AJTo2xU&CEwHKcnWJ{!w%DFL{u`Ld~~92UvniGva4T zahf3HD?E)UrAHmc6ajJt4iQo!1f9 z>n?tk#-H*$9-&@6#<@cOIJff1b9&70}DF18P9|PsDi&+NcCo(yyVZFvxti65pQnci+MV*8pYsuWkuh4g14T_#dV)qhvRRt=d|Qnw|6knTTBRNMm= z!vQ<_MOXtQV(ufOMQ4u)(ItSF1l=d=4rq1mJ40u{F575r-AFC3QEwWyyo! zT?Q>L#Yd0c_=;}Sz@sd~wLSzP4bW6RP}Y8NF`M$6lPXm>T~I9Y!Ij`RIb#ftW+2hi zG`P&*%G9nwc6<{zyHOI=;Aug@8ri>25(f%i9yG``BQv;5c1$(-x|&I~UI^+?=fSfA z-VbzTqfu(MSdK$L8n}lPs%zXYtG(@B;Ln$v$84(P-&d$)xZOES4jOSW0_ubO1QAwv z9w6854(!uFgozv$xC3_W$TV6j#rmldc|=rXEP1S^Ohwax-^EGDKNSScDlJg#Q}UtU z`BrL+?1BN`lSNpF+20T?qdAI_tWsMpizxLCL?AuARGNkA${KyCi3ld};V;cCWdJbv zzbI#lRs&Ib$>9Xif5~$nUX&OcjjK3YC$}mwyG9F+i@>Ilk%Lc+i~zsLm!t+*(kL;Q zk&z=aky&^UByjWUtz(o90AO`v;%oA&uiCi^|BHp0! z_jt!fcZJ0zHr@t^=HEpKyUR=3(|mN?dy230QpKf1o1PLRSovxs;xW(Pe5*S9sHrfI zxmw&EP^7>DSM*CP5bo9ShD&CD~M<(a-KQ zhtizprcpDC^5?Gup0#7X&ly9ZltM|AxzeWfrq#%MPy6o9LNcUX3IhDwo&BCPXP@tl z8(EX;wWep<-NCAyhg{%XuGygik@dy})YEAKYWiF;Ukjw?u8$1^A4{~mZ|T6AVhWA* zGnWVZl~b&#^xsxLGufAlz+_D)r3U+%(^OyLn>Cj{8}iNJnc;6ZyM8vibV<1R;1i1U z&`(Ws#0O*|?}h9vJ&m}ModlI_?XsxVo$qV+k>A`+j#YE&UdYV0dpQi`2UCN?XfG1B z1)<*R`CxZ#xG7^A9}GV(A??&(F-i~t&$G%@20$&Qu#y3P2-t) zyn@6+THm=o{2Y!=bbv)DLvpLHRYykd4ffW8wOR}3A>3!D<&Q0oyR^({b43YjM(hi?nO<1gnl=;@lO)o%jbIKxQ2Dj!WRkFSbC60&)jy za4&R7KTVgvAK+aA?cCB>hLG82XYu#QL{etEB!$|X8&p%94zSk_i z8XF&e)?e2}Fo+kv&I18Z6CEfu}iAE1mn`4vXlapiobKhy? z2ZEMv*Hgvf$(_6CLw;p`;eJKd!nbuDnYDdm3Ic-yf|q5wbTcWw_cfoSBD}9d&l-uC znGZ(*)0vouTait>RWF?_-$^Khc=bPz+k%i7f6yS80&{ED?6pot4XYaAWSgn+9w>dG zu@sodp76MPZj`_C7z#Ag6kSOwHQQod_tAm|Qu#p|+d%pL#BicLM;ZG*VZdVMXTTB; zH$W+Jf>Pv!V89L!f>PvcN@3E?a;0SCr7aAGAp+-N7>uU{n4iL*Bn0invLshsBCGz1sjelYaK7HG)k3PmBQ+AGL?XHo(WtwUiuDaE40zk%jgJ0_AP~S z=7kq@r?JzIcZyig_r1u`d^|~ z{@4Q~i(*|WPn>Fn^s^pG`DeufeX8dLdUfk|Wq9he1J~LY$npcwBx}khWa%#B>o@`p zVqWb3)c;~{&U*lEE%v{ofHy)R0Qs7<3b^;h-ZpHtb!@W2k3+{@AYn(ijH@eWA|8Wg zcCeBiM1+C5zDorwGwC+yvKc4<_oi}7dLrf+d+*!357Od?aV0@jelLQwE`Ri88<6K& z-zRsy{wUIPXG&^u#%)jN6;eZE&R9~>0M0#}^(?~Iuv5=c#B*>9ScZOB==|hCO7Dr3 z(eQPU8HeGxL0%@Elx2J#ZXo3}B5ZK>ZQ29Ec?GwAgzK-sL>PR|Rr7iQ%Kwv4XrTPJ zyVxCjY)DYdA=r-l4tNW!VjHVe?V5-zzncB2mc@R-_-_$zhtkRmdyT)3#FY}ui~ z06&Y&g35*ox$9D>Uq9de>VwFKs7i(3a+}(nX5V8Ul9L9IUU&!4&EdJ(?iXSR@H6?{6@`2q?PKxjhWC~I* zA}u-@iZW}Q4+;%W2M=;bid>#Q1JWzSBJ95cn7a|=j(I1a)d;JMiS9n_ns;*Y2Q+zq zMz3?Eevf%46~1`$PQGh*1vI=0Y2g%-_>8h8kM6XnG-$BJfx|TfhtmPW3l)k?2{th@ za;UYuvIc)id7*)b`2GF0X>G`UTf$cRI=vc)mN3{^%ZPUgk6q!Re9}k|s?P)E4@TN> z*TYUiNJ>kx{_puMy^LxAOmsVh3Rh(m^@yhjRtIW*5j}PT^x)A2c;4FEz0znl-kY5s z4b;{i4eE0>Z0)H-kmrYwjXx{*a&<-2Bf5Y_MnNM~vh^_KR`sF?O#+rYy^b5$ znTNlkm-PE(JJCV(nh{p5oOfs5wHq)8S%$fNjkwlIrH06TI1kW*@hOz(Ex9$529{sU zBgMNy=Z#j^pf3L>n&m+G(;8@E19eA!(%8@5c$-p{s2OeuH>Y1j&b3EnQs#Xkc+}9d z0kBRm6R@0zDcdDZ>hP>*_y}+)Hvtoq5OD?c!5#3U0E#j>n)s8?ER+^UgTiSvzEWxC z8?9MtuaOn; zmTIe6MFTFZ-WzlEGyz~*`#J3XD4DYvUPmeDWlc6VKx^v{XbSQf*-ms2{nJ{*WW(EP ztJNIoP&lhDB1G}=KF7jOP6|Ip1;NFQHb51pOEN}QLUEH-eCb}0MRRgbVmBP#J>_;7UL;HqP~|`2=0UB_l=gd#It;xct8eyaVNbidkloYaYNHo?Flh^Y>iE z(ms6UDRCVdP$*GSNqC59euen0M4F+Vr4U`v5DXqh0*%I84XlgrhJnEun=}MVNE}u8 zW|3CH5N?^%M@XNrR>K=@u5_~0%s~%Rhc6mQq|`X>zl$qjHRxy>l@hL-!dlB#5>x{)Ghh8w|ZEo%ro7T3!4w?lo#|< zuQKoC>mpUV@^NMS{D4_+)iukpyZc@{GrBN^#yp_01pO;*G*R2deW_s;y6@5O9lZdK zrTJS^()@&No+&Ppeulb*8|_)cvSN7W;uq_xm?>z%H+^v&J$-+{i z`+ICWnKu_7bQ< zCNx=PcvzHFd-!$eSuX(-)l*pOsAQ5zJZfkp*?@TT{J+s$cty4o9Yn9H~ zIX+38()(ssh!?&)$;jNhk zzG?-BI8F|$lS|hkdMdpBpTIn#pii{^KcE4yf%Sj!4>c_Qy=*5sh*a1A4p+P>%)?9s zk!WP((Ifk?bA=dZi9-Ya@)r5(R062+l7>v1!fp%tQ(K>Z>%|Dgr-AZsM{0MKa+q1a zw0TfzD^%Ak;#_X#WxBCmdWITLbFI$$!{Sw=_TQH@o@`+My*#W@=Fen1(LtoT z|86Dvt5kt>z^NCJ`-zem%G05?Jtrus0a(8L7}MfQ>5pZw?7n44!S~AJdj9=kvEbiw zGW+WFTHTM&QidqZHk9z7adu2d9Z*0CvhFmd(N}5=4T*mpde(qv%)JRvk0MPX%)L=B z*9I{6)m6Q6H^_FPgXkY~4H*oF!7Fge&M}_CEn8?Y*gsooF>=si;WSbcX?5ZpNtFwO{I?nI-T^|D?&Wnty-?Wi&a{pMc)o$s$B<;JsuK zf}#p!Dq|Ay2>E(UA|a8L?ldI;BHa)y*F+DyCKAfD zB@&8c21=GlDASfmh`a73PeMkjxM<|+oNDDJ&?%irxUv$wsMt3qms5Kpp}RM@UKVz5 zxCvn0yd)+{+2C%Cm<&0~k;+{ekUB75Y_^B$oW-_FrSqXRG5yLjP;9EU+rX+lH5ruI zpAITpgG#H#p68^v24?j@pF7K%PTx)SGmri0>~L#3X|>qTtoHOzaI6`n(ON&Vd3$%- zXN@$jq-PX8k(-}Plib`yLB=}JYimwji+Lv0oh(^fc$?`nD$*yF`fh@`jM5{$`OqC~|3yw60?$Hl$N)FhQ0XkO@L zVfp_Hk!BJxy(td8#kP5Zx)qaQDATbE4P#@WYy?|GQ_J z?i7@;ECnYHzqk_rclJPwK^oS0m98l!5*&$P#u?RZC1wRuzz$-YW<{0Na(bRsiD z*x%xs6fGZL*Ea$$KxL^pY(d4{KX+sQMD-@N@+QG`(`~<|kM{;KEsi=PGQ$%8^WCNp zi!Fjr7vMTkSfrq@swlfmSo{_;ifp7P?te7Q6}4;IIhHdp_pgh#is^Ti0Y1zX6QShf^|vd#lIRYsFM<9}lr_NZq`$M8{{Rolq8+OTV$vMq zViskyN}7PNGjklI%-kZ$7$kq*>xoekoYogJu+D=+@sl@^UjSRr<@Z|%Gqipq8?q5;pRkY7y`Bv-k}pnD1M5>gJam7Tm;Dy`oP^uKsT{2aQC~JV%NtI$XAM&6q z<`YfkDJA^qYmHasPZCZIl>a0B>B*_`2hAjD6utxQc`X`f(?NK?4^uA|OEAC8OWqj| z6X^0^JgD*0ES-f!9kYFYRv^n~dPbI)8lE0qao0`A$pWe+iq1x}eMSrTWL2}o=caLY zVg76X3-j`Kd%$EZOkB3lcA%X5Y@c6_AcIR$F>vEY@gARvm40v8Epewu}*>9 zhLiQCVaw$0O(*(3puPGfz0wAdhOLrJvdfi8!gza+VY;t@kTSwRaCcH7A!SDl^sN1` z`I;dmq=@d42`Qqifo3Nm#cFh8M;z$K1JiYmmG)2LHEggzj~ z9Hc)zi7^)&Fwi)CV$41#PVOVbiI7<}nqjy5E-eKM>+9`4AMTh#JY_^ zam|s;boXp{2-cack4AuCgZW=bjEDv4F zL`f}Ia#6rqYqJ+!QQ+G_lz3Nx8Xadhuai^b>nF~vAe}d6T|T)k-j?`nm6CibV~1@9 z;di7vJh^VYPTVH68Ki_dxt>ulKJUimF7|)KV)v>4o&2EQh-PstLV5ZrM2&4ykS|iP z;vxp+@shbpJcJTKVv2Q&VT%ZbD&B@0$yO3ixX)xk01>qZ7b!WmH{Ku`L{a9`F?90L zR7YSCI!w7??B7pEHs6Nm>2F0ENkmWg{DfXcH3+7s+sRKU2GAO(!X%|V)9fxkk^g`- zJcDH}30TMoZU2c$_<=H#4C{?+ZF}j(eqvG?e0a|vS6&0MDt5nm10#Tz;1R`5?d;d8 zI}KcxZ2q+W4SV?a^hz60S{wFC+(~Fv*gd~gRkL+p>M@o7mT+yL{5QRz99#Tm|G_f~ zcR`=`FQ^bT6VZ46th+7;CNTh2S;-ddJPVaH!~1iBAV#nczX==EJ5_SM>Lc~cgYTp zqO5^tr-P%_{9AZX)=phGO2>{;ox;{Ni!0(%DqY^KS@&$kNm4m>27_zT)>#lWbBKS3 zDDW$_%32lIatf7EFjHjYKr`U3Zg}?{8|cqsmm8`M;&MOcZJ%PD^(_Cv1ilIvhD~dx zH{JvfM`|YGgx_cuRQ1f9a9kL^dVrhWkzwz3J@f94EW{@(>|X#s0vFA8Gk0 zc5hqOv>>iqgQa17Ss+3!g2v7U2on*s(~%}R!`&inFML$PR}F!8!nI;2Sn#rD!#_#V-bFgpgj(HkPVN5~Klo7ULYB z!#B)$^~y3%{VZ^*&*S=wzohrHSxk#&Zo#fln9aZS2+F@n7(Gz_^HWB(UkH!_2BDQX?+6x2p01tA3oR?f5Fq(<$vftq?^On!C&jD7O3SD)4!tJ4JUmJ^(zln@=J*e)=b zX>Pc_JK25xPj>S+#7=g%n(P`T8|8Wtmgg+16p+lcQ)tNrz!UugVB3$x0$@r3;LPdM z!X6!A{p#9FsoR?O*@`>yV*eB0{+ZZ`ztc>7jV9i<_)600Ht+r#ai`ttf7+A(!!vD* zw_Of?CDI%!In{cjM^pZ0j0EKKgY>5-7l=J?8&SedC+hN~v^I_t$c6oY$>|K?pP;NW zYdQJjw0eQ9$ED;mmDOb&8hfsq*T7*y{GI(T)#1Idpdwd8KAyFMYE6Pl10f-xj0}Em z8h63;1O0<(_M`_)2BW7pBZSFO%1+GsHwvtjoi8mc6zddYQOattymW^F@lU1Zi-dwH~!o&MT33Bh#v$ zM9d~i-cejwD9x47dqZi$O2`))?550rlrvP1Qz)BdwmHhtvDm6W;C!-BX%&OLMM@e{ zYm_+FWA@0CdxsG*3O3+*D_KajcpCYzip7)Wdd;5xVPVOVX|x79^2rDtfh^w^dC*hl zfj$?-nn>SG^*jqpOufO29MObMP9eBWW8EGlh5?tgWG6{%1nb zf$~4b@C!N(axSKy*U0&IT*YDtzY2!1r-%{50*4$YwJN67Dk6`ZNHT7mMAz^_Mc_rq zS;`Nc51x{Ci5iiAP7ccnqcbWKxk@iiuo0d)BA1e|`!@s!SuRx9ag29CcyRvFiO7Nxm=h>=cfuK|<545x?n({ZtqB7q43lsq${d3QCZJH$%w5=_s9r4ckTZbXT(R+N zcB$E1Y1}(Dwz|5SZxBP9Lqgj+Y&>~zrZM?utWa;3fDvOWrPA1u@f~-MPwcvT$36Fq zk55dD-!+^Mh#h5`hFhphLTRbO(rR_wLnYtXOzN{Opw>zR z;NO~|hvL&fc{;MhU1d*^L9P=3C08(gY*l=so!)j{brK;+(QB;}i4a23Ys84~cZqP# z78grZ&Ns|uKn%4B)uXN-sy7(Uw?%*(zZFBX71bD+mff>)G&CdY zqY)2m<2SrD8Q+GV-nxktNtBCKXpfLFm7;KymWFmUq;%SkZqC#p+LA_{1QJlIG%}?s zJPHd)6t=+okJLi)q7oYiN%0hv=(s_7KzP6BgZ-^K?dWAeD~g#0jE^U2PmhcoF3m3% z-F~vzW9;0sw_nrLa6ZUl^oZ*bO8Q z;bw4?paQA_X`Guev|@WaJgjY#q%>CUhazZbB`dw!PF6akGdo>lzlI&u=KO6{2FgDb z=^!4@KWy+^jZXS%Mc3Av?zL?TPnrx>zndn*Q|6%*aK;*2D$Vsg8$-du%QgO5qq1K5 z+BQ_}5=MDN)Z~+)9P0Bx`S(*>>Yes#Cf#1v+%{I<2L_LB)F6xn;x|S1wl$cxPo93F znnXbu6wEJ`gu*n+oaigPz+1(E@3gz#)}08M^^Pg%nOrr%I>8U{PpT3iTWaJn6%9=wICcL_3ep2QncvH>6NgCw|>PBz5|8nl=e- z9thKY%KHLq0_W}Q(u3+nco|V_l1|k^nim?LE=wJlx@%diB`64<0|FP zZa0TUF^2Nz+h2X~R#R&F7q7VJqT7OpYRk)l`PJ%8FiKcB>d1tpe<4xU${Oi}sw?WF z5OC<-%Dj39SgllQ%}QxDzp_TZ_;aP1^n2qZ=`gE#VGj~XBB=?y&#(zd6@bMBdAl0K zt!bc?j(4U3?P$F?Kf`(5Ac;@Ta1zi_@fCUcYL!;AC_ZO0w*^n}IRTuuAuijI7HkoT zKj^|8j5!d_`54p4Jbw7GckO%V=*;2A9y>a-_rPHo{k9+G>qL$1nf(VIe&X=HnFD*R zw`2Rcq5V*np~T+ZM|aEbd(kMr9owGClolYz5RmgLI5gyor%R1yBOC2ZW-Laj<&#kJ zwSVu_kRF)TLg6H-3L9DTty66|Xir`BB?F!30v4dKSe&5=WC6XI(!85tSxv4CwXNxO zlpsxIMH>Wa(FA^VZ$frHO3Py!paq;klkN~!D$P87p%3lD4DJlJ=U1wW+c6mSYnmFu z(-4MKsh;2!Q2a8@f&Q4Q6syJhVyW23P^(keUimuunL)EzF}Bg5inD5C zDm%0nqzcuBMuQ>FE4el_JWSP&smEtU1I^lu%ypM_zK)8tbB(DX*hite!L&rD_%}aa zudP%IS-g@1LR(I4_|I|o>b)Axvj9!NB5DyA6Sa^{Qaj*dsm%jQ5L81ZJ-VsWM0btM zG#W!H*!LCdH6+%oU$$c8X7MaTW);W;VVo(LM<#-qmKX;pxV21M|mQC$jl;v1XD z{YvXft_L%tyu{f1w9QtUGfY2dIa#DuxI_SBg;XJkNRdJe1TKu{Co(h66Z)7D7%32s zTg(n5d6utlsRW>mISy4$baE<9{F%*PQ(0JtYqs$gU9PjwPVL2>@C%o&nzq> zB1LY;`m=*z9 znowr}_2q&*e?gv?v>3FANwFr);JO#&`3v$qJ0;mn*!NorFc;)`;mUkLp4Un_`ViD46CT96dQ1O@l0{@V=(+8m3hvDT8 zR1`Oalx51bwwAw)@&!9TC{_L(Ucv?Eg-h>LmxXSV?L_Sn<*!EThnLS4E0vih4BKbJ zi`mpcP5u{Z5;s+4J4;zL9?PZYJ){RI%;EwE`Ru^{Zbj}PQ~P{Ts2Ta-Imod|NJ$Sb z8|o6adZ=5Q8Pcjs9jmdF(*ZF_)5e|_Ca!{M?bFi|_gG!y-NkfJRt6vO;4bf>NOgCO zqO0@S0bZRK^Llku_iPk@a|gg!Q|nm7GZmkUkLu4dqF?ENT~9A`-+S6OHod6-AKACa z6U>^ZLLW@@J=&)Bakyf^f+Bpegfj>b0Pf{*4T10Sasi1D=ZmLG?8yNaiY9L4R0&5i z+OL_^wxrM{#1ZFl%dr5?B}<5L4c&!6K&a%|4_$h9pb|O)dXjGliXgsY1zu(F_b}-9 zj*Sk{TC|O~0jRG;FwD2Ao-$9LGYDA|JBOY0JabKzt_`=!H8i-a-gsJ#$Kp?!?P{>t zR0}Ty?&X+V8XGme6V~DarBktU(i)?060`H@k(G+PKT>Pqs)(>Z1xM>}E24`Og!@N} z2(?DKGexO-tE9nH{yxIBf%5m@Pn2CmNEzQ1dFcsF$3D(g%+{X%#O8}GI>7PTs$`t2 z&^CiOE$&KGU?#)XeiU~KVPXjmlF67R(`dHlu}8^yszKA{tu9S&AP7x0X(%&!;&??< zr^jrg29q*wLx6IS{}q^vx`k1wWKO{$6z!059))pvBI0cZA27Q1flB03rtMqgOF%S& z+=)S!GN2aK8pYAB;2vTVD*w6C3Psqi2NZy7xdjg~9xXI`tqS>kVb<(0#O8;`g$|oa97%_*YeuGJU?nhYA@KW91=Sr_{GGc ztig)s_#uL+=^@f@cZiqHxm2<1289MU_Wyb zS7#!Ehb=i+!;LTN8;POtS2;3>!llStja6af1KJ+#1;@G)`EBj;k2O)>SKri<67YSg zh*s7BQ);lEIo;oYZT93^e0VauHbWxUC{k8a z{Dg|@TRaiu$Ws(S(Xz52UZ|2dKL(S-!B~3_`FEqZWmAaE6gnMMg#@hsUbt%A

H#`qU+^R**^3kfi9LyTNl#+w z08CEB-TZ{xoQk`7An$Y?j&~`CV-Pz?DM~M;^b10P!zyT{o+6abOwTa291ZXT1e3Oa ze1RXNwVG-pr?zJ?@*Oal1;WCCZj#(4q4KGhP6KBUCW~)60(TJol=MvKkNy!zj=L9y z%3>bnvT_jlqxpGw>&d6#Vf!h6R6NSzQAa$PC1~2o#fUXA#v9l;Jj>RC2SxwFw7LDL z0>a@*a&xN#!?R2W#N>q^G!`#f96MK+7rQu8VNH$!a}*Gbc1$pB>y#j`YFfQgSx=8{ zX(=sEUtqU|3s+wDp5IqqF0flT69|&r@dD2UcB>5=7uYSVOBdL!V8Z7MyVb{Y)<2$r z;-#ArvZLe7F|DytdNfxr*wu07<*(Bw&1uyRykO~awIwC zO4S_8KCp=5j`Vzv1VR2{-GMgsgoX_D$ik}X2@#H@9BL8uA~tAaxlLk&zI>ldUa8|o zSa0;ImA{z)hS0__3rEf&6lw64AJCwFj;k2!ahPKRA*C_G4V5B!6TwDFq5D)CO5nnEXbUYsJTdh1sCkoa4-6WU#NmHvwl;g#COPAfF>|4D!mLI6q|Bf1--W zS~$ewzBJbCu|}m991~px3x&B7m^0i!+(4PFQY~MtjWx@~Dh6JxkCmEZ%Z0|tQgw`O zd)LRP!7_q@o)puqi~$f!81qtdxiW05{6G5!g^#6*f5T1zo8mK(P2q+~ytWMq)Hd~e z(Hk!Gd_-4Yh;-GNoQrsc2kdx-eW9BjweWAvo>p^boByV~S<&)`pwagL$8eHPHU+Eo z|4U9cDz8NO)=z-v7ivNBs}kxm9XODQ z9jKcG6-YA3=3*WBZHP-(3J6LM&@nMCA!5t&8g9?Cqm>C%j4&*9Xhq3umFXz=LMp$^ z0nzH%yIL!2AcpLgHCwBlBzIOg$+S4mF$6*|JcvoP`bpUWQQ~@wW1o$TWXNfe9bkzj z%cjO;)22x)!eVK884(UBYx*q6F#NDUzB8vv4G{d&`)FF+AIY2^!cSrML8<5U+$H;X z=*~MwGlyHGBi#=xQyoJjRsxd>v1_RPkVb$#@F4TSS&ijfPIF1tDGv9-aG5P;z_P}{!2EF)uL!H#>5xI{Bo>bo~Km4}54W^6*+kL6*Su@ix zr3QUa_V_Az=d?Es3Svzuo%Z^gQ5xjTo>408^)n-zw(4yX+HXCrcW1A3fVQ@akG+P! zd_Zsk4&yI5DFt*CF2>nais1vkf}`!cB>g*vnWYlzSSS|Z>HNYYo2Fn!ZsQUbF>Rqqq_O zHXBWR8Yq8LWYyvs(x%_mx)imXeE!%%*lBgZ$l8t;@mtvHz}mt;646m~)LDe@#;o<; z5@8A(xL2`8_HG%AFu)^62giq#rDl#}3s`1MEzvHoL`%NLH;=+3$CHMjn zlYKFi{TRw<@YvAdBCujj=#h|b1=n&`^0?|pFFDl;4rKEf&e-@}JMOw?a(sN(#pfG@r!<^}F@Yb=Q-k|^x^Y0=c zv$w73QsGF(dSz`}>n)nxRCEvW@!K|IwyhcuZ`(`NL``4%m1penz_&W`0{>2;ZbE?F zfq1(`1gl3lNF}{8jpc_OBO?~{y_Ia}`wL(OH$^~YK|g`)GZ?8I59limpjFol=&j-u zV2xS5=|0GdA*$?9$RQrgQpt}4<#(huPM?lgO+=B+cr37>;KhRPrS*F3$TS;)5Y5&) z{2&}2vUlQoYvB}~bqQ6v`Ct$J(*54rs<6`HCV#Q8j7ZAF z^ZDo|ZsL-)P9{pj6u_U7EHdEgT*lLL{O+LGWY{s%4^3pgBlNAXCvnE}{fd9LR%wa!$cP3px;pRH#Jy62v5F zh$tWwbEJmL6pI{b1fMD92|T^#5iJ9m)T_uM%TYO?4r?wo==h&8igCJNBq<|^9;Z2n z8b%RPk=aVNM>c+FVTz|Pblj-28YfaK@I@pqsH|zpY`>W`&cGXM@ky? zHxY#L5iw7heW@&gHKCLm?0H6(mV0;K2w}}9g~s}s%Pn!JX~GWF+*|7*g z#8a<}L(R~!-PHmd^X3c2%Wb|k*T@&-Yt~N=kBU9!s>Uu;Rn>vOyfD)??=Ziig`_G9 zp(uK*@`P$1lC2VYy8*&&vIpjRtV@?)#jdZ1hH&gO}-Th-o{up0UK zSk}O7U)4)<|O6j!M&x~?CZ5eCmJy!IM(HxvE;hm`Xf_ySeL46)5zb~@Gy3=+KROLHDK;oU0uzC%!#j#-A|Mp z6Q>`S$zBs}PXV?4gMn}RH!RopCI+=&zQc<7#*W+O#dnzH`FKLMN5?0t+D8*sBPMHA z`_6>bh{;;jtX0-bHvAZ(juj;9bh#>>J}jn3JW#Pj!vP&d4>7q6mua9?>k14w=qUtx zWe-_uc7ZM!QG+cUq;Ls>+QS`^IV2HBe8M@{p^C=%afLNPYZE)KvF5)VS#uu)^OSkW zVPG4Zvo)$zn(KKs7VCPTPxxDN>AR_Z=5d+E1GCr7J5S5;2Z*fMuimXtf{oK0*jgkI zLeSbRPY3WVD+$LqVmp+C68W(sK5jVC^L;SZ2tXAC%!Vlk&gTfmYHbID+t+!zv{u0T z8FT-OD6SiJWF10~O%SqSIG(GU(QQdOq=gbIJG`-GyljXhL5a)r@c z-b-?Y@jiWpaq?ceKtXs84RW0W@v`Jgq}iyyY!5KSlKX1ItZiS-U@_mk*!{+`y7;>J zRqc^?Jy)?V@HMRKB*-8}nnGv{EpU*Yju-=Mq^+t5&NaA}Jv}^#4E1#Jaqx6-9DDuP zIO5?9%7Y$V#N%@uu2#&~mXR2Fc@~%DB;&*uT?Ff@nwgR%BvTqw?Pq}9k=}E zNs&1hiS<8t8e=InajS5C9&I<8wK?c9sa&;Ig8q|k8P?}%ou!sj28ece-{5h|2Y2Ss zdKMEI9?WNsVsyrtOePtShip=FYqv5)!l0Pa1Tz{5Vb9>&AnZt$7F48J9@)hBxQG?8 zf~>a9hIKn|T)CDc1-?Zfd#?+mxdRwrUlMg=bY4tLtI%9LXnf^c@?(+eUFoF?fqAh( zK-E2y%u6K>S##2_VuyN#g?MVj&O3MDd(s^E(Q|?578K4+`*9#G|G@-sLom)wYi2mS z@K2-(zqfba(WdJ#u4L?-W^XfdI&B)}IW49+r#<}>Av3%%jn?8|6JF&IJ++An~r7a}o8+^pR|IMi?tWxPcIM#n%|WuI_!zayTD7(e4U#!s-?cFa zSR`blEns*4xke_Ldti*j}(0-@`iR=Uo# zSJPIFZhug31_cKecb_~*D{f$Pu6ppSXgs~O!0@LXXo9^!DsF;R?N4;IE|%^0J|SYf z)QiG(cx1#rvO{r|ZIHfo5&4UFi|Ji2pDkQ$AZbsrA?|Y$f{?C@&u-Ay-36yLl1~gP)GthUiqCt59h-T)4tX7r`i#Fl7CA2cquM}eSc{)0S zF6Ze0fpHJ?wBgw9Wj5S9d%(CmOf`dX-SL~;1}z$@#HiM6`J=-Y51z?T7{=qXwaPs1 zwc^j2we?Z^@puz*OA&9^hauD+U#J)6&P=SIsjg>^BY^tMnbOqw`iV2u+{8KpoR@OC z_)KZ|#N=p35wUaLkH!}kf1d$xyKUugD-w6r3oB>86~9<5dG-*A#Mt8Hj{rUZ-sGZY z-v)R_Wg32pRD85?2M?c^9E6pSk~u2nXg&i^!a)fP8=QdSOo=s_R8@M~M7Qr?Dc4*= zlLby`DpZM9tq38gw~Au-lP8YIQz=IyMh4SiWY^QRmdn)eEpBDr0lCW&e)P2F0^-I_ zi>#p2xqu6@Z%-3sw{YgZ&KPD*Gqz|f2redcRnrv@?$lmLb81)G`@))s`#yHiTR^SU znEAIm=WE@-LS*+X-IlO zm~;NWSo=OX=iyIjI&PN4;+*dkCPQ5K|JCl`6fuducES4(ktP!{#@Rp8098X`#@Hd2iYlX6Tl^-nIxs8dZldwb zae>I#AhsA7crYJZB(EjE-YdEgyROtXLe{tBlv>hE&miRV6s+G*Ti3@36O|-~2!pr= zsPpMUTVS8#@`1cDLNT9MkyoOWD^^=tXCR|6S|&r?2kwC_m5{_&C;>&NT9Jg7uF$BF zAR1$4uSj3Vawi=a$($`bzoGrW>9AF4Qr6%)B|1jjTUfdgIa4@N&((?xaNH&bZt}`O z-Yp8bE4;=#n9Rv|A9+=z$Nm}=p@%GD`1nGzSLPz1I2C-sl88Kc64e~C4Q&du0V0hm zD}qV9&TmzE1iJsLXuBfB!~rVB(~`d8T8_5pjArPXjXFro7Z=HW805PUEaIY0)hcPI z@k1#L-XfN%G>3J0A^|P)lo9twl-h!EO*rcH1V(EAL z8!kU))?ZDVj~hNe9aSLtS;K0bsBIf|ea~ZXp%fg&=nZfgaPTxcnC5nj?tq7>@bm=r zMMze{HS$fBCcR*=It4$BLEF_*gat<+zE%ZFTat(^$yg|WsTJ;4IA@|2sA**+b9JA< zkk&VM7^Q_?9gASeKOR|f_hfm6HC2U8&O41vd}Mzi(pYEwA@%lW2UxxR-DZ*1V0e+= z))lnY^kNFePVy37B{eTn(hWw_wr&BSORy|vIRr0#6)G^O4R?hGdKyMht&?&7!E)49oE z`f{h`@Fo1m;V2%!!D}=)o##fi+w>b?q2g<+PTi#cIZ$6}Y63Dd-t|WW#21+p@Sxjf~daG#4Y|l`= z!5-OMb`R_hh41QOeMO|E3(SbGj3VbRe!m8z>Y2&;J2pH^jU3^|+wj04o>o{>Q7eh5 zm=%0}WTb%2m=TU`xRAn0x!AK$$CQCH-wKXKaN{X0Qa)igP@E11ry&za{|5*ZRCA%Z z)aEhhnU!~nq7bE2M>=uJRVO(u7DUr6+!k3%%hhLdK~oC)X`sy?2FfFms$H>w zoq1mRHH`wQUqXS-V^X`-t4oS>k1`WI1KgU`1}X!2)I5}Q>1?8K)3lNv@Jj=bY1zKW z@Vi1!?V%0;44*nzSfj^$?z->*F@6VHfbi|W>3g6rN8C-b2|-4c-(6PR87-FR9j=br z8Z3C>K}_dO)s|G8_phjH%km>59GZP(WEhvK=XFRcxJvP5E>dnIDm&Cp!sQ970OcBs z(zZU`a<9U@KJu?>)6)pqp-sWS7tx?z$IY-gg!dC+0Oe7lu(E=Y+O%TU&!TIwo5e{zAu&xU;pGcj zaf_f89r41NLu^q@u&oDk*Eiw?DP(5%rGc0G9u2=gEy#t?c=U+c77R8}xjR%Y;a#K1 zqkJ-OP@f0NXHx^fCJQmkL)Sw3@vK`{?6LUKuYav);HBHEw zmZ43fp=KNSh0xhe;aNER0Fx5<4MwHOpNY^9n`_M(u7#gwCv2t<1La?eRE_6auUwy3 zc6`;Zk87!K-MN-2rMcFg{!!1GW*V&lOwn<=GVtb!U5d)nPYq<2g1^uUgq0NB0m##j zVTWGWRP0}o+NRRWbpg7@dnGmKdNxlyqq5N7Fe+0m98-qi+>RV83540cRLv3P&w+pB z6i4ANmeqb-RSjS{7YYXEIiy!0QcJu?)#bBDdyo@js#d0YYiJKpUi#dQojC-T5E+IL zu{_tT!8^B@LyGNsqnNAI77>*kmlt`YEAZyW13B~{kgphOHoO|SktWN9(=*VWz^VOg zcr{LIOV*1bDWSsM)LpP+=e;1)q#|=mv3lbo(JD+5<~lgk8wY zE^4d{w~8p3XP1V_j=w3qdai~0G1cabe19msQoVr+xOnzZc$o|>(>NPmiwl~FURi16 z*$)b_jL(K!pX68_XTS5TyrKBOq407UkE2#c{mZdhSSNVp{SwLqA5?(T2oNXD9z8{>c&Fp^x37Y;njs!vo>R1g_}#jrmOz{y3Q>&iYgA{glkJL z+ioe+Ktq98O9+%|@y11bz!1>CAOuCB!(OJ{ownO$XSPxU5{X7C69}?4X+lCoQ6Ef9 zj7FpHCK`>##0QNA6CZr>5-!18O#HrYy4#EJ&|m)NbY{-XIp6u_%*@W2ba626R1}9A zRR+3>TfkIUY_phh(BpGH+kG-C@4J{z-8nKyEj8c?#E|bgQ`eL{Qh|dHm}UcMGv;6( z8_q3<1DV*%1?8TQSsm8mV9@Pzx_g0*)=ezt@Ei7i3B9>K@8%tS^BDCiN!2nZnLyDL zb=X#Z>dg58qZyg-QM10zv^(61CH>A)*vvJ!MXupsj#hlwCqS)MA%E@d>$_1c^@ZNl zjAEEGSxkGm7+NFhCe@EmFlF7KH!2M8RqGvU2roz^(F@Xe^?aK_{4}zULFhs4_VpNPS zb-2h^y0K=H(oxbG7~*WYs>@cq;Z37+3sKKP+tGp?tleMaJ4f}fpCB8cqM{x@3b_oW zvbkO~df5WvGzGtyoW5MGi#e3i4hHJ{{P`%qS+iyF+;gYbA{J*b#-m%E{|R(Qk4;r z2CxZq_nA+Fbnf|InNxq%<~dl6=MfxKdosxaW~!LJokeQ)LULXoKOEjWgjbr3z!+j9o{S~>ivU( zzn8&)-E)bri7$yUynpYVBjR&_7_nq7&`kV35BP;RIv*Gzo?HMtPFx}`5`GnsC*C36 zCYDzN%ZQK|B@QhF4iX>M0Php`)B@eamO5Y~afSGeD8_&S@qGjEE%7hzC3$}nCl&$6 ziI#V=24)77$$sT`z~M`5N?S^DFfMOG`$(0X-?b)*R|mX>Bq3GQ^c-4z$3&5dx7_WkONA$fHNK(8J#%#;)8A%L?vSA zgL%3&Wd-JU!{&BBfmMyTu^+fbYWQscU^8*U1FiwZu$sLg7|BPyO#Q*s7fk)Y)CWxZH)-Fbdz0o(dN*m^ zq;r$TP5L%z+oWrgrcHV_Y1yP>lZH+DHEGwRTa#u@dNpa)q*IedP5Lxx)1*t2CQW+8 z;0rP|>CmJ>lm1NFGwIHxIg{Q@S~Ka)q%o7ea72)mNmnLKne>FxAtUo%>L;lMlMYN8 zFz=M|*{G%lH8mz`L-}ks)T+q#GPRYdolI?nwa_$H%(UQqA#=)Xk~3w-B3tUGH_4X1 zVI*4`pN&Yi)I0}~Y-uT7O15-$9wOP&@%e~kO9$yqvZc2e$(DxcO|qqT8OfGb(wk&U zXBf$rQVS8umOidQBwMpIwse^;C0m*mMr&asI63f*a(Ky5UM?D2 zP0fjx!=M~PQNEL38P{sjMKdxhOIJ_ZVs*JxtrVZOeO&ex0$yh}&)n9;e`bYMxS*Lx zRO8|xftv>oMlQa_o!GT_<+^2Bc1IIN(pIF#*_vX>NXP+5pI-uR>?dZnH?NTp+Sm}S>~C{^9hxZ`51C= z%=64K90%Y2ci+$bKHoFF!}`{{-nG8Bby}wX@Ei7T?`vQC+RyJO$NBAP_a7)WU6>Z|+eBaI(rFh@L)(Y)lW&Yr_ld+w>IogJo zo1gpImD6Sp4z~6pJUrHaUcimEGv(>Nzf=JaIbeHR+a86YyN~?el_HUBjzWnOWt83*RBH5+<+1c!UGmzOG;U=w zsilYB!an(U;*R9rGZ#rOLgLOzzSN|m9Om)=AhVa_r2OeOawbjlbl={y2R#hoRao*6 zW20jgF>P9kS`G_ajt-jNVD&ZD4uN5z_6QC4Azx2xP{G7eSN9Dvz?ZM5DZ5dqPbZ%m z!Vf3C7pUQj(P8W^6zblgY99FEmE`(f6w2#bAab`^)@x?rxCon4Ik&Yrbse4XZE_dx zYndF%>%35L#Wk;xx%I{2(m@|4H4TlV+}vLs=lk_JIXRaw6(a8IIU)`du@UG>?Jj+1 zeXVKSc(mI}t1R2I<5;4{n!d-zXlUtZh%P0GP1NZ(O+$}@tm2`Ajqk6W&2ZXc(U1ZJ5M7+i@hONODOtd}C!w#*xna=O z3>Hzb@&qcSb~j4udz5HbeO=v0c?AWJb}*a=eJ~jfC8gJ-Z)XG68+Se*Rn{WzY_DZn z6AHCMjJ5aEGUsHKFUe1L>Y_;z^F_F=OwushPB*&e>4Bpe=M+NLSv^;On0B5E)q7B} z*6*gst!p{c{jszAqmf6z4lIz4u&{6^nKZpMrd#He$q&;e4m%S)x~6{QB^Oxz__ZPT zN_uH&53+BWY}*UI8m_LcpR7ABUAmMDt7jy(`1vSyb1?2Dckb%UkKqUn9UY6&kr6G4 z%^%I~o2JWSjh#97Yo1^#3;lU;lw3Ghqa;k`k33UdkImJB7cX8=dfo*E8R=st6FUue z$eQrF>6bi}Ir|QsY?+uwdHJ-=?(%wMXK(K~ThKL?+rZuag~2cbwotdr(6eH$a?^%* zJuxv614iN4=h&5sCRPR}eEX5)D$Uf(N|hPtSp|+Dto9M5)B6kvrNax_pQhiMjh+AJjTCKZmf-ggNhg>6T19v z;E1h|yD5JBdzdO(6c(Jk7~pwD4ooi{PDks%&GQ78PC{O@xNb*B2XZ@UG}Le^t87jM z49;WOswZF0AX^h`{v$D0472iv!@9p*duwwIZOfFYThwV3AeN*oz99`a+lfT=lsRYr z`s=UNHqI(4f$$6wE`xHd&9!;W+Se@Lh0-eA)}{g#xZiRFzk4^)z=Oln2=)#XS&cVE zb!6#_B2UJVJF<6Y?SQ3_*M|f6T^8;YSaj6S7dT9JDnfNjK|DxdM3B0+Vqw3 zZ&1c#JhQSZ$!P{m`rM%y(o|w~^-UHDtX^tr>aMP{Q4=qAp$5TqUe$oJZqgK)8nZQH zS_&TRU1;dITUL3wUf_!14#}_*n;$qr!UTtfrPkNi@9Wal)=ouS;Hgum5dZZnFYh@I zK@J@b8Gpt^xLYY*BeRXgx%UGYg#KCC*BM=NVCTxPWXrmRj?g?#E%gVgx4uOXvP{4wsIx)!L z)e7f7@xamZ6N|NMT_Ie$?xyyZ%`MxjMYs-SPjdU4&W*YYF=oxtV&T6Uzu;N#Nwn@iQC{9zhHQrxVf!=8pEN4!-(CE)tf?RXDtpG{H1 zQO~jL=}(XMkj501ltLVr45YvT_c;^Q;a|`!7Z+DYkg|Acw3yq}E8eCBvC!b)L~i}k zU#!+%9KXD>)W9QH+~1k0o9?-@4IVMH+#s81KVP|-HP#$MG%I_ePWT!ZcLq0UQ=}{E zJcoxvIOE*OfSQqB-+QmT(3&vyl?jbcQjExVoYs)L(3I0}*TAJyXcgu()6G69E#xs8 zteJnmW*;%f@Ufml>n#gnWd-8E<7ETn0}oPDt7~a#q5Bcj9~~XFE9<+&$=UjjQ+4Xu znCcc~TtXA}g_JC@B^NK_zvR|5XglGI+l(r-`Vkj=S(BWE?kiQdA8#V*yLY~sY)?j3 zK_}-X>X`Kax+44 z8N?EpoG079Wb>f4wZph^6uSb;E~T)rFrAMNo~fv+7C20*R4mnBUUIpPxBv)X*SE$5 zajn{>QHfxA6W}n~Agri^!Or!SO>b|`Ywm1s6<8Al8l%N@4c!-?HHoe1MS;U0k42!7 zZ~>QDf^i z&m;(Il-h6EpKl%ndz20Cl=JLasmaO7@awL2@eGq4J(V71(XL~sX8S92VGi<&iX9L_ zNkK6b>A-xuQF&PDR0w8KqRyS>$x%rR1};_3f~85(+3I)h3|Ph+Yu7h5b;2g#%bd}2 z7rtJ?F4bLt_24!(F}cLctEzl`Q954YIx3d)4;*Ae2his)5b5^&4^5rfhS|@k87=_= zQA$?LGigi{OmWRrPuDPSjk|@_6jEp#`tq{m=1I7XM&3^xck9+IMhR>=a%yH~tVYMj zrvp4G@nDq6`T6;n(n=BcAg{BEMk$e|Iu!Qq-TTN2xm_o`cx{7=hes3MTA}M@Vq#(e zMn=QE1|d{&V`C#NE2}fiG&&U)yRUFpH$1Q9*DqV&=NE@-&&hFzHlRxzMwi5#)7q6K z^v5IZUq^dxyCSwhM&inJHqHQ!$su#XftJ2g@V0Bm)_8FR@w~--Hpm$&t9r|X4EuwW zFPSxfSy%;$jY~^QD|l7qy%pBJH6Lnx1L}qs1;lv63lcbz+uC(i(NcHNdB1<~odl&v zbU1K`7F}6U;3O$YaGX=Y%}8b^>lWEu;^Lxl-DC~WJrr z-uuiU3{=*J7@LE`XZ-cy?N1vqm?=&cQK#?dp6#v8j#Sm;l)|IPGV9)aO6={`v+gav zd+yx1zyyBF&To?r7V8UxfeB%}rWSw#DmRL3`qIGv_N}?JCn=Igf|MPG{RO|~BjDN0 zV5omqGLXDFn+{>$atfbXDOwJ;w_#&@S6w|ZJe-k8DeA|h8(=2SRcx&i;Be($Q2=vq zAE@#kCLMuCv>f92%+f!UT28s>;dAIsO#D|I_SJbCNW-n@bv4eQ}G{-YAx30`XsrEAhwVuTBQIBys)E2 z4r}kZ+Ol1NYHDf;r+7`wM(cyYn`vONmFJZrQ^22ca&cu=jb6HZne?5v`E%9>#85kr z41J#mm&ETdaR>09LZA-~(|O1gC=`lJXoRS9mQDHM*OU2&;aV<9I28bINR#cgYu%F$ zX`0yvmo8p3vke1xng+>)V6Uyzm2fJ^0#BTt>tn5NXi&d<_X7Y^+$8v7aLsO?j3LId zV%P7Btq*wH2{F@KOK@_sFG<8@6gJ@rhwy|ezkga&?0ake&g~dCgrpkcuz=D{KNFOd z(JYH1E$)9hh&$+~m*)lr!SM!;IF)TBmG#*W0YMXShS?RX;ea{sH>}dJiNoY98%p04 z2L?vUE29-FX#R8+D!ORVFJcsvhYwTq%bZ}DmLm)xJx|d|bX3MMxk^B3Hga6j+g!7rJRJh8%W=?&3R+g+d z4IvOMAN{Ywttvvym zcL(o_v21qsewmhbsRNJ4>)JrjAwvS^zJtS%EV%Sb z`I4;nqfj@N2Xn48yl_7IzwYIK>Y@K{@JX*mTm)m0G`PTa@aFI4V*xq42;}~Inl=H3 zNGXJJcR;uXfT0ElnaQSDQn1fK^_4CS9Pu#@_G4)((8zgdtzkE8en&YzgfQCgx?Z=NFfeH6( zd;kMnl#q}Zpa>G(^K42+PY{Oz!Cw*+(|22)d5$$gJe}1!>2&cTRUdu5+HB|CeL(Yg~WhyEv4quLIeom-Bd|fmIcph-k>RAO#XE zodFP@4&DU;lRV`MPedH2jx{DBr!4HH+qxK*Qp`~|YPX^2T%uPw&S*Ctmc?h<>>}BC2TKKOK{Fd=mf{NC*QtiU_lUeX~xR(Uz}j z23sHt&+bUqw5)ovH(b=&2JQ!+)2!TxG#L^$GjcLiw%;2B8Hu<(#N$-54T3Vt}LvFtPxRPQxALjrC zU_2i3pRFVoA=?|K(XJ-n2&*pTX{m!XMhE`_H%?6X+n$s54f}UdQ2?u0TBrf0Zf!k2 zK)emIBR;#4Z|@74U;`QzZboh(Ij2c;bS8i;^4vlQq`^#08EI+q{QUft8_B>O0i~x7 zs4lo>@C{+M6CRRgnN@P$fCsi6)#z1|d*yP!zhCmwQat6yXH;FClJUd*#D1UHev=PB zKdG5TQAgqkd*JHB)c-D9`JX%f^pDoLxVbwgTH_sog;WR>-&#>^j1m7@{* zDV$$HHCag=gDK@mHEoW@fUTUdLD2raBEOn>h~0emKB0(4fObblMs8~&G-@~Z&_YCb zc-o5>2Zfnw<=TLtc?SsbQ4RbJRd&n$g;V`I&hV@EpHz~kF85)kGhwe7JP_*j&6_i{ zI3Pp1mYXHAfexvkCQSh=r2?Sd+Kdr-DUYtmc(hZGIirX}3QIazG%D`=kT|Z!m#zyK zrxc)GMSGB?l2-gYxCSI^2LRek`m-_+ARtm9l8E;Fq`vF#1NT<@;yP=<_VyZX z8jM#{Qu49*_PUX%29x9HsI0o`u)j0!VdHvPvG(qc?iPpDSCy)o}OEZ+aRv5TmXVnf#>=}H57&e@g?lvL# z>nElAzQm2?am#JW1`LGUzg9UO5gihe1gR3?k1ObjgR@BL>ao^ms_FLNwdm*fk41Nf zJ;Ox&{yE3!?DEnjk_cFB0#>69fQFj7`e=l>X;b8FP4Ka(?n^_|{j4qnyMV7UgCt|B zJvl&phhjgs06XIK^w^{Iz7qT7xw3h2$hS3yzxuZao>NfM)3cH#z}+a_Tg+Hsr8qdx zox25?<8Fk2HE?>{eoR02lMC#iuhg*}9K1t^x~BQ}_t$`cZFe1yGX2Z#*L$S&bqlTT z-n#V+yI#X~NmzJDkL9Qym&Lv;S^;X(A0LIA{;LC|PR|1ws| z5eDn=(uM;9o)^&Rg8zCBcs_`8GQC&fZJ%uDW8$KKyFwUvh!yfkJ_cU)Rt!?%;7IVR z&IaxbVZ4FjGXUcY?wann#h{-6bQ%GF9eIK{{WBs|4dCD@L0k-igMh$e@Hfhw=&iuIwF z2;Q?YKL&q0OvC<+h*Vc*y6-~P3t4Cn&J#CZ^6uR+F0GtTtAFeTCk4)>1~8~d`mc+w z7-hib<>ir}7SaB1ml(J&enFg>`&r?4@7}FP+n4o0(z;qm3-8IfdwbA%X`~JTt4M|d zh~Ju6sSjj;nz3;raED0{Uqy9he}QYyoM^jaD(1F|hkA$_@Dz}Dr-Sz&t_@hz|78t^ zE)(w$mw9pBWF0Rus(sBU?p{2Y-(*c%EV zSMI|6($doKh1R-?5dw+~dbW~(qN66tiQ~u3VLdRwA(4P=w8W>}Meq?+IF8tEYv-9v z!Y{kv#O6Qva?oREli1}+sr8%y2H={FFw7x5=qVNMB?C`q%Ll38s3$ruU~WNEq(+Avo{A19(>TM2e&E3*-efHj3;4F5@!7%bUb zIEu($0bEpI!qn8%gaBeAjXJ%X6o|sW9@U9s9{!Y0GkB2oLz|?wjG?!e92d>kVIg>rl1a000TR`wA$XT z>sx*(oae8x1SDEXNlEg7y*5Qe2Y^LxS=WI6Pz8pD|92W&y}RP}O}jad;7q7F_`6UqhH?oe=tr)H7(;4kmD5D^iP4g_RcLIN;} zCXI~3QA#vy;(CF6_G7;(#)$U`?m*^@a9NW;k>Y{lgjIC1hVliQC*{%gaBy&5CTAck z=0M6d;bLxuNlH;uWi`462GMP8Z5(pISOaUn*eC(d3l0snq^vJur^A5O{EhGvZhwLB zdwgEzv#RUHUGkw*r-gr{zIs)NCmjAW=PV!M~ez&~YAau7m0(n(OR%4CAayO1!(LRAaFe6(DE5pLMOJ8X(&;4#PmjDRF~fqYioR{;NK+;R(M4LDG9w{mm2E6 z1oO)01E*_mZ(s0eSO&n_ZKU{p7JTnujjuu=Qul)rP3^~LaHZLSzomwMk1cB82R{oov%{Yl^q{0|^-&I;EbRi=`GUi zSBg>>GxHm<{amIF6Q%N6VA|ECYZ<)S;vwIoIutP`ET|xBB6eo21LtSd09PEQ>SL`Q z!B}RtU3k4rT^aZ;KUsS6Xl)&-vqz}l5L`gRPVuS9eFqP)bE()BXY>STMcS)gm`nUCh?@l@1&fuX~5FHBe>ymLdC4c+b3e!z8|SKepKg2X`It zSztO3<8B!&PfV+;UyU#b3eK!?RX8;c+L+V39tFLmD$d@1)~l&+KSYb(T#G8MA$+Qe zTm(#V`_5D1ZDHm#R*JVv5duvlV}B;6f6>eQmt0qI#Unng5ETK2Zzbq#m|Hrg<#63# zjwe!fdHQt0lJ}pidye(bD_+~s3D{K>Xux7`Jkyec-HBUyK8)-H7qTy>uRk>`(+X*0 zEmfw?7941&&DOMLt2E5b_!a`u@J2TkPC)L5-ktfG+0;MCb-$gxgZfiJ zgY@jo<)5$E(>hyaGMDbjzNX$4InkNpNqjDzf1!-a;iSv`n)|AO$;;Xz5)u8*l=JQ<_o}dp^pa1#vqx4uK zj**;Gqx2__m>@&*I(dfj_4dX4yeH0Heaa^S)^UTnYA^bl3X=enYEUp&ZB=gft51hV zCXT&Y17<+yNHzL&NaDL{MT-KmW8yKJT+`3q_o~OX=q}HG*fbK@Eu|W-dbJ7ES7zoD ztF|x9oP<+oB9N`BI~TK=)b-GKdSVYQ6ezs!viHnv9bS~0x|i%>hRfNhaE|!?l|ejj zc7ZzB(VvCu{0*wE5i=h}w8*rpJ#( zKkY(!h4?JEsazEF@C&)e^WtS4)uG;9kPGDBcuEX9GL%GjfN!_d845ytW?CepV;#P&)G{mHm1tza%UFDo zTjNOD&ZB&3qwA)SJ+;8Fr4t(@?JI1rJSMZrdQWE-OkG{LS%6!WPb7*<&SqDrfNgxE zf&@O_mbO7YAAcN*>%dId%U{i`NcE=Cr&T#9n7UPg+44IdOO>DM6)C<=$S z3*mmJF4O^H2@_q;4!AV6J-^f3Hye&CC1tH`$h?&LSyO~Ufo_-ibdSyp85{E9G(R5Z zm&<(}M?UO)%zSeF=zQf=_}C`nF4PTgD%@MDGw)^W2DPet^&7vk?O6=bE9M+4dIP89 zxwkm({=vd#vtowVkDLo8Y-fzA1V0Jr@g0mkb%x@@o^X$`>6uX@=}d_aN6W6ZDdti3 z-XjwQ0!&=4F$Du%UvI^u!Y^~XP#4sc8>iNMqK;suD|TPLfPF0Rj=J^%_O0Hr&Eps3 zyJ(1ej}Te!lD~LJWf@+z^2mHNH_dXn730ZKz4zMlrTze)o};MecCBXLr>(z@^WHsI zJ8N{avGy}?)Fi3_iM7tD3;b;ZDZaF z`Lb>~yoWyX@qJl5&Dw}MKJRT5_e@Innd|P;;SaRuLn!e{C4Ou0;9KkdPwq%ur{CXx z`WL30%dDH;ua&=NI{OHB^Ngji#-0_rA>&|ce)@Y9+1Pygu#JGvV=bQ~ug#4}d&ilt zAD2vwWd#orSYnk@%N!>Y>nGlvi19Ys?IR_6ZfL>(<3TAdvaI5nlr84@EuGhBy80>G z((oG=GM|yf?#5qcAWUqcR%0b$x7qZLA@I z$@6W(M5JTD(5WqrbOTj#!gIwY=k*c>HOIb`Btjxpf5U~YP)ehC;^@Qla%YJSmpF&p zFG&H6HKiKAkeRzgNN(SX7nmm9K7`>yPmT-j!oTis!86>`D%#3i-X|tNK3@>EmBOq@ z5lF50HfP&^BvH)4vt_ue(3MQ*zt4~}mmOElNECW<&4uZpeZl6qS=+DXz5dA=Id9o} z8Oga+SH*@tMk>qy&^wI3YLyw|s*-7_ePTOCKcc(Y(PM2pckNY({$@s1X{b9l$cx@F zgr<(Yu&lN*6?)=fJsB@s@0!uJ)*Zse7?Rkjyd!BH?Wxq1yJt0PlcTn1zjx%kF0OW$ zXF}woq!HVTlM!3&Mf@$~w!B=&k;g<&L4gwg{hRkBVs_zpy=mAkjN5OGauw8j>dS4e zZbVzu>0b^fUt`Uj|KRue;vJcu#`6=S5w16Aj}Ho}U64ha&CEO$D%zr1-zSzA&#}RB za<@R;w0chk@4Ni9X_;{a?hQTb?S5gR~pRo7TVHXVFtKP$b%r`}}>z zI_dg({8?DkB48ziu#E-s9?SmZc3I*vJSL*&Le<=Kq4IT=dD*wbst?cicwXHs$$Dc$ zp&wx|;}|yV>eGm4-+tH^uoaVTL_Rl{(H@-Cx-s>OA?K%E;-9_Wg8Px*oJH@(8g`#z z&7e)*&u^%*XakN~g%ryUfr@t_;N2eL(~i!s*p&q3lb7dDnfe@0VLG}*S9sXl%E(ka zF=*4cT8pJ@OViHK)v@8g1X;YLPZ=rV^A?V@$guES0VhyBCK> z4E?7oAXKXJyFcyKG%(u1Dtg66UF~E2JLO3mNY#h~6d>?gBZcip>`I_4%5U`b5NVz$ zF*573X?j$2^OXQ8d3#v!wfb~u$kSdK(WiKi_&KIhqYDLPo0{xf&#&UQ@ro>0rd)Tn zj(VQ3PU!dVyr~i!kUZ*8?}jn?w8yp2m~O{BtXGqcdST@GR>;ye5x<)HLlO5{#-(ZI zVuAV5b%l+G7o9R2H6zAjs9pOMD;REyoc?9;!o%h;7Ls+VDWJBNwdlZ+>7+9aDw~6v zw|Z@K_qIl*aPn*a5W(z7UBI?<9?* z`JRfE%;Jo;&r67xW!iH8#mZ7jpX%nsiLoNB6c0P)4RHmsC!u4`8E0+xQOdGEWO$(J zJTBsjZw8z{~ba-pp3weq)M#*+qlRb3|=4%_{jpX8o% zw~0rg1M{Upk_ThfNVq^=*_mZ=O;QqF)Qk?gxkardlr5H#5jgZw#yX={#ZTIbC^1A8F}W*w^-B52F}*(3}uYi6lAITZq{L#mb7zB}ES zsj8!+Q?Y>qwx5Pc;J0S+&5V7t%qaolvY1unUMUQvPdP`dd#4_3Ld z*%LEv#hm_IEontOH8kFJb#{3CI;)@~>Y%_q0%yyuinlYvcet<9jN6~GdM^|_`deTP z!S=Sfi*Rz`QP}I#hfkSj%%!;OKGH-GK#To^t7es5d&uK{j#jz+G40o zxGpL&cKBdWSXteQ5HdrH0JHO_|K_j2SQVQtZ5sb6srgW(HY!IzuLg zwm|fTzMSGOhLdI7Oa+WRXqt~_gw77PwujY9RkuBcs1X%4X-r{*>bZL|Wzi`^blZGE1_&-G7|Y$ECF#KlRxCvT_3H0Xr1 zr^c^I^iC8o&>Xn19<9}suHosLXTZh6&u_uIbE?2~{jpcWmucF#0tuJ18{8I~BS+;u zoUq=4ugs^;-%+es6`+%7831+DWH`h3LPkRKpaH`m#$K_dN>k-qm8M1cmVd>Vz;cY- z`v~WMR$yf~c5Jme{n-dh|&!E@5u&n$jQ!^G>5!QW)3-* zuE@kvZ6fmw$&5e$!sS1`y8qC*-|!EO z)0c{HFCR37_hY3P9{FHyEa+xE-=2B9^w61ZeyIKGmosKDr94rJD>j@!`NnW@J*bjxWwBp584lz{l5+E= zBK2CWMFQlA(s%}OwZ|6p-!0KI+1y!_U`cBNA=Uo#rtk#rY3IhHnB0=n3ES)Z^sLKD z);Jc4KqJzr7n^2uwE=9Q93^_(xJi)yt?9)Y-weZDlapI)-04gVjXoKa|Mv1{u^)Zv zV%9#dvGZ;gTLWA24;|++!h0grZ`(PTmv7$GuiUbqwUzVl5`C}3R~!{| zzamvcM`HbPAj|96wTO;*-wXvFlE<9@F-l5L|F4~^5)y396;hq2Ouv!p6=*7P2=FOn zea&skS<>Wo`kjct7dI1G;c0`um_B_utgZ$~TV}pq>?bDsEC#5k}UIeNod+pHSSHBV0jL*y@t0ulXmY$@<6~7q8GDCha ziJxZ@?Ris<?YB}Ic?8k?*N+9ZtawADii!Bzbu6Iv|w(|OFaEHPUn&wrzrG)W;~ zMJXYeaf%&K4p44#6=o^MW2@-@`ijyY!CQhEemBw;-ZkVm_1_bs6*>?&5n+ z0e`O@(r&JOp~-T6$6iGFx?#g~L54pQm*s|C0|w%BfVaM;>Q0krv!`xVck#7iN8Ao} zWHWV6e4@W%an^ClAP+?i{Uk?Zhfem3v(u z-CftS*YM5vP7h|3%~j?ui&=N!T$iie{GG6-zGdk7(uQ1<#)BRuH;96Rg@brc+=ikp zaoK5h{08wH4NJOrpNUGKai-ozAGxOSP;#L}+ZN$YU)*VS8Mg@{4mVdkOsg2V-ly*k zs;OXT&4Kb-z7Y_bGwdQ$=ekI1!!^IMKM zi153w+sd_l598(cAUhC$DpQc&7(^t~_MFM!7y|>+VR8*E3Qq_RC^lR;FZWm4EFw!k z2wmL;i&N6lR1~8g9V&B}j6-Dn_G91gLSq0FL7)rb*b)sl+Y-4ug|uI=9&$E^(Y`ubpLfAekY@x`&&Ve-Zw#^A1`>{ zE6Ys_)khP3m%(Qmj~;!5UYP<=s!jV!P(FZokluD1RKjgjUD+`OuFED!qf8B*L_W;g z6zK!IX31O4U{L|>2r)XBH8K{&uw!AS84%S!gL?0!oEv@EooW4PkZU0QI2na<2sfCk=CR^Y~@L+b$ieBHu4B87wDSK7jHv#9=7p z8?=UjENa)dVW~rEZeAY0?Z6GWHfT(WgFY%Q-NGvxiZZ{zaOV3f_$}}kiNl~LnuK~E z=qmkg+0%c3@mtRImXt1jWio2H+0BW^;Gj1QdeMUCW#j@K2-Za#9%FBdVTV9H@GdGU z6AWBFP{esI1L@P+n)eJ)52Akz+uaz(dtn0i`3Ew{zg13k+$-k1cyS@ZFc5UjQYS6W z!{b2W&6T`dR^$olQVuyN4!2rgw4%4DIERR*p%H3G`9mYJR8SH|6fHP&77>R#AWMz! zQR+U6$Lx$;khlx`DZV+wD|exK0qPUb+j1&s^~|rZU?LzMe-cf6UEX=72JPxv0?(VO zScJ;sUj%+4gRUy(UrUJ7*=nIFP?%Sa4=eV#OIU&+E zac9dAij|0djZqAf5Aw-mO#|0kC)OO~T?X-PP~F2&7Hj#j6c>HR4)vcxv?8=2R8P*I zP?ULvtjfq<^gjrAY|qa|V!Di@%C(06QqK$)s1}Du{>U{(%o4LYbBeTDWiW`}To|mu z%$1BIy)9xZEmA#srU5pkljqgd)Gi7N3Qk$etXHS5Aw@zM`b8cdW#~o%%Sqi{3dLOn z_0!bst(l(#2_yChmJ?=IBj%jSm?&2hl9GJ*B#y`mk+LlN%U!3yQ-NgRa4@{|L#G?~ zJq&b-@wqOIyw9gwJvhERYb%6@xgm6dslVduYv8%kK zxjx-Me1WCr?2kNI+|Jf~JVER8GZxp(TRy~BuMQzyAjZbV*!G6{dbp~~tHEKmhd*KC ztx$EUbLMD6s3n{ZWKrXPNOxbvW4a7h>>!L;LOGgI&?b@t>32ZBfrgbMTQX;@2Hh*6 zVv?a0C9l+)5BZ($w@XgLh?0~K_vf|~-msMqC_IWm1?t?Q44|W{*v?5Ng|pIUAnEC z5x$5Q`Ugs4*$G%B=m6vQ+;$hEEYv^?57KW2$}nZzHmod0Ia25@t*;c)R0oPSL_K8& z`YL4R(Bp7;Hw2OAM=wmlxd}`lQpN;9pb|9J6m^+X#I^DMwN)er9%_A1_Orzy`OJT? z|M+j~DK~f_z6c;Xd88vv@iGrj7bsio5gpww^ztFp`Qg>CsFv@bgp>nY(Sd$Y3kwT} z2CgCy$EiSgnXjQUxp+A>$r3c|{?9Adfda;K!EWC}r3CF88b716n3 zDKPMsrO}4rZV}XNy;rNS?kFQbWy=?e7$_?K3smqOqN zkE5u%qmqq@ql>=1F+^71(bmew(aPL_+}YUP!Q950jgf9EP$&DlyG5e-19B4e1QhJJ;>HW9bn)=968$#EtS%b3GK;VJpKe@|2)I}D^!7T9X_%V##j|MEyV}(TCMf2s77piTf6vR5j?4Oc zetWelw7H(0I+l8SWvZg8x=}GAH&W}g$?m+RetncJv~9aRoGHy=Ggn^`@>JLY+XSgS z(_wYM{pNhtY0@}OGIFIaA-!&=mTv2qhlgilx~jO=W%qj?MI!C!>E+eoV18MdB6#M! z2Uw&9e7p4{DxYA|ag2A@Ic>(quHLQ+-R2sG@rVVIc~w)b7uD?yi{j~IGchF zRf3D*Pv7%@3)8FIsk_ovFOu2!#_bxvwL@3(>Jl^b+C?oSocBjU1-*2h0^Zy@BdPG)SnvUh8MlD(bv zJe_sFJ|3^Mp5+xdTkvx~hN~St(WA`Q5n&r$67sH$dRWNnB>q zysj_SlN|$-yK3p9f|0x>$ z=hz)SHI)|`iKAoJ@tpViXymwuu6QjikaW0<%)5Zdre1+-)=}xw&}JGY+jPBcqoSiz zI5VSdY-+lB(91M_w!cjMI;}?NcOr)+yZJCxp;nb#B)j{OnIx3rBrD0*V)MiGa&wAV;Z(P6_mn6=A0I3b*2c@(UBdVL`|ko*3v+W>9$*rig%GfH z<`fokptmu9_t@cC4j{4)y1ay1S+VcbZZaA!G=J*cnk?tj_6BQ@@MVf)D5z{s&HC?w%y8)h6e6WPr!B}Gj8ivU(b8dHUmYA3*;uY2wfWd~#sBMNH zpuaU$*?V?U)dao!0|;j5$a|sLcy{y1q2XZzZ~^~*`_s<-E;5_?>yzp2@co8e&%v|V zU9a(*UGLk=`OY72{HB_UsikA_{T^a*!mMX45Tj044OW9CfR)hn3Hc!*7XO~`Ws}SH zlyS26&3UJanvO9$2M6NO>C}qYHkv^0)C#I9P=_Oifpco8!ecO)HI;J4$hkI*z>zRTcBjhy7f= z%;*9K-IUu^I^NgEh}Af)rtfjOQ;5b`E&N{KSh;((alsdDw8`6Ba=7~&rJ-(N-N}?C z2Us&Ho+pzRAR>ivy6%!fXDwK~AvV_+$BA4v-#bGH(m^^=P}HjFs2tbLzrAd_9RYcX zQZlcoNL5xgltwN&N6_;^!a~cb@lT|5tDv{8?)ReVrN@$ygR3K8rH@uxsv$P+RD(y= zPW|D)N%(A6I9W0HOojJVqhS|jp-!D@M@PqYxXZz>mzKjBk{eT%`BdhAdtwgGKW-oA z)YR0}g8a^5HBk~I6o=T_!Ay<6Q(We9+c5cLB9%2Y^Q0MT-W{&=Z!%{qS~*}rO3C}~ zNhFa9xaa)%@#A{_{QNv}O0Xz4qru~JL+|z??g0kjXqs3VBCn1Yr&?H8OxbLLd)}ON z32j`?-!_6GfCvkNZhdtF`{IW#Vh5Q^Z9{fUt zIwme|+^fILxc3?A{iAIg@4B{eQje|D$Wg%^#FB~302?K(%?bG-Ci>;o!};-IJt4OS z-G<{aEnYBXd{#aOuKc_!@rFmBnn^P-F!T%#j&n7g%OdKT%e-gRoadd5qy+E zud%*uduW9eVj$X2?^D6U43V9WHcJ4T=#>8?)m| zg6YNp->#M3FU-@chu&n7Qz8dpBz!Nt$Q}*~K-#~8+U593Sh%pTu&2LYcEj$YU~0W? zu^itv`}NIf6PCv;p5DhakS)ZcUcR@+D$D(ytMn2P=+ha8@Hw~Dl#{IM7L`uEoobQ! z>m8RiFHhU&PN3M4SkE_jxXc*-8XQyuZ+Sf4bi=XaG+XO5k|7y|*p+Kwi4LWQ3+91B zBJSqqb^-3jBuU8Ym=l0Wl1IYlZ1yXX`U2$VACI2Wn5_?IZl|%09)iZLBI(8qIn+{qRCkXtFR-%&JZMyXev;=u5s=CXrlfZNh zY$|yrC8faW+v{C#OR(=FGq$IzEJ59foRY%63)do+cf{bTrT2i_;mUf$>0Hz(t}y6T zEM{%nFs4fM{RjyOUkKiUR^$j|PnYWJS(o`3aP!}3BFi`5`yx_O7}}VrDhn5wbZV6^ zPqw9d)gkJBV0y-^o35SVaQOOZ)2*k+D~)oK(4EFhYtK!s`ADjm=;$2p+Z%T<%@v-P z&XU8~sa|<`d9=M96DK7&Jv9oj`M6A!2R4F(L%D^eC5QQNMoE4z zB>9)&{aap$F#5^PZ2mV&@w~LOH+3$%s>QkuQ!axK!JNFOrlzjlhD}dh_K`YI=x>Z< zhky6FFjrAlMoi-G?*bm|ADSU|&C+rxSH*{xdx-7;B=iwPe04mwDQb9(Nz zJVbPrmky4bV+CBAum1bS;jG`rD}7_b!U@y>4GoP8un{KaOjRwkeubItU^&{RPJ+hKeC>A{ zVp<%3QOB8$WXW!a^EquOr?_wAPBnVg(a0ud^e1uAfoAb3jf~l?8>AZ@0GyJ9I6W^7 z(CS4Y8Xycp_E)oQND!^%a$JZ+y7@>JN!iFP2$)K?T#B-yVq}&Al8*zx6y%}49ChTG3S0)h8BV;D-4*ITGGChp8j%w2+cHwn6@M zM7^kU-j3AAXGIq=>M;h6~CUO6BZ3CK*ro10eh3;^r-JnIeflC)sZtdIbcV>eUhDnJ@e zNO;n-i}kJvKP)!pZ_MDh|GD#{m1&)L8yEe?S!(WilX8{#b^rTVp8@{HMR_W1(~o>=ESr3c4`8Tn z;u`PUq*^s?i*>}jfLPG(W`*)zWh^TmYQJ_vpjsorCcW7%-Dj-bJ6Eoi)bAw`AXj*p zcQ7fxRq(vW^Dq>%YWJhXw+$F)_p1%#Ux$h&;8P1?0cpvJe-EtZGK6;pEeQCP4_ueL z;tRrs6k8IE17{jn#mDp4@lF$1hJ#Q6ZvbJ8(o)GSV#|7E2?hQ4%}s|zvY=K!Wv`(} zM1zOQUJ9#mq1N=`im?7SSh1_xhK1~@uO9+v`&`CX(5}C$QHeQ_Er@vedmnYNpt2JD zS-x!sMFAPA|BdO-_Hl7z?$}c~R%k)#4-@cG)l$|fnU8y<;D$%G=S4|BNYexJe-tQ# zLDoqVkI_YROl5eex*xdRL;BmFOiLiveTj7m7ohO7TJEb;sj8&-`$pBk7%@%|L4KrM ztWtTsi20jWla-dy!u_4Rm7)G`|0QlBxT{B{jbIP{*z2vBxR*pT=Uw z#&=FyM5C-nArq++%+U>34C}es2Dgd$pvnSDgknrZa-KuY{M7T*)Tri7??T$aDN`8! z#&yGGPM0WNKz*@wZ&5W@56LjkVp;Ff#f2GQ7$A@$?9IH4D(n1@Pw7}4MB=;BJJ-vN zUE9Q)3$W9?FTc7>OFhOxEo=PpTtTfp>J4J@i?Fq0LPTlC2RXy()j*Uj7#P2}Lzlf} z)n>;Ypcqu2U@KTe_rFD4x`W-8_ve#u-36P%VYfYeykNBPZ~zPJy|D+|*Xx5gUSaCs zQsKK%pJ>M%;o){TdKNgrV8-}8xw=_$!Ko^8$7LrmE6E$iv|DhU6kX515loXOsc>Y<0{;^oJ=c;$KuA{B?3i08+1x-wHT}?e|P{W2e72x<)(7f>vG5@JwM&?)Cw|O zR++`qqJLBtxm2Ve_v~&U#cG}F!Vc;&VoRbdFSu2yb^TP$AL?9bJn2sCXYW@(zxy{S z`EEuviQ*xOa{cyhRaUBCN@2tIqSTDq;KiJBsZe*bu*+`EU=;7pl=&jns?L-wA;G)h zle*GlDca~g&LocWLF}`N?&0B{8#HoeCQb8`R$>Z>^bE3Hr@dwa6JH_~eW6m0Y1&W8+l2KkA&l%T(;%46Lr&BUQY>`vS-m5dzUFW~hFBaRayO=xcO zCy6nSqtPRtwXMLejZtQ;o2V@a-W>)hz@SQEi0Lempnv$V{Uf*Ed@)E9`n#BuOq_iy zFnME?MJlZa`W%QrLkbLr7t#Gi8%y56tSv>fACi5hqYI)^f!5>PEwmDO)0l%*ovx39!L`cVpwszwU0Jf+^azjfzEHT6QrcN z@>@mBo^Rod_!b;Fx%f+5iHc60U(GA)l@Fi1QGL;O53q)rYiDrzPIcSa z6ubE#@Ud)zoTrZr*NT3h^6Ej*!~{XG$M-}G1c0bf6)|i3Ri3R zIX2Rx$p7M$0uiAJ6g62c8^y8g)b=B4IcWBC1@a%Qa{{e%{T5amh_WawW1mk8cj}DJ z$uE8AV8B->@#=L69^$=&Xotm$C6dSF=putCPJ$T};}~BPWB8i&gZX>GhVGQYLcK%2 znc=MY&8YB(-YNp)thTDK6US&IAO^%2TY;Xa0docMYwog)1)tg#P%NQIyTmsEFYdw` z^lA)4TefQU^=-2u$%q^sj(C8&f056UC*_b!WuTNnW#uOow>hGvKk&yW z0|n{D1C6?Z&z%lFqFIP&Lew@U^F1c8>pVN^zF8U?YCn=y`Sqs2sXxKh2L-Vm-{GQL z>-#2jF)I?zME*6ODQtbAH?f;-ZRvNfgZc?sn>Afp=@^IU6cR4Z-R1gK3$Y=sFH+rO ziJu%aN>HObET5J#D=FfMl*zHg^hd1S7iC(u((l@9Cm8CMtgDRcVt#+nDd%#m=#56E ze|tB@K~{J(OA(hn()uD+hb6le{uB4_eJ4?Sn1B(W_)g3S=AgS?C#rRtqW zb=JmF(SnrOTwiWak$CpaaQzP$h9lvb9La$RBtyQQviL`g*k?h9cE+yfqts)v#z6&O*}y zH;gOzOhXI0NXoueyEDVpe?24S2LoF)zWIPO-?HHxV~k@Cy6NxJ%*qpBJ}mpSLq(Us zo`@y@6>cJPfpW`L^V4zlT5HKTw!!QtL>R58$8hx)eYaqPtZ#jNAC~^84>dYH37c>j|`lby7kIM8a?7+h2>D0S*1# zdA=K3UwW!ZBjXbU00LL%P>4*)VeY%~BZj}HFc%62?~Bg!!&>+VLoElqVi}E0{O?5= zfbUu6>|jE4P)mch+c>0_WE{uK$!~YEZ`0#cy=cex;!S%SrU< z@ICDNScm+t`zGd|KS$AN(;eSC9seV=eboLoSpQX>djfwF=dZO9@j{URJ6^R!_2XJO zkN88-fZ+whdb zR-C^&0yn>{QS;S$@wwz=4{`I1M@1hAaN<1u=$XW@tq%jx6L1I{02T#K19~HWV`HP5 zw09TFHnuAGE=z+I>X|c>Ck2vPW*bZK|J$Rd&)K|uvT48tm)6iwWIeROKajmFE`c`i zfAZmdpZc=_FQme0LfmRtH##~cr4SH$fGlyU_5?V4W8~%jQ(=HZp8ad!*RYkTUt2;8 zK{M(&(P#c-cd#%|=C5tr#?jWu5SOd5G4(Y+4@WQ7xdt(PopY$YYI~L00}m;W_?Qf` z>osxBTF2b6Vm+qZZHPb6yYhbie9=#d3?c8oYw5pK_qa}wk$rXb2HP`IU!A8X=W+1I zgX=SKP>1J-%q+6+?XJjX=|1$3_DJ|dmEyLNkjU$lXWe*i;tM>pM`jF+3euC< z!D2qjWwMBlAduDanAtc#A|S0*ceZUVCDIS;Usj%P{~GZZFJL@B%e;KzDu%*Jfeq~E zw`fJIVg^Ryx=B)STBH9qdJ)4<>p{PKm-cYu9IG2Zn}gwiH&N7vjMIgHDJHDwe@E%# zTn=ea+)vH(m1Ip&m;+Qg1;*!d+HUrQr&K8@;T_iAw@N7P$Cw^XF^Iw3+H+v%6J<3} z!JdGU7fJevoEYDpr&|oT$NB`N7@fMT>yxo19MDr|B6j)p37Fifw_~QMc^@Wn^>^R@ zm9kPN4c81n55C>(^HRFk>xxd<-@q%QtT+EI3X0b~Z~~$l{trVZ^Ut^FmMp4@%buI@ z;g_@)vWnHszf-vE=xi=J+|EFLl}z=Y%(!*Yi#7J4akiuM5QlVg$9 zKR?Li^8`8HfQ!^zbak|fek=k>w6`)xD>cZZ;0omf#6f~dA+h>J2`o+UQI+taw!d}O zzV0EUMZRTv+Sv_J z*6*jkC@rc!@#~sMto&-WVo{LNahOF#@=%_f9npygk>lfIVe-sDoi^Net)3?{%WFKe zs{PtNs$2Y8U6lT=(&k!M#ylqW+mu1*%B^2;@WDpe#3gZ_pFP0C75(_C+@nELV+#9q z`_4xTg--to@~{>s8bp_kN*UYysd~*d#jyo106*87?(W~frzB#9 z-L=5`6w&E_e)1l>Z>m;Q)o zBJ<}grz>|L!FoDXgSG{HZPU9;<(Qe}@X5+>PlVTVGG!c}4>1(NA|vy-s3izN zE8v5QSZRx;{)oW$Ox05V;gB3xq_Oi=d4sz%2_oy;@QEm_LjUN4Lz8_ZM>TD}~qw7}xQeRTPCg;;+oAM_ni`-#nh7U(T zmR!m^=R|)V9V50aCa+h{npR@9icJ)~^)Y*?AJS*25}(-zsxN!V#{FW6Y!|4vM-0^0 zYFN=WnJF1jfrSaU`pwLnHM6<%Rp*#J{k^^o%ZnRl+Nb_UKExACxcsJj;r&5pm zOn%ul+ERfQG(H65HDhFZ5-ywEp1olZ2J0pG03PTHKm1$>i1tI48k^EQr%`Pmw|)eR zb(Ai=e#nYH_uKm`<#AE}W$`b4zk+WmDf@3!FE09psn)r>@8ybvR8w9;(ZSK@K7zyF zl3!V!(YF1~eH^R=$BEP%{DXbK^_Wd}lCXH;VaBy5AsOLnI<(kn!)qUgCGLWK1CWvH z=%zEHTkiI<8FA~AL#U{9A-?;#iGk!N4NsG0fdRaAeHBYa3lpSfDk}-QKv(CXuB$BL z+ef--?;e+8G^M6{&NtD+R=ef*iC2UB-hl^-0uMx@c#@TwDe5myOnPFSUDT=hq7~d+ z5D|p-klA&~G!k{Gj3rrVn%HL$Rj{CZG)`%FHvHYIWyqk^s;{Jve@I`d%UX(pXEbP8^rY6 z+s$CmX1PrHMLL|fR<1>lU06s1XZa13JBy@Vezkr8Suo<{PZh*3p_FuLmW#1Xp?*8L zzoB)1Qo;b;r6;0LXg_-O2PKFmw>Iz8CsCs|pA;O0*6~jWX3x9z%WM2-hZ=?%h3`?^ zAgz!iPi%j@_|VLm?i8ccj?)jgnM*B+gulq5(meOm0eSvnEys$I`Vam_Cr^>fwX`%d}l zE7Jvrek9Z_QGam+#CTP>2RI)%-~N5*!2LIzqz`o6!LzG;ib2q1PH7LDt6aV|HD818 zyHsHq&Y@Ic-{hW$m4h4WTzNe|`Gw_i`svzW7PikoZ3U|t3e$E9IeLSrbE}UXtO^ov zczt8{{GT~!(Ul`3h{>>L8x!i@w*~+-L3_H+zrEdpgCPsN)4@4t ztS!V??0xLx=%7zmP7WrmSz&2%4JRJ~rr_#t>62SS1S{p<^&@`KH?yW_+bV1fog_XG zyj8faT;5i(3&oJ&;&L8A+=95HAe8Zu(`!g_v+*oD;+4c7Sv!6iBNm(y!+T2~p^*cK zLT+haBAplTjveizdv9=UUmAZRkVI>0# zb3t35v-TZ&X_AYSzSJ`aH&(F;M7WcCtFbP<7y#iF3GB#chq73^DTx3fZyiGzs##D9D zt?vNsHZc#pcL}qsrFLTL`cuHUGu{f(uM*L3J-`-&RRi4|E8Lh(5}SZ5u+Q};88pJI z+nhOFktDt{v~B^-D@#q!v+{a=kpD0QW;0)8x7?X&rws9Vci;W=%X)2!v~HxmJv}Lg zkBn(BtDj^Pjrq+L0vde>YFKnsR60UuM}B}Varzr#RE9q-Mj^=H7bw<{1=?K(V-ZDW z!s^Gx=4xf4@eE8%YCL>uNT%HpLb#yW+u$|)S~fNYn_t=0D*$XboIi42{L9x-CeH3X zcQ`vVmh!ggljnY-Gm{no)3yNs5`l`J*R{5Ipu_bBu!CxEG;#r9G$K$@{YO=$hpY5w z8<~J)&)qJ6+~q0cEU*42_;oi7kkEj8Xho6fu$-E;noOiq01Vkyfa9kvtKx&h!}#52 zXg8&MpVzP~TpMCho{g*mEHAI>-c8nkTnRqgXhKaFSjvq4{A7dlWT%#)Zc6o_%F0^n z7ZVZA{%X(Ww2n-IjqXU;lVqC_vhgGF`w?z;IOq|95%sU9sIP2RlsUrjjW3080i>e5 z8H&o-n&RluP#{s-&Q9uZzYhgmc`iPt_SbQCu&c3E_`T`37{_1=`xSBaNq~ppv$&DX zXUKR~Yz{W#yBJAtdIKoY^Gzq~$c6(w1vSHKYVmB_caUOpc@qfvgM_25pT>Lg=a+cm zg~jPmW${u_rE@zi3BlG5qCM++fzx$Wy5~B|?FmD-S<~~b2;F$4DES5`1*`Wpz!p1Z zb<(^lu+NK=OejAksL#>~XGSBhMZYv0;xE$%_$BAAO{oaI(7c%xVKd?2RMPCB;MX15 z5wm3E#w_$0ep!{L2$CpmowMvE6j1jTiJZM*_Z`u&H^+>rDl{N4^i)u^(tP}H|O!+a)}7mPt2rP+go3LYE6XcQQEFW*>Eng_B7 zYDM+?k{uul!D{DQM0Awc2w2*SX6Z?s)xk&%cOzLcLW-8gIAv4Ft__di6X=i2B3X+v ziG6;9T&+zxk+r@MQ~OW#>YfnRQvah=RYwz8KdA7KjLPRGU3ev(1CKaRm2t;l>SJAr zY}C_D0|pPz5%1C1GtV8yij{jlRMok%N*UPGitc63jaT8%!a8ZpUOzOV2T5MdRyv91 zWl`rnVW27im1$&}-PkI~$!_%#q!jd6w*Kl<>;#eg(^Z+)@-#uu3^w&Vl?q=bN4E~e z(}9QsDQ%tE^gj)As1{@jR6QR)SAGjnalbj&yHf#$jfUrgo)v(CVmhIf?A?R`V3Ko3 zmP})m8v2e>H3rb6wI|)~310!ZYJsqGF_3QR;i%li{n(wzd%bEh7$P8TetU`S?o#&> zTOJ^L3sRP^k;o9RtZcEeZSK+iLr)ucNs+n@HjTLjQyfp29yP855Je9ex80y&p8sn& z(nv@r$qMM&C!QVb>ydt;M259}MK-*U3}Ef`vAoaqNpx6Kdw5KKo(50`8%G;-6uXC~ zOk+h|x5;It?IZbIuRSk29TR5f93c`|VBpjLcO`rNgXLBVfdsi3pe@JK@YOpER)^OCy$Nd>#2VL#n}}(ub_u>DTxf1u|{RLvsWJ zt|Lg~*MLg&A!TP-vNCm9pto@;wQBV1KLeZJvQv+f3TA7h^7MAaG^65mqT6|)&%+(pg#e7Fnp7Rys~gwQ+I!L`lyTJ z`Uw(uT#}sD!5xXIt!JPjohLz8iV@CFGw<~Dr=XY=r^q(b$ewHgYR#j*t>+^Ma8JrY zTgpWO(e`6ShD1*JYJzkKP%|U=z$g%0!D9?3dkN8LDvHv+o;W#o2l8i$&clG{IP)#< zk<#6z(uC`k$HJ|z*nF1vN~6zx-!RNrXz^M~^7Xj{3klG5(RkO-nGdlY=m@cFU6W`+ zGZ)gYKT~ODJwYkH6W9I0ZO4L2UCe6VYxTJ4fB86alP{8k_6*m2x81VRba*>9jNyB% z^3aCr@(y0oQuv8stZ>2ui25g1FBb%zAlD#28BU^|KFR%R?9D%HgouWU919*BTom((Y!j&Pt8qL!vR%kIamwk88_U z9_|`yC!$gCQkB4SN$UwLZ{C95DIYYR1Jq1U3>IZ*nzE1pSVfAjZRy^^Fa|89B=Upy zt1HzLEhwo*(M4hmLa`cUUbWlSK8W0~RLwii7|x+aiA^0qQEz{e^ls5&wC`So(cA;d z5w2)A{^4RTYQvf8prq!Evt9}ud->VCb*ECFe)U?(XUs_YxXt4-6Ca{S1Ipz2GO5k` z@HzUmau};*701RHD9D)L*Q%5OfUuuf3=Pq*Qd;xzLzu;szgWLQ10Ef)>8D6tMn$Lg zZ?Y}w9ll=qiw5%R=3Nvz*Ll6Y!)B~O&y(IEvDqav5C<+}Z8Oa++okL5x}9jiQ?DC} zSEqlbE2pPrG`rU~ZvV=oEKUHMWP@7^ur2Y7MHt2`ThU{Gev!CH?J@;Uxa?0vvqQ7t zxHDl&YadJt0HyBwa+jQ?;KvG3z5wr&E;zcA*R+{GqIk(-k`Xf@Fh2JqYqCjZ(QM|^ zcvXMGRW2%^3{(sI%8<}~oD3%9aW^s7TruBEB>qSJG@-^V>Uix~LeUDmi(j1{I@Sli zq@rD{RUasL=u;6y*S72Ql5b8lhy*SNh#B$S<(Rs{WP6`=j+_8QD zyvUMTFzW&HsyeXx$06K7I~MxZ_=?028`h z-9(&HyYC_0g@*bW!VXjM@!TF^Cw2!GG9&x|PMf)cUr{eNhoza|z*1*$!i`j|4b~WF zr}BXLqt|8j{xQCQHi~hk0++fdZ-cbOCU3+V4Fh6&j)L<|PGHm4Bj5l)F>rzXVn#pS znpDcn%;a6qh;jrTFghV2J>Z!3(e=J|y1p218ar(~1$oK~)xvt_6Cs_;FF;=2hQ=9( z5aEs2?0vguRS^CJEq|cEh3-F*QsF4K^Pem=?AibZF;#Q(Tq$Yk<4~KAlz*H~=e?VN zU$DY)U2)_AA(}4wPBW7b81dlM?UADS@kYsb#a%0y`Deeo+$rB3lb3gAm1mD-DF*my z{=>YsEd$zBG$Qorzx>#N3-o`4NRb(*n-?y4(Y+NI7KG==x7W+mulR9TW!0Netx5q; ztigIUGH!JF=i0caN660YXwNu_^EfcWnSaVChmCQicMrTh!Ye>V52DRBIj5j2d)Ql~ zlb-w>z9Ll9MR~oU##%q*t>eFqx4EBVcd=CxuJ76a8oXdaHHa-h$38j3Uc>Ns7JW5; z>h=8Iq9FIRH-IPP=+pTzwiJn31^ght1u*ze zU4Ag>ySLj9$51KV{1AIwIz4|a@u^X~oy9)Bzh|L+lg~>|&_4{|qWp`#sUj0+U@GK2 znJ}z0==kc0h0=`l6P_$3RqLd7BBb_CCzQH7ORZM3f+B?78F3F);|E$Wt0BX10R&z} zrL^kL%|Hl$-ttM}A&)N=95b-#+T=YwquP69-8=A;U-Yx}J%6A5?eqI8WUB#Hx^iHs z9RsfTq#gZb(Q?g_jMUaDYwxtd~lOz1h(f>ONg6#1)S=YO|5=aSan+Jok>!R4XWjV89?E#_ zdq{lxktm=nKkiq#c~pp?=agLRQo6j7YuFmmp%X$u#dE~)khP`eJnMbEcpk96l(PDc zd48W|fLr8d3C9>-^vEN$$;c-}zbzWtHtzytICYLe*qVv$pulMk!sEL1mG0JsWsX=3zz4>=f)BRX-Z-&m#nF29{ZdzYU3yUJyHYoOkPy``lpw0n&iJeP zZPWGDVRtx^$MR{E0D~@lDA7y*JmVAhOw-iAr^Kw9Q)(tcS9>DBbL>&O9L5*vRd5Hh zQySh-d7Nu);I3o&Q+eLhINxr^z+y!4dhU=n&f&*ZKnl32g$p^3^lZRcJ3}3=rK&XS zvj!huJ77M3YDBckpxi5pCw(kSu^aBvs9z{?IO9yeCsVmK6ScR@HvfqG(xWG#N?zy2 zBd=`ij&tw*fpz>V(2xLPgC{0bmHi)uG$C5M5CN+pRf z?gO*A$>8@wz+BK~bsbzHhwUoGRq7kcj1v%X^4@sJs>%Re1)=r@)fMyollqLPUrQF^ zg++Sr+ykwB-WDpE8);4hBWJ*E9NXe?%A?kP2TQvziCO)MhyLj@>%1)=yRwg46{b8| zx_@U+^DL}(BYcmq*tq3cjUOcZcH{x;5_FHG#|Xr44^4`3O|&vS8XPHl$-GUbmXYeg zpHyd-XRXw7dn+S|8;{@^Kzib3r`V<|Uwc(#DDp%lYo`(BP!)WqnWM&@yd07D%ELiK)vav-))=X7v1+D!{CSkcgSmw=72QdzkJ6}P`EkD zmtNHQAZ&wd@HY~KnZ)>k4>GkDGO@JU=K`kj!1|OJb%;+20R7ogKOb;^vwrv){G9}} zZ`JbrvzYe;+_8<2Ah3=e`Wd=`Smgw&*do+OmoI75DtBsb zhUAr zpJ!(e#~1yD1R;M7x;_;HsAP=9CN1ArWkq>li8R)EyyP|ltRMjS+Qz7BDRzw1_JrB% zxwBkYSgj7k3N%g60jt4qAS-gU6g@+(X0Y^>+=zWnblkJg3$a{4BR(=T#@$&$lq4&H zx!<>&M7qUb<=Y#Pn6^jy7Gk^>I3qvHqQJ>+0%uw$ko9qkM*cUMVFJsL%DiIhZ4H=c zP?M7wT-PDE%0hyCV2UPa{vXxV;B~nK>n&mKqmsr$_9ZOpJJW!2wwdskFE87rFu8f?dLD3QH~a$O2W0g?g`J^H z+$0QEQORKtG<$rOO5!$SO;0o-;}NC*zZ|Ko_gEp-&xo2AWbRNuiKVomKtu1Bm!_E( z$XZ#@D`W^2@YI=J47mJZV*H|R@u6yJIEVt~92FfmzAyIhd-C(mq(GxNqj$W@tz4z8 zC;8(Lhy&O^rU$MSE}eHv1#LPP#(tUNEv6SLz}-#L`hcvi3UJHbw@X3eRL3-J{Twx% z@p_OFpw|%R11ATd3!29&GeJB2H~h@H`x1yY@y@<+P)&U|z_nPitvLzE1%NMenQBnI-b-44 z;InNM09*`cy~VE+LwI(^2M#Tt@HI8+3xeeBdkDQ)PP_^7k?xqvT=?*7t1o=j@H9G* z4~??~#nL|ApN|iEB^^+>3SPyR6`5S}7m(jCpOkG1I#*TGj9Y7SH9;Z(yA$yP;HIuJ zm!IPui6v}8&kIYoI(rhe>+F`V?g`feN^|K??tI$!t`jT|?}DVdzBANGC|TV@(DAwa zNyFK<-vpQOqm~O3;}iyCbVP?ooO*AOZeSz5J`{A`ZT=J})!yHKhz4N+@I2_vo%syM zqSAOwyc>2qkz#TyS$YiyNE}K;3ybPeWlXts`BXc55EXHHet85BI~A0>YU#JGGg73Z zXQLj;r3e}mxSX{0Dr`nL!L|{soMK;Jq;Evu4eF!`cq~2+;;RM}K8}(+{Fh1nOppl_ z;|I-e6y4m_I=wzyz0C$TK!cVZ`VzgCU-OD%0Zbxu<9mDQQ(PwxWBF}pcKWg>g&^cA z;C24vZG^4>#|6R36Z^gtLQZIfD;2~?Mm*j>UqcsfQ?b<9qC4qEkJX-!rz-g#B*J+V zByAx6SZiHVzu}Z@F}WQwg3dh~DFp}gKa0fsA9f6Nv2T{Y2W5kBwsk%ne|vZIgKPTj zSA8~o2sr`B4x0a$#_b6?080Xw3F0Y%%QwXm-FT@%+s|nB{KCU2psPiz< zf3q>@pr~8%^OW)X#2Hfv`;9P@bI7JXAE_+0CPaoT2!GZf?s?P|P8d)t6F$>bG0hd~9JqeMk^im? z-Sm&jyLLrbSsYY3{QlzD5l<(jY^!*Akrvy;0ly14Y=V$yLSrUnaJwul`?|uXjJs;O zp0ack>J~c-+?bB|TJfriJ6ObuqnAY@YiVu^!CdA$3`tZ@(I6#EtyT|$U!r%^M$KUI z4IHA?ilM*B@^Vb+JV5JeXaU|j4AgQ0`tQuodiT@U=j^^8UgYa*K`Ti;LkL&+i$I&E z=-DqdB6#tBbt5Ng4%liUz;f>SVCSE9%{M>^AJ1~=a;u;;D~ud~D`;v0&jtw{7;g@) z%D2NE?&z*|9VXVR1nVWjjYlw$42+%Q)bBZG90U{Qej1h3zUg#(A-AU8~yce4V}txCL%9L8~VwFTn{dfGvuy=fMfX+IydTz>(F}YG z%ZZG9xT=gVT?ts=h0XSg&mj<)aWFHV+3z1Gx7Oer24-p;oWLnr%QgAcLhY=Am?#nC zxI&SFtx7vA5Ps8OW+HaOPISb#MXaTVkAbUW;_&ozL{(KU_(BOxlPSFIzLa53ymDmr z{skdu3wX`fvX=l1hfWECnQabd!P1DDRvA+JDy^$Rb7wi6cGtA(bG^-OVbyy6>syk; z+mIGFPLD>1JM>3%3Zyq6nY{%L#i^6kq+bPzHWN?(6Dt7QCLqq_K(s_~NZ!|*ddEj2 z@4*?dSF<8t=l?04LDmlr5Wp*@Go*ChP>Wj8tg83cD}aK;A*90H>C7@rQ1jwQNxZ-Z zdnduU3c%M2Rz z_5N8v5twW8XWtKS>=igJw{K6}8+bQz(kpm}X*E#2yqh@-1vHt6Gl6dseYWRy3kzT& zgYf~bvh25Ti)E0u%>L{f)s$I)p$7~jT|yRrZda;sZU~6G`_2cZN?5WsLokS-hqv+5 z6YIwqn*ih8og_mokJGmQq6@e*%qqlT)o3L-$&nx#1ADrc(=JjIRqy@9aRVBrF^TG1 zOC6WLi*RisK^E@Aa;k5@w@Vy%64_MTUV`JxYEUTDV_u^|QX~8G{bzUHFu>iwFv}uG z|KnTK9RW%Std=zJ|4SE*B47mV)*voNIT|j#yoyQ;IM00Vl8X~@@E>7zU3O0n5qi@e zkR2>)Rau#<+tb5(%~QzB(z#_A1a&JYMYekz@gFE9EkDdhbbn4EQdNeA<69M~;N}`?4-E_ZU9JXOz5;_g(kGcwH4R z-pc{IJoU5cCF7NKoki7nzlYB+js5< zR5L9?qq>`|Y~UjVxMg{M#uy*S&r;ork0cN7+FH2aLt3ilfuv(#lBFu>o6-qsDRygr zC4H|jA;0vE^%<(*1PO&^zg7M06981i!4=y4ErT+5hiW({Rke!FQ0M-OW%UgH>kvhO z2uGrpbH;5fAdC~2Ib+dYrPk-DjiBQrTw#>Aw37@(UWJxb0s`SKuXo&{}{TPjba8aHtFp41gnnFIDl9gzE}{hy>tU zZpz@TEQSN=IM6S0f!O^993E&7VMPvP&OaIcs;Z@kvkf~#1)-(-#Jou*d0#01tyKiw z6cn6nnojE@Z~L3dK?ROBd7on_`4;^7?yLnerLKBHzj9$A?UokyiFG-+KI!iI%MZq! z$j|Rqv(fwA$H7mFK`;riP1q#bRkif2%W2w4eGs1~hreR$hG>x8h@+SW-kdc8FTpY0 zob04J_dYn8nSrA}wpg&E7=eu#)aMyRA8#~+2t63;s{-35bINKz29Xib&E8Aj0*(L8 zV~;KVVb!wh;YNa#pwg9Nwpv(O074;eGvqbNR0t5v^8tL+f*Ls7T>%bcVPcm`e^M!g zY14uV2M%-u9`*jS5IqKfDa+i5gvU4;0#OfTwltjevWJ4uZl`O$EttGh|AADB>Q?nn zDqX*QHUE0~K7a~AT9Qzq(1J+H zsEA4tQX$D6F=S98>)2Y5B@&Xg#S&x9Rv9BXD8?>>!4#4;Bu2*Y{_fHHo^$>G*Z*AC z`|r9+JM`M)GJo(F~=&lWoj&=`sbs)+K0?xH}IuAnDG2;{Iz=>XilA_~GnNrazu^WfO z)x%nC>Lvqx3J^C1*H%Xk`T^&qL1`BcRmQh2dWPBF`~#ExAp0#Wy6A%l;UwC}6Wn^&cN1{J;iZb-^tV&5LMz@Ad~^K@+| zg68Y2?-FeX1JqDy#oX}^AGSH)`MX0>?*0L|kzk{|^{T0DE4t*|ZBH~lXo$RZ;n6wh zpB?`Ai@0p1-wJEqPq%j2NkxC*g`=*0akBU-o+rIG&0oL+5=soV-Jsv6#-ygnU{5Vy zF#uQ0@;pJZJD-s?f!*IGsODh!84z(9^1c|9PJm)=Z+*3k5_fZ@ZTK{qH$UZajUq8p z*iI>HZ!$b0PmyuQaHrc!+Xmw$>P6kjMX$hagAzx4myv^^jxu!9Pn&XvaolZ?^MJ9l z%4el5p178|`3?zBy|V73(un9f^{u9P1X10L<4WujorQXnk{A>_RvFE^l2gTmrV?zD z9w*1F+YN?+6OO-y6rnP`y=S72<0i^DFLHx%uug`#vXd1lfeLh8&Mk)$@cGxO&x3y7 zx9$B>{k2rrV@^S(?AQ678^>5zvXS|5XMZ6xJUrF?;F{ok_8Glb$`n^-^T&~eg=Z0r zR3h?OV=Vi-WaDR&xh`C9Xo_GsFs-sf7n+1X--x*^@{NYU2^=8}1VtQoWgIQn1trTZ z7z_MVU)fn>8CCpktPzm_Qg|ek81vQXRP2o7oNi$9wl}=Td(u25q@=3E;EbQ3oFZkO zHKJT>^16D+yq9v-#X znHQ*ndQnBQqy4rlR>;Rg-dJ@q6zS0U>VVF-4_JlF&+4?LL}Psk`k!k(faXE9&d!d$ zW_cHS-}WYS;hE|fL6^&DLaW_z9Q%G*a-_)#&)incXMs(}c?7oHI{y++-D7j=X0X~y zIB~$r{;@xdjq-J0dPo|E$vzufAGs@Bd+B~2PRi=31$3U@T4u>J)Kl|}2R^{twSxS( zN{1i8zf@g&w?J@<$%-(PFVMky-et+<8i0dSLpb~t?Ac@d_~axhID>s5rG?<_yi;rf z*K$VdSg%Ub&Ug4^_h$U;PYz)svf!ouk%cW{fp{*(_zELR(J4JyC@hKeo0*l%foDEuqC8;K^v{YH4Zpag9v9 zuzsIs6~Jc}5D7!tRC+vlB}sPVfP{B%6z@%z*&H^qy>#@-n=INl zPYIRAzKT}{wS!xWe}{2P#ab*+XV3SK(gCBc zZgmED?}uOdo#M0h<`T#<67knk_?MRxAlGL{;*yj_*tER}vQMB~L^V;FtQAGYMoX5q zM;(GVN6ok8w;?C6dE${vnicNdlKRsdyq1^PY&E0ygW80{&V?40tLo$AwS>dfMF^Ar z%0P^A;gTw3vFtZo8xl$0$+CymdVRI=olZn_0a{&iw1HU$qVahyu^2P+qq(EN$VJ${od!Z<)-pgmmR zZk$etWy9?Ed#Rx!#h33T)HR;T~u2$Bbp z(Rg|TTTQx|1z*WeiW|;1N)kI)6#Vb_Wra7o1b^N3#I0K*40S2Ch6@%kX&jVgxmD)4 zUx0Kxjnj6|)_1`$$6V)oXa8*0w|gS~63-3Z*w0fhh8rZhyy_g<|2Pw_F0C41INbro z=7e}`qo_~!Qpbt$L>EP=R5-TT(i*El=!S)VgtBnur*1tsc8C|5j4{`d91zCt5Ku5; zFg3{2j(MAp89`_l8NRp7Z zvdaRbk?i`)OCFYXH2b;4^t!L;y%(Si~>PGA~!|o56jNhd#G3e9pH80V#p$sDYTA;*e!QjL03i z=6RR>H8nLs7r#pbp&s>3OtJtzlxFw-^a;}RQ2-BxZB>B06CWR6*K6nWn0@9+!JkDn zrW)Lue&<~ZXa5>&yVMvxgH$`P$P)dl#+6?%yTAE zd}g>+9f*1WGb+)Qw-UDi!V|(!SuUnfvPgqLvc`bAK|ned-7MDtRNjP+yqAQZ-#MvT zEo*is9DC&y1huiqoOncK&Mei!^CHEI3JWhOo{tsn{TY+OrLu4y)GvL{)j5LQ$mTvj ztrK0lt_nFy?b*`=*r&ZhyD2;qgk(Uf8I^e!0b-0-!{PJnFVvKtoSC={fE`i3#**Bd zMZz&-Tnia5W?I(Rd4Gs=+mMADF1hSG# z;iQu(7xT0gnd=AGF)|Hu&EjM&Ztd*9{TiBwv~ls!W1FPrFooXONK_UQ89+*G*E3LU zBU2VxO#E0!ew>Ogx#;$(mW_`m5R>(5vM(8sh~jL1S`ic$=Ha@o3Q=iFExwqm0Pu+; z?b-3g4z%{qv(U+gAsk!(Q(j%z!^(tOsSG!H0!TVd0LqbJdKCr{CU2D@y+V6C)zmFf zQd07Qj?NB^xC1u#;qN~H)OSh6R($URsv97_+5l|ESuz6}H?T7Iw_@Y9tjOhaed~mr z{^SvIWWtc~v>b4BF=4#JSD-q)!BmWpy$~QAXO?fMYyt-E3+SB612d`t>ziw%r4+ko zD>y`Kfy5PapLY;Ov7N&pm}>$o+Ka=^gL`0%gg#82pBa7Z%mKqA80v{28gqR$-f<=T zBPe;ak_`ic7&G1@8R;X??Z&P{xoA5-;4*o3%|bD`;G8x?6n|BDnK z?|!Ym&Or3J*ex(zM;;FK^zK|`-1ozq+&S}GF>b872ufBV` zM;?_gr25PjBo`%bmxIyy=6IKWWZ^Zsa#3+{vAHim+lD}q)(qU&bih2yD=6R_5)KCv zTvm8NZ$~IGB47#(OmFH3);!H=5rk&<1nPFEUV(1EG*K18kHBscLW+I-69f_Pdn5DC zLFd6UK(4I=%p`}(+=`o1oG?I~1K=D)du&4wbZ&WIoxZG-X$=Q}u#5M)AXUKPA-HHj zz$%TF?2_J002+?^a|lORnVL-t{7`_%%un?&Y#qX{)oh5xe=(!}H4kH_U zGp57L%*=c*gPaav*HiQrtdVuo=trgggnM7i%-npaE)y*sX%8=;} rP~9L-fgk^MDDzjv|IYv@Yl-V>g$G^C;0+20F&d}Np3FIM`PRPx|3)cY literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/html/_images/week34_74_1.png b/doc/LectureNotes/_build/html/_images/week34_74_1.png new file mode 100644 index 0000000000000000000000000000000000000000..2e80ecf7790938ac7b84cbb5b09dd1d761e2ece3 GIT binary patch literal 38196 zcmbrm1yogG`#pHz(k%^_ZjhAj?oR0tP?}4JNOy@S(k%@VqNLI(DIlmc2olmr!@PXo ztXaRA|EyWFW?Zw0<5Id44A-uv0ljnmOm#>Jw4ZK6E zfhh$35%pFy_SScQ>g{LaX%A`Icq3fgydySYv7;r_j7>1!S98;J1PE5 z<{i5wIyil`Y#$o|ryCM+6E}yc*>stTbB$7gRE_sVOHc>bOqn2gdgTk9pH6+BVt=}` zQ!*3`7RkuSz75-DqQ4Rws@^;s$h@*+xq9@~M5=ZtD`99Pb4YR|_bPZPKW!}_4_%f3 zMV25kd;;Y;1YUrjFVQG?|9KG~@&D-sBhwrV1C2W!+sDTzHZc)pz@v-izXRHaZU%V! z_|OXql2DEg3Q-zj<+Qeny(=j}eNO(LYiopNWl?J6iuv?98?Yl77#Wc;?qqd!b!BOA z|8qySg&#f;IXXJ7N=Xq`5{ry*k&uumBFE5Sgwxse8+WU-4QAZG#^{s%@0U#vV1 zJ04=qU;b|9uIS{%A(GtleE+lXw)4Neg@pEY3GyT*jcg&@+V-R20_RVzzoJFGHae|& zt;cdi=f1pXqgB!_&n-&*z{V|ze_`%XL7~X}3?{(z(szR-kUTYO{qP3`v#E=7Lo=TF3M%I1tB(`5} z(;i#O1_p`=x&B0fG9;f1;n7MZF5TUpHQ)UEg#={_xpp_$f336WL4!J@ap84-`<5<0 z8!_EGZa$4m!wN02ustDYyoD^iJDf^)zS>GiJ}8e)!ls*?U_;|o4Ea;97+rIx=&3&2 z+yUFj%L36gTBaG7ueRqIe9%8$fk--U$MB(u4Y9lyCIZ^5s( zbAk&Ovc9=IT^-NO&1Le8mF4|Vx4G)^t`hg`+7S^nyOVFYDieThY376aDczzVDu!Ss zfTm;qmc;32Ai88zfEB9cT{W@7xT z5^HC+cA!GP?BH^z*7kT~)PJ{rpx2q^#ffgv$)vLXM&>ib<%y%j@d*FiM`yC{?=5!5 zU&@t7g3T{+ki^-36L3oBu@-=v@D(X?bQ_;Gm*T`S^5~GbPkQMe^X(O_Rs1;DSaQ$O zcTUQatPI|jx_?J`$33NJM8oz4Y`0e{lrG{w-4ENIK3y%C4RQ|VfR+)4@Vx6~1#7fQ z=z6YGr z=%D9O=qFhiU9TaM&Z)e7@N@XJ$X-^T^BM@4)!B!TJn261_hv6iei?0rI`(na@*x@~ ze&1UzGc-xxQNh{eKa<6bU-qf}=EKDInr$0;7-MCPskN?u%aA~H3XJDn3F@QrrH-Uv z!>DPn1^*l8c~Cj^?PH!q&yB zB6zbS1oH$|vwo}5Ekmnb25vHv=E7A=BqdXxmZX_sUT9G8iybP?EK+Rfymvm4rYb^j zKv`DL!~o@iM!fpyoQK@tgxLo>N;wI0tr0Q)1A`kd{k-C$ScY7rIpsT@v>hQWvsaU< z<`)tdU8>VQCt+)$F!=j{?>tUQA2(dD$0VpKocewwESw)^E7zSpo1J`G@+i;Ws=TCR zg?CU-g39dH8Hx1TmrU$Ab;j9x3_`ah>;XMCo@}$@_BM>JcX{5&J|vfkR7LEKk5bi2 z-LW0)Nw>E{)7;_aO&auv={@QFU9R4aVLxSN;2Zu!;xC~`aCo=)oLPWMXU3`P#A=t$ zqnIIHZLFxelp3nL`0{W-zzXUtpIQysYbJ)4S5d#>kEXBu=aQI{Sb-hZMhvN%O0}EO zW(sJg{+3tos6P9$TK;^IC-jiS+OL(Mbhi76*J`qayDpBkLb`pRIr*MhPMok*Rb)?} zyteFiWx;)dTaaUsl6%Q4dRWxv#OV=KwsU2Mn>2G&Y^1VGfD$QZ_G`ZH$DT=s)caYm zfjaw+=Ni8zj!B`%!AU$!Oa(ZFSXaAGn}gr_1@0Sv2^nX?$(~ZoPQya9`j4o@cY^6` zKK*8JXe~&pYi{nOG|A1+o$1t_{*ihxmr!vCmr;hBLoN-6^V8|wmumTE`7nK3Igj>I zYQ;T8+v;iyF?nxgkAglWuIq)|)vLc3`UUtDOP(CfO_cuIc4Ek|l2&={;D`i!p z^HZD7u-eS*YN2~bI%y-;_;0QbVLD&O7w!~eZV<_F$o*+1reWLHLBx8g<*i3sx`c35 z2z6~lOFf)tUlhMsOZ_bsDqr3H^mBW%WyE2#02d`~o=azP?e;(p-)4&m^?R2-=lnOe zXIo0ww`1yko0#@TZkfwicmJ;08HM6#VmIf%ISLvJiDTZ0=q2Q`ZXZqyd%cj-VwORU z(cyx@B3V;CMR`9q4QzyhOMXaY91?!dEAbFt+VjS{ea)?vlD3f6uTT&?UT`1|SUrcW z`jz8&qbwe*9({5vhyA{;BO|D@X+t~HfIW`nI2AR_s{u)SUW|-#c#aCCr5wZX_`qkX z?~0ztAxpIGz^aS0u?R0L$Ti~oopTV)DH2(EaVI6hzUbH7 zpYJJpu4JAL!sH0GATRYO;EA{ZK=i~VwJ6K+`&seiL zS!Cm34uAbhASCmS&+ePgOB{MrbLocc0_8h@dWKrZCaq)!<3Y{}a<>sP zeeME5S1y{T{$`+{}`>wd4*&1F_M^Fp15KMrl*%ZO7}(+uxTF(KXz zy5GG}Yag!;e$8x9+m}iuOa3}rEt7W#7s@C0;p7lasNE5_nRe8f`nkiL2!i(-w@;x0 zIF6OQTdRi-vUrqj)llg5t_Z}~WKaGOfQL|{o|Tb4?&(I;phuQ+l*H*-hd;}HL!kls z8d91&mQu>GOAKMf8jF$~y-N;)ImXi-j~oQ4EYHuWN|ms4ccmrRPbQ0n_-=G?@Cwzw zb7hD1q}BL;nNG~$rQUdVGFzOnJH#QT@v=RUJjik~e&HFVhkNh_ld`%B(kGAgQl%9I z3)Z6#YiNDz8s2-vX*p4&E8)>ZVu`b#On3!ZG_F5SHySQVoJ~@gJM5!EZ>rlU@Aki5 zs*Z!GzAw#6336n4Sthz&7rnvPXPc%F#&@_?nK5bef&p<%F_p252Z#%yDVVVH{_V0E z#QECdu?in%*`bBoL%iJn-`ao&^wH+SASMOM6cXA!em{R3S-fCR3n6IP+*UX)8z}`9 z9E<#1cQlA|EctBfiZm@pQMr~t!+$vlJFprr5*{fnD3nv@B zI1)+eIjylE*0~yEb#AS`lIFrgA2zGweP}e20$1O!s6SW|9_a=sFJg1WFgBGA8?R6= z&{gY=VG|O;w(2=yO?1!O8$;fxZELUAgCC-@DwB@pw^R~5#y6g283c~g7bl;w0a%Er zdW3}WU6D|D*m|LXVxrmaXOzjJJ)_dLq@WQ<8P=EvTC}L2AEFY9L~>jdHIM)(Y4G|% z=AH(JRoBQk^=DtGbD#{4ZYaUBJ{0CDyUcs`ruO2_E%)K#nIV?g19KNVvSb6yn2M;W z)u78o7mJo1wZeRp3`_OTBNB9B^5uk+l7fqC%PM>)MmF#%pPm&k?G zVmWyp{B^C~MBMVxOcdV94~tmbSw>dA?q~Zu0Sh_eE>rpG4BL{+Klf9}0h(Ckxo2T5FyWn*mgu`?cBY&Jvq>NU}ALV?VOJv7`Yg zB(`x*=Rg8cUAj=)q0W7OP*7pm6^i8&rQOtLe;x_xMxXsM!q8 z{XFX!r$msn3_(y{j8*Tb4B-AjXHON&I09B2PmjH|QH!kCt_8Zn^?y8(b$&VO-Vt*^ z%T4x@hhqV$KbYeoZ;FJUFD<4m(tXShM}CRqIW^N~@bR?*GLK;~!T zI^XP5AMqS+%TvRtp>X-~_|#fBgi+>|Jw=jIUHWyt?9f)wyD}9H*7Ur#>d~aF53-C* zFKz^C(ldP*&xc@%!e zhw6?ena{`LqC@QjEmEH%4OuCgaw ze>B(+I$vI1A|HBSP^@|%J_$X{nP!_b7*R;>5GHubt>qzdR0I-j&tB4#HNOp11p>PV_ey{130G5NiijS=8l8Fq0U-9T+v2!x=p*Bion%Vm%8;O$ULMQNzX4b-*JId4lG!^o4j5CEi_!VXP! zIyR7q}T)Z6Pn+)Y>k^!n--b1msLHzM!(o@lk@XG(mD1L)~^UXo_E5Iq+cZpif- zZA|;3-6MKYQImFRsUUJ!SJ(5}yGx$h9g*bYR&G(zA4{}RDWa_07&m2}=omRNokEps zf`1XO+GI&ZEF_FV7_p6FumK~ZP<;gTI8O>bc8eJB4~Wx1y3# z^z7Qg!ouL-VCd9VuYI~j^xbjDE&j%MzV!NY;~FFIbGSJ+Hs%pbD7V*aE}_*5cfo61 zMhVYUeT6aOIBD;*9HcMoIiJl;Z*f1b0C&~{N#!Y%q8#nF5E>kB2}KGMgdGwQ;nZr6 z?&5qzONA~tEDaDH9VRXeJB#MMOAnv}NZ6ilDT6;Q3kVDMY**0vQc^0t z0DmNJkMr-A!_a9C0)xI+o6I%XV@sVc!`gwI#!H#hyj@vkEref_BJ4Jg@p_19@!Pj= zEVqRiCVJp{G}%N-xoAHySCH;*=uMN@s)q+Pnc-UI1bFf-6=7t3KrHN&^Y}d~>2S<; zUZ&(1L1pb~dbulS5+=$nN2*U(6P$;tDZF2$Q`3xuOl&vevG|lPZxdZoh}i6ti>$Y| zz?ah+_W;Q+4wDQ)cUMaTCFz#&%<6BK_e1W)pN?dNky-lN44(a{vw?|;iS?RkYK}%q zW7BJAoIcf3BbICIQl)tp-+#2GK<(v>CbeJ$77?fyv6w_zEDZt0M&Ndgn_SS0xQ1mc zVhFBQGfSysjivk79zI1Pz{)}s8#GZnpHU>CrNyr?5+5|3nwt|DtmG%i+dRD2T4pK@ zCw5L5EvqIS5WmwN<2iTnk>l`^m;Z~+x~LTLKNJo{P}Vx{qwD#(?|}*SwhA?4S>%>%JYIY+?8Q?kMb6+1cWMB=ZBe**Q%cF+@d0|LpE6czEz7Q;SEZTLum-(Z;0E z%Z1h>DzTJ8`IcW$`W=QVZs{mKbTuZ7ZYgUZVj!rTMRTQkNxEKeVqhzCRrY`=rbdwS zmL;VQHg##wKQV3JXN7M>FUhSjvW|>QM#qB+0>obH4tHZ?l5y{^xbvJ{*T4QWZK(<7zYQ3&>_3->x<(#_t_w{dd?B2~81(`ry$&(3r|38H>(LruN1B;;H&(x?h z=Ir@mrK|fOK0ZE3P(p5n_iTC>EFu4w*TSgddlE3#@hM01Rg_;d5C-cJnUquvU%uMLzjn) zZ8{0p!uLt50vW)A;x`}0VdQ`vviOjQBgaXQiiT^jFiEYb@pKN+m3vBBojj7a7j3df zw>gDtfmBl=mx#}ItAuRP-h)d&7>f&|8-WWF5|EL|qF|>d45<1N;>&Rnp@}Eenq@`Y zEj$i;FNJP2q8@G3^XjW?cwG_)BR5Xz#PYk#z*hKE7VNZIZQ=6S1v+%&cz@o7th^Dm zWI6i;)Us`HkC~HmYkbz_dmJS&Q$y-)LOl`V&}%8z6`I>mYas^XUsw~7FY_4cUUd)G z?eb?c&T0h=9k1-+U@?8Pe04SxN!V+KYQeH?073}j3kEG7m)6mhNq35f5{lKZI4GV< zge)blt5Cgn>e-6s4k7}mnt^pY-2L>}&Pju<`_%MgE~ngOtLsEz+nz;`N-Ldw0$$x* z6ILg)rhmG<+o1v=f(LICpiDoiBAN~2w@20XnwAN^t*@iLm1Ry!QY&8JWi(kR^Qh2Y zoQGBQL?J#KCBDEE8aeWNUk~>sF})oOmuJHpFA8+zD}*8|8{MfZ-zqFnXp@o9;{s4< z@xu3w@EJs4j;rTb!uL=V%YNoBSk5>z9!=bBa*)z!ZTFw z-bRN$=&O*cQ?O5KQpPIv;9}`KCISSc&+{zY`|!&fvF)sGZx!F!L2Dn6*yZl<&-o8U z6VlIeU{K_CW2#e07iY+NQ7Q{-gmITY)6*)Ei3}9HLSp~{l-kF|w$-9;_0uj%sVlk_ z#`5?zp~wuZidcB(ucKMjw;cug>$*Mhn*p_yA|N*tw|tdGfeX1$lRIbt4TcH4Ih0l7X@v{+8yTm2e?t?$)4f8>>Gsp%>4XkRSQ zQ8ckEPd3O#adTijVj)KV`Zi0rFA_v&YbCvRkC*ufuBTm$UdM2Xe0wtW{)07192K{v zXA;*lX62$U#rS2Etq8@ZE_E@*P7K{7plA=I&`#gTNYrr??IU zDkx;1N`nyQ4qe~oUDMt)BZUHx`)GBZ3vM9fet#@&0e`(Oi14;)8;a5>PJ(C^p0|e$ zO%X8`%pd^bF0NP4O*(jo^5OEu&LjF8F>!ii=kzhUSr$GHkOr(5*=l;{vv7pmzp#YK z6C9OQe2q`G6c?c4<)W9E>!FC$EmM4&(+Xcv#e8~dOPabB_wBRV|?9djqN zY#Ocf##9RO8xN6fQ)f0buo5GIuOE=%6>GuQCvirr*V%I1x4SZF0WM1xG5+ zraEZRq4j1|Jc4U~F6~n0Ncpng6Rti5pK8b)2`e(Hj52;}_aJf8_%FjvQsa(~6>2gK z0Gua|`(a`Av~@s_!c9 zi} zH)1VAdQ&lXl-HNZQXnK?OdWvbG+c()QOsxJAFiXE>aoO z=W5>p;O?*k`Tf<(NvP)mus%WaVLJO&i%{wop+mF;bQg zd2p(zT#D6R;=Xo4Bf-9fb4(cHc7^CENW`fJ6`M90%;lI%=v>z~R&G?{f|!26vNhn! zN5%pjNH9Hss4?y*x`#6>u>>B{GmlFq9>oyyY#yn2-3^0gwOuuJuN^5=Iem=a%3k$j zU5;t_{w`o9ky>Br<%`&_Tr>l$RY@V49@$4GmbajoUnnxBHa`QMw88)7k%&Fccl<+0 z#vt|zohEPd%EEv!sl*zvu)bguJuc2N(z%?WMC^3C)X>eWqXd!5o9r%+E{G~4)?gCO zxCat1_j!gp&0;FI6lhRP16>?G%yp?sM8vPOu#3*`p=I%e4R9X576HJi)+3#y+~w0U z>#QbOJ}YuUhAi_+z*-y|Vt&m?j0q8>07N-Tu`3-=Jt(s1E>&D*ImlRGJ#5SXd#WUW z@L3(v#r!l|PW4W2kz;-8u5B^eNYXd=zbgShwUY3n=oRj>(N7`l(R8}b<95!-dRCX5 zk)wCsgNnc9zjOb!vhz0e6h?u4(p8nb?Vnme{3@w{LL*feTj1f57e$R0xUfkl5k8~x z_6pX)VqI1)Nx9ElEyf8*g}EpYhzYui!hwV$QtN-^XI5Xdy>5VqBmF9Wm&1VP-v%#j zAsg^)vNFHlZ}Pjc%!lByo=KoS@Sn?3O?c(@xhBl6KPqA^6oe`;LtQkaXH9Vc$E>Sizd|yGZuC8ce<65kr8R4664X)(E-4gc+dalNj~rN zm@oLXHC6WcJ@7WbH%2mH5nZ9cIAMMD@9*s`dC%&Wo3ViAA5L*`!PI$Nq;7uAkY!bK zJjctedK(ZK>gIAB{_JA^Ll~fE^WQ5yrRn3DwM*gh+?Cb{%V=bDsDTL3P_=>oM@Ig@ z%5P(YVtRHqp(!WlR{w7p3>`6R;b#MgiAeCZh*6DcRNBMGs}qIt0M*JrYHTZv0rwJK zvURs7T{iK}7B+yHkZk_##K#o(+k&AEdmh6aQw*k_`9N=V-bx$sd(vOJrUw zvZ34Motbx)kLknH&)^cGv5 z9P}3e`d6u`;o;%PyNjQH?X&=w22S8O+^yMW?yS+Y1`v{y){BJO5wzdhH0-)SBing~ zIb-4cE6|?VE%bI&4fm-nJMSB_*SWpFm%pRjr`*({Nb>mO(Z)rT)G?{HKU#0|6XH>l zwH@5tqNlbL^z>3^*Hl$i0iPg05@cs*_y7BGw$AkzFbV9nWr%vQU}0e$0EQ7eDZ!k~Dkdq9rNqV=`@@IwYHIQ4Ucte)1!vd5 z+tGiTcWY89ZAMfu zS0l5!^t|L?#&tpqwc{(`yF>uzJGs4y}0^!akZ9+$@e;*G&gkF`>;901MSmptz5HZ zLZk})L8_~fr^?i(2;LMS0}zy{e`^j4!PVnJsNN$%4kCkNHJB&v28LXc2oVd6Nj;Ci zkd>|P0#R5V!UN9Y_j+x^$7}Y5@HM|MFAz5hfrm|7`bGd~d4R$vP>cK0F*3p+9V@GS zL!$eGRzE<~1K2(RC1pH#idA5_5_W6dVmZwEf3j=*@3;R5j@;d9&#!ZIs_62=DyljX zQESX8$|W{M*u8)pv!O``{AFv$!$MFek6RXQ!YF&&_F6@$4!I8{D1MYX?vTvVbV~y~ z-?`#lc)sQZ(W!gDQaTUW@g+|PXafHf(&ztbuuqsBm%jQje()jx#%21feJxoWF+lBi z{={D@Ml)}i6dQ=mMy2X4dMl)_O%9K+FKuHFiPwvc#7AlfXl-P?nn=Lic-(N(M{8>_ zC>O7Q%2t^EX)c@lHcJ>0o3dojH3b5>+ zsMddr*7(XKbqB$+y7NU5Z|9#yb0vhW&w;tgZ8A~RGnCl{cau=gpb_(i$9!bU^N$eQ!? zcN=jkDl0E<(^I#~6exv)wZlbIdzv5yNZ-S9upc+OJnA z?WVqX%rZ|9PW;;XZ!s(_1l<3`9J9MKpI6mi33hLo$kH@dJY2N2{UxZpU!xo+%=(;k zJ@KN`(7v0wEz(1xFu<7%m#;SqR>mZynp{XB2m_cOAR1XqB~Ivf3;Z3F&Ch2eW@45A z;G(sU$RSXU9d;Wy^yVSEu0!~wO<-^H9mF;<*TF`wn8TNZtON_6>MQ=a(8j4>%IZex z;u_-@g57$$gN+%jtbCpGQ?O+c;xj2XO;Hq5%HyUdqMi53WX5g{)qV86Iby#ryz`_s zA2O1RgfrS!(N4IpWUpxjSmjW&el<9oqzhY9ZYY0Lv*fr_Us(K@cl_%QAYJW67g;<7 zC0~E8xBt}l(3Xmcbw>^oKfH3jfR!7{3j*@+RqxmI6z3ZK(} z5cBhsC3NF>;o0|VVOc6k_~Fz7feQ#EN3MUwQ&fua4j%yjMVhi+gq}_`P8yv?FR%XL z682guas@I#h0R!X?7y5)C za1-qc+yaAA-RE88RmKyr1bo%690Og*y-hEiPs8@2Z3X{=2ZY@SBHj>z%~dg}6zL!` zN5w7R{cosM?gO!|n)RhjhN4>BTj`5^EJC~TpUjd*tlzmt=^sbnSLo>gAQ#()cNpIZfez+>Gh z1EsoO2k>c!E$@UO8-}TEW^@~3rfE{M24L8cbI-GBqJbVuEB1L)a`J=VL zqfPuboCwI~rPEe@VP#(tTbMkb2vVqF!n>(vcQ($!uAkvljV37h&EJ5M*c-V;;7B}1 zJ5wXz&LaOxc^+OBluJU({Ygt#Fa2Mlf5Y`fL_Dee_fDHS8QT&}TS73>jV_z(5z1?K zD=EeH+FoBYs)5sSIO_K@45pD96LTk07`x;`mYP%UYRmTniN4E=c(o{>eA-yOSxQe` zROsywnKF*0!}cqMvBHkrux|=lAgz6AoscEXQ3U{QL^BB)67c=Ea%1Zg|KVCQ2{X#5 zr*M2v4yfI#y8YJps{aloYO%;j3fd*(QM8}~fxG(mtG&JK0G})rlR*&b2O^K5X$Qsc z+p=%LG)df=B(+NU4#8sc4vhnl>s4*tZO*-CG!;i@d+9s}yJ$R8-U;{--=S zvnkW27(i_X!}NN%UIIK1-)wvO)H>9v_})$H8`d2V^V9F?!PJ-{A`jmkCjTP5rPbFg zp5`Q5>`sG?990uRQ;>`_^7=YG}oBN{Nsopv` zP@^t}(aboh5FRguwCwuG=x1Dx34*)vw0i>~ODnheDo)SbsUI77f9NWkI~7GUEPqT$;bG(O_d#p?xW_={5Va{EOo4^O?U+JhytwjX2V{AoW+SBxS`SIx13$^BoB2gEF{-Y z^Mj5sfkQYNK<8z^N}EAR3b9ku1U@(NpE^09>dzMvmA0PZu<25xV}JUg9fxVy{wsQ8 zblqfA%ID1NrX6hTKNcIRrY>v%iP?G;wfrL8T^Y>U^vgwk9jlfM_P`p&L*mBo=aOif zEt`2g+cQ;%oiKa|x~JIq($@s{1@U^1wSL)I_n(bJJDW?l|C;=Pap15vZ-P;s2}#x<}+rL$P#`Xb3@DVVv$^~5nx(HS2LbU)4&u0M~r_d zcB=QiA8nVYnk%PMB=9CwHh$l)wF6e!qtT4Y$<`M~zyD_iyxE-lzDY*tds~dYpP%rf zM~`@YwhiVzzj8%zO5}mEx3{lv-~uu(kqDP0$bY((=XTan``^~yn7r;LYFIwU6CBIN ziZU$4CFXS7X~X^oVOaFe^5j>{HFH6Op5vLovuNxxtU_{D)#>=ZQt+wg*DLrTGTPes zpu>b+O33*;r^}B9RN9~eq{ZMXSJ3fd?lWV^n;nzq4aAGcAv;2M(2zrPr$fZ_ZK0l| zVCvkif>?WXc&Qu&JNT{^8QA~G2fn)OBw&Uj`dqoPNYM+Hw#%lQYd~kr`Ei(ggQ*Y+JK&_3xMa8X*0>_s;d+UWLO$#RSrotED?B?$0NBoF)w@A)utq zXWNg{e$b6A|79FBt4vHz(%;)FSkA&&Fu26N6t*f}0JP~#SnK)$P#@F)fuoZ5pULpZ z@L$BDslNJ_G;>U1cacr?WZvkjbyYs$+9%@$TH9DsnT~9=J)e+k$_S$WOZ7U%Z!rM9 z>Tzwq^z9)&r$wuGud^T}?t9SvW5Xqa7J-B&e{eIA>9GO|)ZA#|gha)|f1PxCR3$`* zF~B8-Vz5DOP|**Y8KyM|FuoCZWKlf;qb?1SPgASzgPCOl^SpuwzL`6EBJViilR2U~!ev~6xruHuD_#H53W zheyfOd9Q8)+fy#I2DFn7*!o{53{pXk_+K0#;S@Q_Rqxyd`J4V*MZ*Y9a02}*=Sz1t zR_(W!J0d>Y#NTVoV*W?P`oG;1_p7t`e7_IWfzj^^)u7)7$Tu>|$}#?<`r49F_lN2C z@82T6yW~c-mPxh)$#MJahJ!8=0U4(crMg2A5n@;#XJ@1gHWIoSX8aZ_6b2(R2eoTH z%OZuBZrOd@&mYUzS??rK;^{+mM_BYyGky1}&UbK&GFj*MoltMzf^MCYDbNj}?B>Sp zyE_+_n1}<(ym;~Xv1ZO=s~04o?MM1zNc-S*{-sGq`;p*{?T4gwUb{FgSo3!A4ocSMO&7!Ty}B_m^K0v!f7GaPJo ze|&g0eFj>*qTdO2w zyMK751>Dj4^;4QN$z8@2DCG#&#<=w!-B;^wx|fW&x0h`Sm-sr`LXPZOW3Y?l;v+sz;0#A(vu8C}B; zNH$0)sN=AuR+sqQo!gX7OgaD@23AtqY~&0odqM4Bc7D$H1R0{&`FCOQjO4*JnMzh` z4Ul3d5S`QHzsKP+b9KfQq^z~~Bg;u>R9M&8_=M0{h!uC-SyU7 zxO*A1%HFS)G#q|Qo|m2aO`P&7f`i2H%hUJKpx!$<{A1p=52oH=2aXY%Q^Kz&vD_S} z{uG0XBy?y`g$zEOuodaq(N!4ePS8e-zEUsCLbQK^0SDvY0+syqAd^rlcu|Kn5_J4Gl zXxQa0w}6ic7*JJ~yAnDE9V)$7zN|b~YOlf+TvtQw&?} z!Ek=Q=xv7gq$l*0@04PSKw;9bhLlN(B0};VraWm5)NPJ8HJ|2Ko1dogKH8(x+`!4) zdroP_t-iQ;mS%l&5dD*GU)0Wn!qi3kyx6}sU_5T)w^gQi`Y>HJGkSn>yZQ50vobNE z(A`Kxd>J_-UK#dB*U!*&03rTV44qUr6grQ)COf*v4i!1S-WWOmQ`=PXzz_k}Sm9O7LDl}B2+&_bTO_OYMmOza#2e2_EG8!}XS_&iG^byjjs1zRkK^#C zeDB>}`LU#3x_pxDuF_i%O^RP-kB*+fLk|5qWeY*TU*PX?COS!3$^t+s;$5ZYzp_f1 zZas^cKtk>0M{>0@ogg@TDSj+i!@&~qc>htF>Ft}a;E{khR)Z?m}9`abyT zbX6-rp$RcgQ#*b7X~^69C|JI+g9cklB{K-14aEbc-=N7@WjQen%r|gsLO0w}A1B?r zUNv>nlNl0#k6ulbB3STuu9`t#Vpx zVc|E@%*qtNyB{HuwG8x%gqXS4j7*5|P*A#V5I;>8s+zoSrsDT>G@MU}dxGdkOIT=7L%6%=uur*PGa}J>1yEf z_cN%(%F-lx+ti!E{c_O8VK#wWfO=N`y(%U4lP}PD9?B<=$pr@oF%Cz}FbVS$`_{*9 zgU=xz(N0s_L=3o~p#gCiK1d%O-6GhuTMN|n1D>FutE;INi&XF$mFpT`&rHp#zXfM6 zEudxUld^xn&4SvD+?B^Kw~=eq)3xDU@oC?z{9c}hWwlm=f*PHe2JJ4rD&BZ1aU#C9 z(DHA|O+)7k{qXcKP+IODrv)COP_ad;8cM@`auIoaaCE@%4*?C==2{QNXKv2h^i$1WmFTMArd4&rO;Gr zI_V(snsR-Mu#7y1E|mrz4Z2CT3SugKJTVT&#ozWhCMM;hI zLgh|WtA^FH#oZw6zAxwUf1r{~0&IdpPqj>=`@dkXSp z)-;CCSGXYFNQYO1_6iotv>-+@A)>?Pvli`QO|lK79oYaJ;pfIl<05@G5WND>Gm^{< zgYuptgX%KMt43or651pnA{|H&yq{$d%4v`WPF$VY$(U`@ggN`%gXln=fSzG&(5%iEJ?BPVU{+y!gWfZl zI~wMuRvEi4Oolw9Vs0}0Bn=i#I-8ltTTF}EA@-((B0ai_+#Ec{diN$QgO8@bp29FE zpGrk0eg!%`2#Quxo;(6thOE7{MM0JkcHWW?sbbqmgIX>twibaZ(C5M4+(s1Ee+9y} zO6oSfmI3G+%P!XEGAhQ!KT>%I6FjG;@U^~=P6S-8N#Jy`wFiqqz$=xK+l&OxGYK%6 zOgpT+li77@$RIL6Ys}3 zh?vB|>12@q*Gh7-|IOm(=daS!(Sga^ajJ9$^tDrXkAkWY@3S!~avZdgOn&rVP3~v2 z2RB&aHhPThg@rB|U3GT}fc;d4OQN`pfgH>mwG|HHUe7A~aCRHgi5bT_srHIH2q%fp zsOe%7(k(3LIB zpb&Gf^K@^5mcQtYMB(brm4UKs2*WEn4&Ks2b&W|vd>TFa3fPmOP|y_JP=&JR7j``F zHEwgdJ>w6&$Gp$~x!iZ@10Ht1Mtdi+ehR*j0PEY-RK4RA%SfIi{NDE+d=NpqeAaRI z*?+WDGrJBjewVpAqz_L=y8(65(a{M7wP#Q-JEt0cN10d3{Sah+&<_~jMp^Td9T>}j ze^S_v#Klt>Bc+o<19&5DcPcfYPSBXZOL&e-?zqhIBIe)`|90#H#^rTCY%vgozepX9$? zQO0tpdL1+WQ1Oz8Y-q?N_rb6m4<)3ZOnxeSPkVmuzvu<(_J98TA#fKI6kMb1kha-wE*Xe+ zt7051enr}-iwkT*cyGtFP@wAz8D`M!6X1&yOC7r^3Pw`0KgpUlMsz7D~>-8kv(Mp9^5R4rh~pp~j?WL;%M zHA&AWpIJM@Cmc^dA`UcAfo4ayD%eL?QqXt=%2NG_R$L;1pJ@UD0@&Eu4+GoZy+cVL z^hVbF- z)xPcQdyBay_k3^`wJ9p8g!=SPVTiR`hO$sEugAu-*(?ccXu}kB7)K)AV?`KAV^6F3Q7nP(%m4Sh@^y+C?L`; zEiEnG5)uN^u#g zrOVs(zWsC6GCkwXm(DH!y~%rPf-#UN&~SUY5o6pj_pyhQme{Yk*!9c>P>n@K8)+<@BLdR@qnJaPC{3AGVp#Yz zf|1pa;(37o5Slqux=p5u>AEZxpP?PKD23=6eGp|i+aq8kZ);{7b;I{*nveZq zE|;O@zP#oD$qKZOAHJyU{}mne8UrFD+=yKL8fWaB>W|iXWqD#=r5*A^x1%~YBo>Z?#-bV;3b{QoC88K4ba9 z&-Yv=GqO^!sGBX5r97;{VB^tvELMBOwUlzqh@@rr4lHNe~Q5xngUPz ztTWD+*h)IODm~hLsx-SEKOV>Dw!}oGr^8SGaFS}zNDbU zhJ2y5sj(X`fB#woc+wA@(+ffBMcNc~bR~Ad`|@vx{Xhp)GPhPZZ3xz<&QUpz&aEq| zO6J+I4@e_X5e%A6Qmu_?cAR#KJXc1pI6UCjPhr{^z$7U*;JHnYxF#Tq4z3bSu6g#W zM3Ph>K-L?%Oz0A1U&27S^IDpIo5a5!$Bx2PIM@GP`*R(g2&eQkUqYM-#avzP^LFHk z1i4t&d#ll$`x5xeX_=Luw8Ef&#KdiBGxD|5V)|8MENP*Q{hLVtN>2#fEOE)ywKrPc zk5tQA_`Qo-|9Jug1SA+K*^$YGLx)Gh+>P?~XbwJBwDn|ywXaa?v#99FJA}$JZQ1NU z%$v8}AAPjBp098G{^Cq0v}uR%IE&(_AG(kaaR(Y9So>TzYHTNK{9n5n>fuXsOuHK4 zHf9XoEXY3;FHgSzLlh3v-%WPDY@3mYBJah1I(X4Zl)*1%Jt}r#aQ~pY1igE zxlB}*#~H{txa~=F(yS6H3<`i_5B@y&X;4GBWRg}W*!7e&i57qgQIh{JYYexzTlhOt zBKVkCRi8vS9XK(mPSz_-9=+PP>Mm}Jdv4(D*(rIDU(;|>TVd$LRCYic=OB+e&oPub zv~AQ$w^QG&;Am)Z%PvR(9!}hHaWUt=(3t{Rmp{hGIF%mqcJ=D9Jl)i$RbM61$nQqZ z@qRrFRZsQmvmtkw|KWu46|AF%I+$k8GVMbj)3R&%{%Yt-axE`~@e1Fl(5DrxYD{ED z*%THlOOH#+v*EuGn29!tJN|L1D(PCr=xrnrt4JJU29E=aG_U>k6!CiJ;i_i zij=V(xwT-ofGxCXuchN53Gb|qyE7S2n_`ku46TS>lig-HNF@9he8FS@`n{(QcNGQb zgzEiUpDEDGZ<`n}U7M1NX`St8x<*DAKFPZW6`D&dnpicJLxi~zZ&U6j@{5zp+r7Z{ zR3l1lX;*^=h&b`D5((IsKqiS!ypEk}gHL69i7YE?q%23q&UV`@e7-z5WSHc66$4wT zz83E&3A-PM&?5DR&AU*NgcdK_d)9>=c$C<=zT?WKa$)Bg%(pxY!7amMRHSM>i6TgN zr@xwZ6)YoKlgc8kX_(P1%J)ke^PiRauK3}~huU~Js!@tq&+5iCz9Zcrg*|{ES>8*x z_~g1M@3eg(l8YwM?Vx<`|4n=Q6-g14qtE5;tie8XtPuxKSwRHHJARw;)3A50U$vs} zMk1he2n{<78WN+&H0X-MKO#C;=e5>}zw?+$N!M2hVt{4W@xh^2v^qfPrTDY#d9EJfhgS4QnDUO_LZjT@u< z(-jiNalMjR{Cr`t^v?p3tlVJ^+Uwe?Yex%NIx}JMo4uG?hKhv!TM1bs%ofe1YE56= z%%)G>eXyjZr#PRu23*azp)T~kXYoyk?U$zlbbfp!G`Y@C7i_<9O_2w|t->9!5#*Aj zguACxrmw^0aQtx~?xR^@@lcPg6(?%@^Aq;rAvMzjUdGE$mI}nM)Iw=BGOoE|RMEGQIEigY4WLk!V^0*)r}F7t zF(_IAEde0{lwWG0F-Nw7)2?Fg^oOVWO|pt@Pomj=0O%2F&CjzZ52D~%;xx88llJtp z?q^9YXp3G^F=EN|>zVu3#+19gf}vX=9t%;%ydVJ!YP)v3^v4Q8z_9wAT(}u+gXAf6 zG|}=@pYonkpPy$-B3lXWvK=Xg;!VuxA5%n>dt#2*)eJ-IcBn7MYLr%UCr!HV+|Ekx z43Oh0yO#dpdTNpL(vTvnAvo@9saK~07uy)S)4Uvx@A=++kxGU!7JRHTwMZz!7GH>d z3clCMjpg=NAPbhITbuUje9kz?g8}4HXb|@y+(b34u1lF#gK%UzQ+pmA94O@Gh>Z9R zyBAg!J+~5kFWp$1&N~H;VxfMki#D{^(@H3P`x`!fY4u;X=3@Ji7jMZQboI&al8lKp zRQoau5NH8RGO_tZ-7or$`Y;Vu<_hrpgbo#~VN}`xO^S9>a*rcRz zV8zh@1kYsD*yDHEta$Ib3p86#cC#>%k09ka<&2U|ZvBi85A8>@Yl398p6cYNrlgGE zcyZaXcg;~3e-}n>*7W~@vz-2&-mB}wNkR6Q%57G~U%OmorYc+2g$Wq!_H@%=&ieYZ8*{Y8yweFiQpRi+a{+%RZ3-ZLP z(HL|d74zK*CSf^Nof(?wESKwwRbW*A^v#ZD{gj8bl-1T+GW%m-Knv zmKTYEfq}BRdRrI;55nAIl=1d_wI5cR_dyHMKrCYHFw8hp4h46!2qK%89 z0)y&rZj;n0Jn5d#HGQ`lC+3<%&>nWkye>IM8~^Px2|56WDkC?r&ex?GLIRZ_U>rU< zX@va#;fl;fm~oBE9B{kj>qpz77+Phm<2wVvF}mZu;XA9#`=d@mSM)%+T6^&}@1Np3 zC0IPX-(kQ0m=xW1E~`58G+{IU`~6qLePY?I!Enrh^SNiL{2@vEmgZgB8P*V;9Y;e4 zX79r}t?LQbu3z5)`jHr|W8n zdTzk@(>D0)(!Y-WyC^%3QX*|Lkl*o-at&P44nKdJXkO%)!_~{wLf+TrhX-|23I>j` zFCeLfoW5}--6tB0 zfEtA&W|9m9S1i9tt=rhBkfViYa*4-*T}x|g)Nze^niL-F1<;aZOBX|2&5Qd=$pSfZ z1_Q&^_l?q-`jTWKN`tm4CzdtU9@I9xYKD8Gqa7=g9!tj(Oz^ud?`g?wc=s6G9(xFt zR1=AFG#`D+_k-AU{l+nrd+%IZnzp~y1}nKON^Ck*#CBwMwkbf|ZJQMg=mCayRjro; zY!EMcz1Iy?Qwx9fNx5g1d%IqeSER z#p4yPfXi%umkFesx9PDM!gM}~dw&Ypg|{@Du5|~Gfdo{7qVEb+Vfvh{o7Q~>zXzeD zWZ(#9Sy>sFp-D6i4I{b)Y6wC9KOgnTWOe`L>x6{)UxV9Ny+tziewz=u ztUU5L_@WL7vy2X__!{{M{REq#$?x8+?(b>>i_~mux=xxdeF$t&?E`Br z3N;J})MlnK=>g$kM7A;qAsf9ulfY2f0k1%>Z>!z#S3H8X*NK{ek3}eJL{x^%n=%@m zXX2P?5R1C9%e3ZIKD>CkqK!z}a$X^z{FG-7&SCf|a%6|6*NsJmCos3-rTCrwgk(== zu&5`!INnX6^vk{0kNNNOG*&B4Rh(in?dHk3<>%wAuNAAUdJqJy4S}Th^nj>TXTZF1 z=F*w42}m-r41`d=eZ3;_l{;vK37YrQF@968{-sXBHR0mL)_3UmBq4EaXd{!I8wYap zOw*L%)6Q^`b}&jQ*iubozt-91=?~52!=sr^;>RPCn~aLU*?T|~maIZvFv>3RJrp7( zXl$~!rOv*`Of(#X;A>Q1PoRtT;SmQncElrgKlw-oz0d+MKr(;Cn8RN3G%j53BiM|8 zW6e$YOfO!3m6{PHt}}oy`&iM|srr&j7mj61P)0^63Bc~UHDd7oI(W9pR#@e!M^wlZ zA=EX4b8fsFi6i9(fFXQO5}R;O`9T(2Efj1?DY%DhG&7U&QGAo>moJ+bl(0A5UM+J# zs@6SIa_etmMvT?$)L1eW-S5GDvF}SqM}ZzsW%4kzMHwSfGHUEX-3Ee0aaFdv!^89}D8|0^(H!FCn4|5|xbmH5N`rh3bQ{+%S8DiNzWjs@hL1=IO?>E#+}O=QMev#q@)W!qK6C?Fm<`sqXCu4l}qD)l&BdmVlx`WQxMcu!rgmKyRsm|*daR; z@kYG5^>|%WdZv#L-YKHaREWCGbA~Y+=tT`2IHP%0Lw_{QY_(7nF3 zB=g#)$+6l^r$&LH_1RGx)M0?37@u8Zb}cj{KhUAIJOfuaZ9hnV_Ae6aHk|Nr#`|la zD`A6SZXxjl6KE5|$;GMW6+fh~>gUgnhfOcdXT&x>`A!4d!MA1STSddHB9g*j)Ftem z64cQS(3vEPzeOp|8SeGHkv&|STntr}XL-v<{`yi%fm@oA!aXq*2(|*?U`LE?l1BeA zws~St;~^!KF>d+ zz9D%Mn8-d7=fBMaG?7H+&)II_->D_TPB8S89=C^KqrP2~##Mj4_ptf`{2~Fa?N3T+ z8S`b~(_Jb-(HJ^tt11Srl|hE^<>6F($eo$j9tVqmxXH!;y&Z{hGLfvZjb?Zynv~(U z{@NWbRm`iYskpM+qRRQ1;&_a7tV@y@6d&TQnVn)gi8%${e}_dyq+W!O-}c+9KL*}0QM=B)@ntqfwC16VeU-2lCu;27YE=)gu=DmAR`^} zkSYeb7yiS+sfvr6cBEItq0+|iG*;`*!6J0iU?D&okDP-D68p^UuiI14RrpY>Y(FmX zx;Ec7v*pHPyl8Svar-4c!6hoRr>L|B{4SzRNys7wi!&f&W+Wfz=Z(|qZp{f#0eGiN&yv2``KRQI{GHk=^99k=tlH!y^ew({tPDeWRL z3E|)*`WJtG@fNqLuBo$;)~*>~58VSJj2UjMPvh4i_)Ny{c!j?ui?wokxCYQZRAqQ$ zsuDvK0@C32<(5`;+v1SxyzsPFU2*{ByTlHE)8N-?(xo(L=+ii?ZKa=n;xXx8Zig^MjofIHS$s6U_PA4 zu-RD&`P`FGR^w+gdq`vN8o-LJ?rrsOA z4F`$VC>!R_c&zpd)%e*bp6svV%fufBuFAhMC;xjXzY>fk%HWulcggf9wRxRhlhGj` zsVnT^O7_;{1jQTwn;HX{2B1%eN?%xWftwPYJSl}@>(A4kuV@J}RdSo#jqD;~RFKX0 z%DOfKTd+kE&IB1Q#dyPY12x|^ejbJb9 z_gZYfLw{CtDq9wGK%iWOBG7nSQ7AQIfhQBGenF+qKkx+Eb82y4*5{ku$@sDkb+MIr zgyYwuns;1m@hz+!4HrlDuHunlgG#M&_s_FG>(6&|Ztyl*s&xNRLRV5OrcGI+CfHEF z|4Qk)kbqhQ^;TD1T8t5Gq6!yHqG4&}(1z~OgUx4m7E$evi3uIcQNp4hceH25all4R zN)!S+B&;94heAGYu28GAgdFS6el06Qa&N(Rs_+YG5&fK)Q1a|KJD4GfUZ}N$Z0}EX zn@!n#tP=ArPfsaG$sH%2Rb-TxkHnyV>CK`@z$hdM3jPyViQIlm4lx zD2Umy)N=<-Sro_}7A_@!k4s@ZYfWU=?sy$jxR=#_;`Ll5iG;GBV&zqG)L2>2zB14`$-dakqLN1UEb4m~A z3k65-Ef&c(X~QeuF(!3T8sy*=S6yPb@+0>pUuRx<(F0lHtL%5rgj#!>Immv&QG zRkbCO?s~CR552IcXm}r!#B)rS8Yb{B)3DrCQ&X_B+sxbgh+Rfi_BNkIT|il z{*K7&Q>&&YAZ;}_pMT#@K|wKqP~_JUtQ#nkDd_0v^78T;K^|^vMRd-~Dwafv1kD$g$zORJd=K+p2MU;-hdDMt!7QvNh8 zjG|_`h)*T;`{!QdJ7b;TckrQI`cRZ0!r=MyYfnWPe*XMf``PyC zQ^vnuVo%3>A=Tc|yLaz;v4uH=?i1rR3MA6aHGl$hv z3AHY$XvEn2oJ^olP{RQd+Cr|(8SkCDceV8Ng2Ci|0tL*pMBNY0i*EG5D>9t5jBmsf z?^YNdd!%qb{bolN-3|ZC)Cf~mK#neiQX=+BsgC3Fh*2UH0jh(o!b;Zc$#M6mjq!Ls z+9${VJ2e1ylLw!uO}+N1>W_Nt&D$dBD!1xS1VP%*DDHrYjg6hR1;W?%Cwnh~%lALG z=-xLDDlQg0g`%dv-sL`T-UWr|4+ke3>O)`gTk$&mDcmw6C54oA1d^#{$tvE+I&sr> z^eD7WCJk&x?=FrI^7E@T^a<$vw6;|%E{HiP8@GD;lz-ZDw}-4~hwwENMp{Eyyv5f} z=%W-7Yk}Kb5DoWDMTsAvux!u7z`ZlwijkwqME;ADnG#r?1AtRI%CRd-pwn)f9K%8- zrrk#8p%v^hh4wa7AdX3&+{ZLo`RwF$wsn9)J&)`7M5l~xRs2Tmg~u=}{+_GsRLi5# zM84@uECkWx9OqWq{F2{%RPTm10@B*Rz@eLlSMa7jk~sxqfXNW3wK^3KZm-+IgQ8Az z6(671?1i}2>SuE)4&LXlKjN7}#*4ArDu8&HYkdldrAWUDdp+rLG*^Z&=W`ZKov8cQ zT-N*Y+p!iKdOpic&vg8N{Ei?_90DAOt^C@w?Oh&BbLvNMM6oHqR{72RVP|{aKXHz1 za%};E2xX9@Kz(Ps)|K?_>kPtzS&oWnIlYj3Px+DozTw+Eua{IE_?!}mg|}XaMN(N^ zNsFaa8JTC@NvGIHVLDQDuPp$+igz53Ox$QVcvo#IF`k}kV=glMaZ^SV-z;U7#>7mT z43=+yVE^iT2BbGGB08dm;M{j6Ct-(kZ_KBj4yVg-XQOw}VW-In-eOmQ%`RdxsR>l~ zk!m&F^HOG+w78y(lz5Eq=6M~mqnlUByIkTU__m9?v1$66N!f*`YWqhb4x)`;Ib+BP zH73rN`+qM`cqgtujZbqm#wyo*lDc&UGbMxH69Xd#L&8USwu8`xzE9mUe}4J}cD5UR zkzHBi<|5ge)^RE>V;X@fsPsnc7{XmrTmOC`h?9=G=UXCtnE6Gawr`S1JN3G3gJ6z3 zg`0lTdE-%&H?eb)G+bJf`oc^#Kb5S{(}-MvOLpHBxx&UsABN4%m+F^$DId{iPHXx) zZbO=8(RI!TwoXVQWGeaY{m}IbBIJXQd@T>XPdUW)dnUVL5Y~B}Yay|3f5|>5^SPJ~ zm2T!`-vW)%SG)T$!ST!3OL(o%uxR?^2cf{BFMdTGUZl113PR<$WOw&BR6^L`*6OcM zr-0&%(jC)TCDw_CH{cFFw2tSTX-1rPUM*s`ZbweAn=vv%WL%2i)pIiL!pro1K`n5G zPq0HMZa5;4?6-2aXBG`yqMe|r1s)4Jk6ZEJA<5_OEimXL{Ij}jQbG!!>!emUG%9yC3+MO7lf3Vta znx^m?+{Yjg>{SnaUwH+O6`!#p=ypLOaKJEcp(_O!6Rj7lI-&wwl?oViy_8r$2@O{S z#U=>+3t9$;5dhKl64g7QxECTRS*o-uLYEEl%5(>S9w%QtJg7iGj73YqO7mu88~DV@ z=HDO&pM7rsP}@%F2 zy6af9cf&z6kQvGLYr_d)dN1cBfzag^1dBS=muun$xl!ceUz&G?O`tC-dnx0*`C~4U zwe}qnkF;C8lnczhH$IF>KznnF)!0?uhlo)!>@HgL0+5yQ(?I~MD*u)fOa0MZ5=H@L zvS8Be#7LO0#0E{&4#K`cx96|P1_1GIxFJ3|#^l^Z)Lgp!y_WZz+?Z8~isTz(weCA) zlSH`;k^xe28Bhu-krPX8`OplKFRfW+vV>a-Z;J4+q18+~r^A)vpfP;cw$wa7+tm0M z3dLCPn9GraY%C@J+4_G@0%^mHYmT&RE_no_)0x!|KQ31eV{_*`rw;+Utm4y|xG)_38%K zogYh6!yV9;e$3Bg)_uD-ci9T%|G~E<${8%cEml?9j`xJROtm@;fhUM$}pU20^Jg8p=sKvsV2V=W~CO(o>b5DIiH>lyk)z-g{Ym zm~fq(tnI0>d;Jh$NQ4rwNPL(-vvskc!N=QxFe)*Dm0!~IE@yw@hcBg+jtSa)OU$A+ z0abbGqpkqOJD60mw$#cn{M1cpJpY5N4VG@@QXC}+LziEB+Do%LXeL#n1DP_+DQZgG zcqVAwiW`y>bH05&7X4i=@l>HJCV6qa7$Bz-HpK-BgBE6*qqug)hly~g);JjXOi{^6{*o6q zw-@BC%!g#|VL5C6O19$Mz=!J{pLm6=UQ&j=j$7I?ZC^*>3Q36OMl2Jc`SOd0KP<<@ zP}CT~vw%{&QlMYXJ4ysZ4XrHj_5}HltK4EDaIzh7-9yOgdo?2CySKN=uK0Kn54rBK ziAWx%PNzf7m+A^Ns&U5Wbxa==zn5ZAP|zY1Bd81k>d z0@}(_Cr^-B9ywi(I6H!BiIe8OB zLe#4cwho04m~mN=AYX$MNqogloA#-GBMTkFBxoI zNPhbPofQ8xXf3FjiA$IiDs>8tjw!!#;=Vm$M5&M9U7^n&dmTgi5Fw&qb(t1RjE)fj zqd?gL_lHFQ%bu!RbNJdm!NPeNx4yzYLNJaLXKPHX5}|`H-K6dWcOUjUw(hw8{DSs1 z(g?er43tGDyGYjzp6kpP30|!sel7pv%$NKDYmqh29f~7>etbQq(7Oe72^;=dtS0yH zYOwmY6i`NF<+1@T$QJy*qmHddHM(VmN-pHF3%bBR8ybf9@gqk=42}6f;fE#s{N+o) z_=1@)^G82sM^eeKdUcY-0>$8&z%7+B(rHqkIjxSKx6dZN5jDZ+ZyWw-7He0ycQT36 zhr<{tOxfkLT0T|3-`}ccVa)<#A^IK8I;k<6SpBj{0nB_S=TWnY)@*;+oq}T4ZVM6r zJjCS{p8W8sz5vaPeN%=z#}n|B$v@4_AbJk?)5($KjX>PM)lRlsZVUJQ{-=HU9TCQM zBKMQ@izAkuB*5>w^zXXPcvZ;Agl8kV=22;0rG%QK+}vb7+9sdZ>;fcknGOg3!l=gF z<>mK(xAGsa`T<){E0@eLhV%!hh4{+{9k8oX+wPm65`~$Y3wr;*X8!ZGfCH+tD{45G z1b93sQ8!;qLD#+Zv&a7M{j(!2P9i5U!HRGBFWGPfX*aQJ2Dx15xoO_)9qsoxwVX&k z+g4vtrR-A=%Ovh62bW$&wTBGXZCd{aPm)?^G$*NrD7{NpVOAdg4cE ze2)ha%0k-=;91MBitehb|2;l}vPe5<(L+Eh?|ze+np*VNgR*Oahg;6_*8P)2E6J1j z^>CD^yl)^Ax|V#P8+!0teNLE}FxDyVna*Nb9Dkba_tg8Utqt{sN*`N;VH0Qmn)%XH z%U4JU-Me%aiPT}L8l@PFNEu)UW97z7gOBPKqRU97E^MO&>L?No zla`XCa#n-0KnWz9cUW^(B5VSVriS#1ikQ!6Hm@kE#W5J!iLFy*G@54=>?W7Ig^tP*C4>G6BL)5#>6tyiI zy{`&M2;8N3Dg8#b0%z&q9(<4V3kh}L+uhEiyZi3lyF+NN|I5<*K*r9-`(&@THQoD| z7m;y*AHsUb_G-iFbI67hd#sGSkCp<`X5cQ46q#Ndu%#Yasu711PeeQ zxpj2*?-|2ipcdcR*#VUX%YOCgsi(8E^XzV)OuIi80SXmD%p|9zgoDt^*4D-Brz0Ux zH+s4HA2_W!A+lYsWk87CyW5$fAi)I zq95O^-Bm~Y$@h(4ylIe*&fakrGm2$PjEWg`o^2n+?nR;sj-2S9T$*#QX<6Rp)=Uwh zNk(1&XbQ08ECqR;{u6pK3mzB?Sxt%YJI7|T9MMZN;&gKn5DO7{M|Hmhx`O6^8tWJ+ zyPDskwTJT&CZj15PS{WmF^PFLRZRuc)oNzWHs2o%t?ysDa-|jSc?=BJgD&~pa#@**j-IDDJ0Y9pK1Xm%5ju~kD7Ah4 z(K0wihQ(GxWKP+<51Uzt-Mt<>pocM>>+w6hWKqG*OLRhGPC0l=1wS1+hkp51}-!hx9>Uz7M+4y!B+@ECb8Peon+vI(I!5i9SWK9k#i{BsnmBZvHQZtLvtK3fNV z6g8vOJ_c)fc%oR4`B~^(MyNE_w@Do(bZ2@T-;1N|uVWbNrRk*3fw>bI; ztP4R~p?skU$(bB!5syoNu>j~y3d~k?y>o0$$leZOH5_2WI#7- zHCSOYr&f>HefhbVH3a>_61{ZR(_jKExp`@Hrgfeh<-6b;WKd$cAJ2QX?|{w`nSMIU ziDDz>d4;8hcLd={20XF)+URT!{GvGZg;eVE5FIO3CB5QjG)<{IbpP-H;!B6wk`77>F4yhiA#yJ5b;vZX~6x zrU=-W{gW+@(*VWhuv|x<+=wD@%0UK%YRr5xotaqZ{U-z1TwokTM|DR{{aP!H=sT)b zBPcx%VhH?#9#Oi`xuz6}i^mugw?Zur)BJ$?f4J}JWio&+>!1-;)c>MWZvwvBXK>h|5wGWepB|5DR1=8qJ|q*0Y^vBNR6#5PR@nmM;a z*nv|FnP~K=Gpo#mPjW{S{v4*#Ca?QFo#)ixr$Nb%`?}$q(rDYloB?2q-vDH9v=QA{ z&XT^`jcA>+Bd!-7z^Oj$^1FkQ<0Z`c+)#WOi=IOM7BwT*;8u>!{=^ovr4(jP&>^Wr z2KoKY(=^S_DWg%_tX4c1kGj6zlk)mf;^b#OxwQWXZ{>Wx&Gx>i+T739{Z-o2q#xylF6dr^(@9e9bx~biOaZJ$j>%UUhB>J)CF8Ex-EtwJxbHich}7@{!hDa&okAt-gIQ z+KBEG=($tqw-ECe59fQ3I4G{UZ`1LfEB3mHR`b`I=o=<^6zW$9x392UU$|5I$lDSL zi9%4_gxpwVOqKVo4+>)Wnumlmb{>< z0}PIyeRq-4)_1=oLp+Q5w zVPT9KT7+&bGpZ3BTKrB@4HclD|C8Hlwl&WEI?ZM;zwGd_?hPmc!6XAg4qh0Sb@vj( zb1hy%QS$JXW&87<6P>ayZNLw&dTxif{e8qb&A z1`{jt>1cj>XieF`$fr36-H&|%(_ypPf;Y%enlGyda<@Dhf2Tc2QER;&Rst+UtobFH zs~u=K8JUiHMu{4eM3CVW-!OKYL-Ns^*2&6~mrxLOed!Fno0GKt=o!u(4dawOPqe_j z-S`)Bo*Ht2XwYNQ*i|$$61z^;r~;x6sF(-;5wthdnb-luHw_p{{lQv=l2g=YD z6-E@yOTb1Al^@LWj6OPPxhj3LWJYx(MwNm6Q1G%k%>7fUdhVRKwACC>gz|+fms|{= z6GZ$l+W}m=V;<#CzHPy9qxEu0he>1AJ;)os9FtkVx7?vrD9A*@RAdZP z4-FQwgMm4X1DuA@CN=l(AJ1!-K0ac)3~Vc|X`YWPLMF&{gK-tH220lvc=V~B3WMev z`dug#Ojr#+6rcF&se9m$E3Lp~J@BXugLIN8Eq7<SN5&q4r#QD7^!Y8lBcOWXCne1cfrXaUHE> z4p@0yi%pH2^IWE;kVt=8m@b52A5Fs-r3^r_3IEFGgOkAsDJJMR=fTZyX19@6Txqte zUNk%#T`TNZxAWrBGr20*IJw}AfgmZsA%GgNGtxJBD0}Jqo=h&y@0<&<&2CNX<{rPO z4HQP0ai8AeqOC#Fruw*4Qh(YX{OK@$q0NUV9;&aXu-oGBLB1S?uG!_mIi14 z?pBB`3RPj8grB9pK3GDAPh|Z)h_HqjCW%zaw zjQVB=1^-na8KWCHgO%1zpMF0oL-{7Je3&iM_?|zTaOZGF<1)fpi&3jlq%H&J6m0Dm zDWvB0dM2@onfh480fW^aBr4ef1G)W^XK5ik)xBr~9e9VZr_RCW*x^dEtTkAeuqQrL9y1qi@%3{eAz+rU$_wG9@`r z1Vf0e!$0d9)GB+soKFwSuPS*V#$LOWWWwYdYKL&IR3`IKeLZU{m#yaJ)^jjW5Ycj2 zmAd8lHXN2XsD0^V4QEIa`~K)*aw1H|SM7ok|E`K2(&k z-WN&=3Yaj;_syxf03hq>N;W0?5%dr=f;aJSP)qG27L0jpNzXrcElFExFjXN=0gjU( zP!b=u?NyRt+ys!SLq;9VmM2@w%*M?L)p8rB(8&&Qn1z1HB&RWe&k1& z()T8%)kpm{IzZIfxBoEPblg#8`~r*EFJ}SBo_lQHMjM{I9bXsTq~n1FuoK_F1fwZ6 zx7Bi&@(y0#p&c9S&KqQR4SzNMyfz2pVA%u5=Vfm)$e>)@b!51;3QeGUi-v-UkZ)bm zfv%O9C<89k6rH z|FkV??TqKC`&{ziliOmf5-Bt^x3?W8T_&F8efqTD%%e$+p|KFF$n=!BX~`0I02$f> z#oNi<+yIk^Ui)(|eA5vZ6iPrNCiqB9OsnY%)pZ^PQgK*jCWnV9w3-`L1~2qCDlek? z1_1Ofo2maBIoUs74mwua^3)38krC(pd_B$fM@f=Pt;eGOp(J^0i{*bT@0EbK+W`;0 zTX&KB-*M|5aOuyr(wMf;n0opxE?V*%mJ=Eq8_$a%JOOCYmb_Tk+nA^zfktiRQUbKP zB+qt)D4QAwk&&Py)js8tM!z=779zguZNQ{%F#jXV9jASL`z?msf;V_B*r5N-dN#m#>*CQ&PAs|Gf)N0c;u)U`Vf)^`aWz zmi^X0T0;}+jkzr}M$IjA?>!F4!CBeZ&=5JdBM`9BP%!5X>Jt9|exTc?eIW&&eh_hhZeCVLXDC}HJ>}@A z{A#N>9;zdfKprL#%^1Im1{R8MD|cvHhHC3VzOU+~2-^#5V4$b%z{K8x{_zkWX0pQk zx2hs@0U-@+(lP^aXMxwm^#-Hu>-OG_T6`?}{;lF_y8pyQC5v8#!VxUAg%SpCE+%%rHkkHC-=sI5 z6b0p2&OT^uGb;^hKxKv%jBl6!R_5X%8uoHoZRJVIB|EF!+b_cyd<*&uQZ;(<)x2-6 z^ep%BIdl=6V8d_ox3(QL*-2WXP=^YV7C?__6l>gei*jsDty+0xQNSRGvF%?iW= zy^p%5Ri&kaZ<`ZDATg_S%aHrj+-d=80GQe=@bv(lRJE1kkv`elA{54E*}6$;;+32e ziGNeGks5=DD_yCG`Ogby^7}j(pM+xUPsj~#^0)g|RGJkZ$z>)c;{KNeWLDg+1+jYY zh+K-^m1g)OCG)GqFVeXOOdMv{-nAG$c<}SpXg*P&!pfw5ka!r-g)Cd}rqafj6>@pL zQS`Yt&)X(e;vu%G$GxOygiAKZQ@`InbS&>TOS>|`|Ivxnl9!wNI~2E|bR@V5m^HGZ z&hGA%{)ch#L5iNxP6yfo8H@eXr<9*0ft*-8{`9e}?e3R((C(On2t>Yq3-A-g-e)h7 zS$^un5h)qesSr3b^4pVvdZXMb_WpfSNiT}5Rt3ThS#_si+QcS2+}+Sm*CEz@soRhv zA%7U|NRaYR%+Y>g!4G&#IgpUtxpPO*=iCz|$2|g^ke%-9eJB*5?cWcs0^QypCOg2` z&GmK1`4&8>lZ{gO`XZz{^l-Z?ap}80HIuI>|F>kn-L`w%#Va=>%F%?ahk9k#yOi1A zEfdb&1u)<}W8@2;|3( z&dyuv>crkB6Ep1m{Dg>B4Als_CO~)`f<-F4vQoUIt!*B`#Mip-7CWzM`*<@d2)*uf zy)=OFiY4Wj^=xg1gnT4T()HW#w{4RSp`aThb8*xcz&umuA@;fM?*lUwV~E(a8^*y5 zJl|{y%lWx!-#!-AhnJU&@*7o>BIE%W?AAJZu)jaoaBC&afuB#1F zRuvQ0oh84u$ktE#?`eobYG!^&+N#1$w>ka6GUK9Zc%(5T2)2@ zL&L*tP|am*hl#{cM~j7BCnZoipQRR+D+4ASKJYtricg? zeNTs$Xu;Rn$HMfgsglD*4IF%YcSFu3YJck+M!L$5d} zU%UfAal`@72AI<1O^_$d(7ujpZf*{$o=U)!c`7;)x@BATV5fs4h84=_@F$hO%O}7% zvmT-$DE)V193y%mxYlc@m8Q9V695-1&S#kl5$NK7O5)$WlXzShBXfQjS~Ta*40UKu zU{H|3ECOhVQ1fV{5%!0+D?NI|g}@r2Fb;<-`;8k68d5tgc~08jXXfDXh=kt<_~+u% z(%hhi4AahWoyP(D)vF|mS!6KKr1^4DE&2P{==3NCvH9>#Qg>)sT) z9g*g<)HG=?N!cvs8%Q1ey(2ddPs|~`gqKfazb$^d<6)K`pBm-7&7C1JeVGRSpUB|b=$hZaRw7ww_bPvQs)&+o4EYB(!6DK+HE1q zq-dhVLdjS0rZ-ZfEh;%*yWc>+b8MaNMcsGvfYHQrN5Kr{ z(?O%h5sJ{1R8$C|sOhTr4!p6{sA~7nLu3us11l8l{7T^@``RcEk}+Xyorye@Fj~$hd~oBqjq5np#S@z z%F73}4UveOSAAR@IjRxfUt)ee`}k~sZ*TU;-K(t;SL?+WcPR`=NPwJDvsq2>QSZ~Y zorQK}dS#(e6**Md|4cQyo6FViv@z6nTGH2#$qZ{Jv z?C6+{2oG=btm|5ukB^T(g3`0=yah~Ih4-5`bzh>AiR?c>Hf&Ely^+vS<^3B*#ShUg zz&`f!CIYHG2X7$DpX2ni4eL{o#?>;P48(Z1Tz&Ah!`@;K#sx-6p}gVlHJg-zS&tbt zGcrP5>8JLJC(X^jVeiM?cF(%W2AQ;#BPABE*L}Uc%f5lde*V0L*rWqSq2A?aMu}^Sw(wtc zumk>!KtB8nHfUhd?ZHW6dovD;r!UZo5Tt9CNsVhtsRWaZ@pMh_kd(mM2rj4E%Q&_{zp63&n4m#IVb^8djKyPeHo zGy-@j*r`O&%5yI(KGClA)TL|$V? z#~>U6t#?CTQ>WL$|^k@ZYxg|fh=hwrKV)am0#2wIU|rkrL6KoU$WaIeS@On#=qPw>=K=XeAbq+m|IfuNFOLB| zrpSbZ1h%5Ly9YQZ*qNA-Kv|0N`|1fp$L-PPx%&I2pEC+JxTRkqFQajDe<*Ym*&4Fi zQUrswSEc%53Z}hwl1i?{_8EcW@gH;=K*ma57qH~+)}}+-54Ih2&=ns7t1D{DbL9pF z%ONdY-wMdZQxC_u-WvHQCeltiwlihwT8m?Qi=Cg=2b;6cm>C;0A&OAMzlkux6g{66 zH#Eo6Ze5&dj*{$G4!Ir9g&@XS^O<@t4ihn_W!3H!iF>CKpnio8-{IMoPYmo7KWCa5 zX)fy(l5X`Si;;lqwWGJU4P5Qkg{BSO9itulCm4b9a{%qGEj-4E9Xt?YgdO9oE;z2y zuQ(L(3UinshfU7G9d&hPO3vYt5$u{NY2{BZ>)WhKGd_Ll=?aHQz@dn?cJ(0oZJ6gW z60tZsIwD6|+u4b>w)U&oShUh{sm)RKz)sIx{(HZ`mo1Oan6LxVYv0$eDR*|-+S<$- zEPMo~`}FVM4_W_QLkt&+OpSsUKtE$5RsgyG*9X7f->TyJpa8$d<@Yz(qHb)C zeqTCC{y(2~;y{J#($dpBTW`|SV-TPeaENIdHe#fYap8CZCYF3XH=K=?WVK6SE z-UAqlM?Iynv`>0H-jT?%HvQd-JF=Bl9!z^=46eeZ142sk-o1r=oLLR--PD9PZ_pZx zEnx7#e~)MV)4gB=LuMUui_YQ+ruDhaeZw*WQ8&ywZiamcEz0V2dR5jZL_TA59q@P| zu-SV;%t_61m6x|ktl_AK%`_ga8OhpAV!ZYSlq%=LFLMjOJtKk=2jtSlUMDU9E5-q1 zI>pO$09U>DzHiYAHj1JAdoa+v6ZXU3C%cPaJO)WTJ1EJQ9UJrnW@g~whrTRmqumVp zWsZMw%rLKcQGYHZFDNKD1Tfw?1f%m1A0t247Zyftt*y!s%@|MMFus4S<3r z*WW)|HBu$t_`tK;dH}in62uOjERxR3o*PX#EQ`77so%GLmPN?C+`9yF`_Ww(iOjp@ zAdY@c7h(?PU@6r3dQk)OpCcd*{DG52pl?@dW_Y;uVlP)_>0oQ7@4k-Cc8%-S^6u{L z5%0a6ZOBml82{#hS|QzMvufqv)z#Lf_`oRlL)Ze(ft0XY-MhQHBh`1Y>DsWM#!Ybn zTg4wlt$Z=H<^5JeMJ0A}Wx<_|!~S@0^>lyKv~P4|EeVV zO0=;o7l-Z0DR{6kzOZj>z)!uCfzjM<836V#D0g6uM9 zs~Sf)J^pS3iu)8K;z!^!35tn~)agEGcaTi=eECumd411*r{^c<)z#H4b;Q1vR(iU+ zz_TmgcK~yZ$-C%uufKFf!#4lU;!`k_93qFZr?N7x&-U!C^r&RaUhyip)o!Q&JGisd zMfn8<|6`AU)3s@KCMHw*dwZYC1DF2qTfg4F26$9r(uM4*>gu1s1=ce_QqIoKU?;6# zw~p_Dy7bKVb>DaY0k*>0YvmaJACu1i01P6r?$&tV=JELxCLDNOVr6G*J9j_OHkGfH z67GF655@IjPR#aEyO&||$vFMollc{oIQK1Ie*Co+@T%?4PFYWY?ev4{>gu1_nj823 zdbRrJ`hCB&4lI@jW|WV|CKQ)F`03_%|Dm`3-XrNLDIco#&f5>XAYpINiub$U+x?rB zz3$_#I^df5KfoLDD)+{nPfttx1iB%B+2+wk<2&U~E&sFC9Rr>y*9K}T^tvg_3kx5v z2i7hR{sRkB_PP(r57zB|_b50hr~(*~%a<+t1X_LaW3sw`*ZI?be<#-f6FD%!0-LnW z^CsW^e*Dr6_ItpzWjF8n@ojQ(z;TL4Uwc5-{Q;@~C57{|Oiw>=>U&&SzC{ewDHCXM oT3}KL>Ypfxuye}LnsNL;|B~Bg`=1A`_hA46Pgg&ebxsLQ08C@AEC2ui literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/html/_images/week34_86_2.png b/doc/LectureNotes/_build/html/_images/week34_86_2.png new file mode 100644 index 0000000000000000000000000000000000000000..b12c8193f02443c48634c61a03b5e8970c2033c5 GIT binary patch literal 66200 zcmcHgWmJ@H+Xf5|q9~%I(w)+xbSgD8($do1-AeaJcT0(Yba#w&gLHRy!+UUF*Kn6Nwqg2)AdAY48}0>8;LIPeAk zaM_D|wpXw+w0F|6HGoL#*jt-h*_)f_k~$jL+L>5czGq@%VqqXPwzs#o<7Q^I`0pE- ztZa>#|LPnyfDb{o7E`r@K+tp^9te2?xh4<<2t-`?qoQ;2&a9KOMn}!X;l88EShnV= zWsS;=h51Y#(#Q+;cr8I1fF4n{lO{BIe-vuyoGaT17Gxn5i^a z2&|vOyP=T<`o)zd`~z)1PL+LA`>A~U2?=lS>fGw%gS@-B?W zAJHFv$bJPHCW8{-}iicM*sin57-BZIEe}>*akHr9}1Nt z;&DZUgm@w;&tzX;r3G@e%nSD--%&0}A8=sUEw-44Z;maui^(wOkI^a8y*90&8`Q)= zdWK==-4qm$btB%Mv|vTPhJ+ol@;`Tht}Gl$@$Y{GKVyi8|IZESh){C<_fB8`fBy#- z={#=-cV_F7a??D$TU#ykjn==%$J@``YI`-LPFpp7Q@5<~Q7v0b9lN%euJlVxG+i6e zwpK3wpNr2Bq~!8s*KL0j%j6M`_nz8zIjo8}&RxSU?A%dVnVp5D#OT&@y2{Gn`mFH3 z1CR6~V&!py_WhL?#Q9|C9>rSre`b(1h>}K3QAOntE^wEcEt$5A=5;~BZZ-EOns(%3 z*UD=koXC&&bW~un&WQuYf92(Vxk@J?5zK<0<-8iRT7Noj%YX0wS~LqIuBT(BX1&aK z2y@Qe4zEDjLrPtp9*x%;2jX>qjfMLH1L3ax-*F!FYGvx0tei9`9C} z9XQXhvW{X95lPF)Fg4xo?hmZ|&zuONmF^%Y<|$??=mZWn?yO+_pwHL-F2+Ct0fXu5>BX9JETtzhA%MKNtCT8`avC$0(>=p+fS7 z_64&-o?F9QnZZW}wMBZcTm0fFRF)PLd=(b9TRj^&pe?}2*r1}&J-U^uTHf+yA?@Bh zkS?DQa}Z;FzzE&&(8hLBL)Qm0z*u95iMnzx;RW%XF_AH_M;Y zMlJ}qujmB!zckOidQ(n4fefE4Gr}ouyn0u4rw((`@B0Y4^<;LQgMd_dOAt_dU-E<~*GIj{eGJbuKv8)m-mQJ0G;v zT~5#6wUgW(;4CdKBgFH$F}XH`yX`mlOih_m%Ovllq0#+Y;$Fd~{VO&cc>Ad#_85k} zK1I+%E)SbuU$P=$s0fc~|EWt?Uee*Dgp%#a55?uN7v>@LMN(jmLKLqIgGlk>PTu1W zSuh%dKNiLgEC_q5vW9m1R}`ZEQwMb*QB)ib0rT>hgca<;E8YY?Of8!z|E)Kz5UL)` zQz(D)oZ6$W5_+UH3!?V^?U3`Go&PPdNTjNR1lvR-B!U0) zF{2u{Y`dq^+*bK4tC}HXwfh7!PbvP->m%}Ysj*Z0G1KBTDwud2oL~8l82``9!^%}* zIX{b|r;!tmgCuZYHxpl`{f%m&l{{ zKs&IE!qZEYFz*g+cKn%Uj5z!4+n?pbgRAX(o!u#`z`(%fWt%KzGh6{ra;F&^r}R1} zK_Vieo1-`ZgZ;VLo166}!@HX&-}WN^StNs!;b+f(1*m-8K_I2GVK~IYNte%mG*fmr zR4m@FqV3TE%4XQeX1it8vNdP*aX(LjfS0FNqs5@x=~&$7vNift&nuD4tSp3w9e%UQ ze9vqgMVK}8XE=MnrjL!h=ASkFj7lcyiP@p8e%F%^`ZJ#}%W!-xyb*7cSLzQt2^c>A z{z_k5kNLH@=`QhlzbWmb4-$+0#^9a%cKX?@)BIb9EqkgKgm?ijg#D&_$L1@_e-{=> zRThTX;kFnt55#{%L#hTvF2B`nOrnXJ3ytptz0Lonf~9OKFP?bh8+v+zLGf_q0zG)R zV6qI2YlD`F*WIN`jKKa}Lr8Ag9Zn=AS*qH?)nnXJ)a?UeYeTR$kXAnOS-l1733r z2}WA_Hy1(fCqr%zm6_P{;$vqhUXLU*<8lX1oSeLTufY8!F2rsvc>v~h^A?hwO$Fu` zcHH}(xzu*`srhuJhzl%Rk|ejZG})kMZi(dHIWjgD8y_G4>eA|8QIUSyT;V{!L z3MogzTKg#Z$?OHfUV_&JRhZk!5LC6smg>oqC-oOgA;Kaer&l$gSyj}YA3v!Y-&x37 z8YrnFG8!I>l1HGcc&-}Bvn^*IG>BjAJ}4q3Ml@Y8QG3)~ASNZXQZb>mc)6bT@X%TL zBgcEOAdAOg$qQBsJ1uQ~Z!#Cw*S~*_nHrt>>6PnSEUISv#|x7tE6iw`u4hTW=(=oG zOfcz={gfRv9dkdMTpyiKS1nx;EmvbxEBv-y-1H4}cbpl!0S`(ls*EA$G^Z}A;~~?r z_&#b|yJK1={y!^SQ6^Sa>m#x%M}H`qsOY+%HD`*1xSUO9=Z(>=didShoy>1++&{_q zUkHUryG99%nrgFJC>?v0J_e8A-=QwQ+8E(|?|wnmbqPu@-0OD2tHgd&9;`iAD4Pn*j?myWzRL zTXWFWB_c;s`qqRaSBP&-OiCR1=RfM{@e~%G^*h!b;Z~_*0(SjG2OYd&9)EcWNdU8A?f%q?|nSdBnIgQS>}Nts!IV zr?^;PW=RQA)pfgeXjBEsc2$#yFO=jqDsltT4yVy!71enw9GRxw3@Qq2WK%gZzOxfa z%k$C&>{ykR)t9Zyho7{xK8lJ`L!_mp-EU9hG(1i#zza(a=W{s~6)Atj_Q!%Lpxppt z=o_pqI7)WM)w!h^OkO+fI!gY?F-#vbrKP7gI@z#;SaTAxva-H`VcdYbZJ)1NQ2C!` ztX|!j%+SNC{429ThXXre&3S z&F!R78mTxjwQOqlm0;j~e zCpa2X@{tKub84v0A^Z<8n5p5!785JVbxZ(smh3pKez2&0+7^j+UH19pmel^?OSo(* zUP9`#LyzR?q+BM@F;cJw`JY5e#IDYvMV$=&)@gG-n>0$`b7O~)cwUgfc#obHH=N;P z(a7wKnSCxSyd7)c-|Q2Uu_{zQXH4@rL$0V`1IW5sg9#MO>gDh{To&C|kn6MEMS!#5 zmQ@%S7+rr$)~(l5&fjbsZEbHm-yAl{X=(Wn8UINC@(gUWHqbVj)djB(;rF5n3KUFn zQ8vre2hE5AEnfn^wE3U%Uhl|*Zt)KEkh_!ICa}{5RO=jH4h^w9Odx}*tsM+=7!NN& zD#?+_s_B**z&$?{oD%EBrLt?u*D_skEHuv=$o{*#?0PFcvkxj>`q}c(uG3{=dI*%! z;>P!}G`MPjj`(n|Y3@hJrWN?&M@S=l=QCN9d<_`kfUKSP|6dMRWgS*ic}2t;?U6Sl z^QS%@_jmFvfj|CnIUX6bLlDrpUMAFwLZ?45^h~l{Ee_NGNia_@C-LSvN_4 zp8tX-!9eTD!p|XKLncFnJ5kAK*KH#f2>%xrp+B1GHv|Zg!lgc`o*=VKyjwj`A{*Iy zz#H~oWJUJJMPv7KaO|Q@1tdL0sw3Hp$h3l~nGZ$gzZ}k&l-gzIbFemiH}B~OGcj_P z!Iyb#7yxJa(R^~UKH|LNkalghzfozlA{|I;EEc;+)s_!$fC&MH#UXP`s@j^;L83{y_h#p*~8N-)6y?`O4Hv zxRC0H1SoDZfOH6-G&MUwx9tar;CEy%??JQX%~@5`Iv|e%t7f=ia-hJZ51rfw*e|wr z2#eyoY-X=t&A9*t1L4oG92`!SRyJ0e1$7g(#~&G?Io3I(OD3kN+d4biYJ5HOZi#!8 zu4G3E-S<63&%3agw2`6ep6T@AFa6q~E%C&E!~M`BRxTzpMJCdY?kmLCU8huLF~J?mq)!fY{~Rv9Xf&mWWT4#@#|UEh(-eqIt0z}5`U5}8#c;i+nu8-){Wj>KcHCoF z0c=@sD3j_JC@ivkr_8##y5h0}Uk(kKVA<#w8NAi)pw z=a_d-0T?HUO48DB>?f>bdUteG@p%6J_8CA&_j+eLF05>9hs}siPY$90TzyOvx3X3j zeCqDFVm75oBvJP<%Sh{tVa=eAQqXNFvM@`UqT1j>S8MW``rW&No(atSu!YF_Z8G@u{0f$EBMmY8;p%3GH@<$Fkt4MRSf?cQu$AA}Hx*;1uOlf*<}CwBH@gdh z5;(hZ_c@&Qo)}lmd1bQbb=qzWejDMZ;pb1a8%RNJri#-*YAzdKqGDkQnk-wVg(a|= zp{9A=<;z~vgI)U*>{^9Bc7CQyZ&7H}aFLuzd<093r0qIJ7L0#)TS|)WX#l+I2yy?> z!QmqFTjQ6#eTkFWnmCWir2ZzN$$PB`tvZ@_OX)`C0y>t#2-E`q0p-PC`DfSH?4FI? z>*6xZzSL%1StZr3Jhdh9eL6OMrGZ*Jo8P07p44>bCK>Om$GYv4GszYGvD*s)B?0Is(Q83p-1%|&d zVOsUcl%=&~+8yF2H+iG7qAh6Ho1nsy80SIEUX`Mht&Q;8{k{&D)s^~3C{GA6iZ^k9BDmTFu)b`yAVDIn(}N*yO}lw?G|2es@P}X| zmDXHYvPEpl*DCpG*g#Fl%-uGnoHk=TtZnIx1?b({FBQlp3Oy zbuOtr6PdK`)X=?mbqm3*h`6!8;nNq+y$I6=V$%xFwOUO zE9|VbBP!OfT+>#3XA>VpY@IL1V#MPrR)%AX%qEVGnr}w(D|-8rTl~1nShj9>Tnm-V zYV<~>c^j5RP}yPz7b_3-7ms;)@3bPr44ckGo>@3n`G7WNt6TQ=f>s>I9U43HSyK58 za=0jtkZnsCdCTV0fBJf>H$#tk)t;`wq0wh^2Z_4puSbTd-wEtL!X#fp92aO1h4blI zt_HQPa}CSO6qJS0R0<5VVeL_|kduA6#T0}(xG~D^I|m`TQ4H0-Su;q3yWow;)P5{P z*U(qZEIQF7w8)0!u|>o`o&7pa+IWBaQA%Yy`wQOED!bIKt5&O+1k?xg|A(A;lCNQ@fPwnZ8@^WE(f%$jt3m4HYWentN2;Uc7C zw1-V+7c4%Ri1jz62L5$#IweaEq=duUTRuNn31A51dsTV1R8^1g)ERkP%Z-F{ZCvg; zLZ$GrPp+j{{}Non$Bm5bSrNVyB=z|$?-9OLfCq38DeJPZ1dKlDE6{VxI^tu6Z_R4(EpcEoNI#= zTF}A&&wvEhU>ngFpcWfyj`7+Vv7-XwIw~B)AKkr%(f8iRK$MSmWd|wM{ZDq z_lBDE_N0m=>wJ;+<4yR`u2JBC681Wct_rJ+iDHU%j(MM`NH}s?Z|MsIkT-$z;?1>Y zZpl;7(`evu>BPw=lg+1w)6r3L47$|o6mC|tO~p#3V=z<0iM;vCpimg zG@bLsU-GO?5jZJmU!~ge@ZY-YD&(ABUq2x3|1(ryLb>8r3^iAg_~|vJlL|CVCd^KH zd$?gPlj|6GNF#Nq9C2V%+HrDYRU>!l)i|u2lCjZ^skGo)wWN0q^^WjpWtJA=LkcC@ z`Y5p9lNo3fr`@kcSUrHUBFm(CQziVI9NDGT9WF`{)Lv7qyHGCmW3pdV^pO4Qc4}D~ z{^c7_kAZK%jNzw7?fl1fVvN_7iKJqmFkg>q)Ba>-r8vb#_wxO_9BZ4s`riejj=xLd zbDiq*lH+A$liDM7CLZ5TWxjtcb%@PCH~rJa6)wrjm58vPQ{n`xiSfRu=WkLj%ZeFW zoj=m-K;xu}8!l4LPe|nchGMs-%T9gyYd@n|n`PaY{y~&eJ@bsqitLPvcQE2?Yh)ceP9IZM*v0$qDQ4IF1eev&#@tTJnf%_?(P{;*n~DC8%G;Jx$Bg!TT{0 zf(O(*QS$wZnaNy6x5OE(BrNMYc^Q=(0)JBO!#gi9C|9+@=t7|ew}n;WGR87rP<1Fo zLv=TY*W+3A$T{Ym&0xZa3xInhuOu{|ySPT- z{^^ERL^{B&98H$6^|cQZ89g<{2fpJp(9&nPms~+(BE(BzTip&E6}C3H^ZT}!I0*C1 zo9YF4i>}K;#vJ=Rv}TMZD~9FDxCqxCdEID07a)Vi{LcY>7EmzTM>4&LHx?;*7xKHNyVLu}wcyI0c`A))^*}q)O0q0v z0sM$(C}ODgB}0CKu%-G%N(?WgZF%MN)(Y{!WK%`hm?A#TL+h&>^+h5>kGt)Kq;iaB zc6k^}*h*!@1la!`S6Xu16mhHCe`bCqxT}|>vaQY!?d^RKHI%d<11y3I&YKuww;q8P zwKsB3{A{Kl`#*-uT!oLT9#T%dNSJCkKt zg~}ma(Y-=wx8CLD_Uz`L2LP7M&f6f}9=AYPko9tnt}vT;qtHi3*L?|GnGb+w0A1`} z7dUm%F+?r8`K#mV`}zR31L&b7YI%2ePV*MG9>s|Wl?TRm=Jg7P1ZeX<^EYefFNnAq zzIZVMzFI8mhLuKm7x{8b-PPPQ-ki0B^j^201%}#VqSnB|=Yi=6OL0Oge7w+*2?JP^ zlS5D9VSs|wZOZ^wH2*dEv%IgLH_p!o=ftc9Ig2S$@@S2PW?N%i`L21Zv_e7-wmYu* z(0I#D9>=ZpQ_!#%%wJ*1@LCd#FHIbBCP4;8DD{ZTA6LMzlrpS@Be8rRba8->x+R$# zfJnbOfA1NreY2EXf7ET8Ba;%YRlE)tXoR}lSa}}v{QOB5Go6s2+8tfZW~O|+E^tp& zsN8dVKHudqF)QZg#`ECMg0&31sE610y_%!|U6YKvi=s)pfg_X9JoC2B0mQ}KM*ro! zqC2ck{WO=;XkdM`EN(vV){4jGbgt5fBjl0DCS>3gr^A>oZcO4c-@9E zsGmJIoA7L#sqvr}aBg{mlQH2|Ov!uPE1BzY>t@~`bGo`INX8(Ni_JJ{XZrU74 zU!J?moxQB==hK0ZS72zw^94+*u3xza%jDH@IuWzOQpOC-A|fsa<~Uf4PeR$lp4m_M zz90Q!wX#XXEH|r#;(T6GW6*w@9L38`dJEpt+rtk_z#I9jT@l3G(ZpfeRZe?N|0s^SHIYR&%4x)#oZSPv+DJ^Z+kDE#3(9DL0+gC8` z#~(TQU@MCOTYc`X7xalKj1qOqg$BxN_DtOglF|LR&BRE|F%lh0K1>~Zjt3rNqLivj zezQH*R=v9~qzONp#k78OF{2vTaSmGR$A~i4dlZDUD{L%aW=RV`KY)zEWa2Ta~-tg^bD&X}=^NXzDP(c6xSt2#OjFkVO2( zXMZtVhEAat3>OJ@jFty!Bsw7=V;p|l{u1~pUr~eOgpQE-DX2;|l9s8nr_Tk5&nb?) z7gj!Rq|7>8JKGFwbMo|XCQ(FHTkQ~05=iH|ze>A6nngKE2cJUli!NdVy- zRcy^@#RN;Xn)BRUloA#1dsjG6f1|y;3r$FX`Ww4)vpodOd0cHI=meJT3;fubb&Qy~ z0~Nme`udN6cy;&mh)GK$re|j-08i-X_$k612KHBAfrX0hkI5E{xh~=Ge%lKQZCm`1 zHS|Vd87xHi#V>joB(I8>#3Wr{XDm4p8vGl&!!LBYA96vT5znqlf1Cj4Ndn1ts>Q z2%m=CeF01mlaa!XTj|qN96msZ*G%xUnfJNK6eqxHh@H?1V=MU=}i zlYU3=giT#TmYh|6NoaC1uVH`k%MZ%R*bxyCJ5v>iUm4Zehek&Sx6cD%vgs9LJc@TK zQLr+T5u|y~XFziMbu>;Fi{04VR4Fh)(Sr9X4n9CV1E1Bi#6a%H8_r{!VL1}b{!!Uy zp=Yv)GFqwrXy*N*K3Iu>H#`&{s+NZq^AtY&dK6j43BSUh)&%QfpH^VKGD1>p_w1H$ zSgpT|M|*J~>0TtmAnpjSO;vLDyj-o8-9omzCiu^176(?`e0|ke+56wkC2H!3vrn3x zD7f6t_trLZ)6#!i6HHEWQBqSM7JKpeh>j?UZ(Lvw8xp&2#Qn&2&@8+L_u{-;5hpG# zZm`r&Yd%w*3Q{vn^#_>POd9PE(SYYDRdS!+`U1JjdP4J@cEG~id?BvshE}aGd}G7- z?gHNQ5Pt~=(E?I#Zhx~P)!xC3w#$j8I~lp5kDXxz0}l}bOiaxF3p*4K8b?H&n_Y9= zqJHIp?#O{jdG)_pe0+2<^hlbme}3p7Zb?{qmDUk4FuJFAaK34n#$$cQv>3-Qqd4?}7s0^ODJA?9F`J4c7df6?NRs zmJ?oYOu1h;Ph@EFtUNa~kU_H^0PNMkactK>Y10(Q!>%)Nx~z6`{TkFcp(sCX4~($v zh{D`}ejb;#UAjEHh8f45MdC`E94=<-&XA1ODQJxe%qVqj-Tu9^Shl;L?uDh_@$2h; zYA5qNRNuj%l*Vf%fDGYPd|9s?#9$JuG|T(cYB=~K>8+=@Dn@1tpnTKaGjUghPeYk> zLS6K~x{V9ys~TFhAt)1R^)1Fy@a?@9p|$~+l=CATEm(hDO0R!hKD1)x=^CdF`{nLfR)NBYyWU@>hvHG6h<_Jy19@TWau50oj=)OWZt@M_Ns$J2GahE<+u-Fo-b^ zEHyl=rmfGw{qqfNK%#^zujS{~FAsQqvLpdvE8m*@Vajx`?f&wiLi)vt+jc499$PF1 zSaWuQ0e>ozR$A*`ILyvsK*)irO*#Ixv$hjRD%*gELRm{c90C^*i&EIoB%i8s0)JiBmTd+}@M=aiZLsD_frprF7zl|O| zBpxRiNNt-wAZ^(Cf$Bm0$YQA~UV6JvauY4My+?d*1lPCjNWXj`Nl2JZ{5;n#uf#Xk zvHq*%(Owxln*}+?i#qBzSSg)RoK&!)`_>`ZfRk(XWeaQIfS)RsQOuaHXac;?DLIZSrAGCA&=B-GTefs1WA9*E!aHkRG(y8V zOUzm@ex4>28yh<^IeDf)ju29z=BGwrBjmr6}0K5!jcXSjeE z^X()3T6#cc*XFWy57S+l(Gy64gbG%E$U2OJ&QATazWv3$J2FL zNapzy-UkYjpn{(MKqKoyE$L*Q^q^w4TS;}JD zl``mX9U+{J>p~Rq5lw_F+tCg9x6fZaf)|gCfzS~OHp9YV0LfZrh^v!`cJsnXy#t+k;^wRd2;9=wg&jAG(_V{5&d`bVHR*N# z5)}-+gMl%H46j1&YX7*ZrtF!!AzvzuPjBCU<$JD4&|P)7(y0=?FQIIX-^9Ocf8;#cAfeP_Urwb20cepu3&;ZG z)MWFy`xF;xApHO(cJWa+?sCy}Ol-MR4IUC2{Q#s7hlczm$8X9fG7dGfc8 zI&8uj#8W3LE$9b$k8f1Z6`m1l!;?t~3vy;#wK*7-iX>zu)BeaB&B&M)XtrzMUY6^# zkzf}-`rsElEvh{bT|U>Bvmu*d8yaPhFlI1+*#9l2+j#wVd&|uGGedF*-R0@KL1U|> z<6*0dB3;BiO9kT!x|i#QQfo`~HI$;%CzlSguk2pFE-?9d*MAL(NgVawe%pz5*tL#+ z1r-!jEGAyB+B5|;92dB30mQC(vI9|+c;e4f3*C8%yNpmMsl@sBDm-tu?^||bFK+km zAE(b)m3+40PNAg9U2>G{P4nWKuC&-ZjNdD!?}0Z}q9?lJ*?C(l0_BRPlE0Xevkeo; zZaL5lV1M$p*4`(oWt-Mh7o|G{58lcxwVHh>8|;Ps#McG>hf2&7pnPP}aV{|(J&MeL zSBYLIK~R2+>cBGfxrnwp8Ku9~@muEf@3rtTCU+IRT_ z#a^syY77GQRGc5=&t7k2p|m%d#UDsU9)o0*oT_Tq{QaF12#nV`x%EVb>Sdn1< zEGW2GCh-~o+~zPq;07&;!%Dw9+T_7BHUe3^RM)LE$jzQp)5qcoyUSITGWW{@$b+j4 z0iTHos=cU7-)fJx!E)!z(Q#$F9!4i?kR86fwL>-n#q6NEL)WV1e8-Y!V8=bkway7G zFm!`;kI+NmlNBLtG8Z6sK$L3PvpLXQMH4G+m#mAk0eC`$>p;QUHMzci{iWmhuH4kx zdg{PUJxZ%7zQ@~;_wA+D2alL~6WV~3;jUgooC^iAfWkxG)B z-?{7cA?UrZni@BRIx2|d{(=~!&-bb)IIX`3RL!1I#4^+*fFxTu7+#LvM7F0{B~C)_ zf(z9tedVJ?xecBK26SxVTyYq9%&7I=R?d`Y%zK_GK>CMZzBruWICCVw{f&`^&Pf*& zEFx6YylC?~-;ls^vaQLfvDIh&Ch(?u=7SUSMmeXtizA#KSQSs}sWRhU)cKp3U?I+w zeWMj6>5F-C%q3uiZPq<;2$X`qe2&0X)$3DcA2Zhz?EVb+E>umi+eEv#!G`3e0e#H0 zO@3J*i|pP2jP&sbkUg=mu(;X3zhjQ1w3&67IIRZH_wTP(`FQZphZS#MmaZIJn#C22 zi47kL#nV-MyC|Q;#S%T)JdI^^^OF@g_3ZjjWOI(@_Av9_idLqZ9hP}!5hf`e`S zUtmx%`NT|53QP8#m$R8aog^?S-Dz>X@X(Dk3zAI>k=T(VPky2rtLnL+dB#o%A(bcS z0vL{F4jl*RE_dn5`876wD>!FR9Y=RtXlp>{lDt_-uImcM7;6dtX;F1B8XlKd#Q{Oj z2wH$-}+CLSk`FUb5hsgvSb&MU`&RFhCpVcSYUpMlQH(Rn#keyXeu>oZbTx)z#}ax znZT+(21Fbnnj>TGR1dNS7hShHRenRmtWh!ZMfmiPnB6Y-)V333Hqtl-iJRlttl}5k zpz2F^1Ac|5h7C@CZ*D-7zEFh|W14dI;@%KP6mbeZ@J#Q1$eC4JjK1WpTQ)f3AZ}f& zEGs{FLX7W)mfukQh*k%A2dY8KBSqIaO~D*zL!Kn4X` zb5`G2cxbQVHdCd#6H6y+Brw$@h?4So2j-N;`&v|D=9TqYahKjRy&6GlQ~WF494?(Y zXGOuar4YITM1FyS2^mRAfyMH&_u0xMULEr=zY||AYc71_d{wAh8|p^oa+yRSO($=H zZ@G!snNalpJ^J^Os#kJ$3sk+AAwwc39vRthe|ud=kjKnW}33R6tXOb`o3OZ-VNi-lJ;9R^?S{DQ7>>66l<1-0t~(H*BDikcx@ zurb*(OzoI*fLazSW&o`gHf&6ojr*8Y~aP8Z9%(M!Q zMyGFd3BpbUx-n8%svCT!A8Yzv+C}}fbhu&a%pD<`bhT4+%4kn%?}PaRay8i&&P(Wf zgboxKx`k?DZ;elq)A&M;-{l+wovV@e=w)9&Hsow{A3kOOovLIAm)l4Z95(nKU4VY7!&ykdxLRNBh9|$yT@2Py zd>rFi8p<54z&Wfg>XqI*h{G9DDueqs_Vwn3shtSkdY6R;V&wjl@~#=4HTH-FUs7P% zCjY|Y<=$0cAhvyt?C(6xW?BRbSHw&ziZzJLN|6j4gfo;(k z+2rekbkRkH;=GmbRSJxL(NNrui$koM_0*^TIna|?UjEpJ%o}LvUKeN6Ru>0tPC#s5 z1r2IBBfztZ0>=9g{sHj->-$>)!{HzKM(1&4CqX01l9GK+_N=?RLq+4ln1Cwd9_x?d zx7iHvumh9a=Za!q&x}8hmzvIoFi7=SE@-%^6?t@W%f-A$p}e+!gaat@YiDvC>HOjd zAzuhfu&=bEvJ$+^Jw#QF(}<#4fYo0&Prvtsdv3?_Df6yN3g78P>9*-Y_7}BpED0L}fjNV)mtjrayB4p-Cqv2A@M^=S1l={6h>dl^ zGoUcv==pA%&8S+aGpo`DUT$atR06Rsl_0eAHN?-gsG%)ZrLdNnUGuG3T3u>GH&j_aip zBc_+AY^b^;YZ>DSGWV1urnaodJK=G(Rt1kfWDQ#{ADAU1bI%puqF=!ZFVVr-4p9(D z;XDzqA0PeDkhNWJ>$p~l7#*n=4lNB`47D7 zBs}iH%$}3q0xoR=StdI4FW4iJvk~XfySU_f0bhU!z#}uqdBLP!;7AMW3zS&~BA#nS zQ#d-|e@tPxHYX9(jjE5$tfY!?3oIK^tfY4-#v{J`eF*%Ty<9v$f5+DQ!B;gk^5sp$ zduzP&d?Sp`Kl*;B?!si=aSbC%sVj$X>0Mm2sMc-H3&4q&BW&GJ0$1gIJMO3W!OouN1%GGe2I6od znnAb3p{dsSQRw*;t{?vVj{?EiL>J#p+vZ@q!%&E;}X=a6xv9D@mM)q@MJqd3J;(`^t++M7*u2gwr zQ9}DZP!h|o`zg6SY>Phs8r_59HZ z`?Y(R$sMqPEAT-Ri%0-*77_rkUfkBlwAKmpO==yZ&B>hh{S$pn)LiZv`Se3uO@>aOJOo2nz z(O=^cO3Zi6@AaE|B)T>tmZAmM>E_36bWC zt1hpBQ+3nubngD%TRio!8reH5f+}PzwI*``%j8>N6e(pSAqkei9bytg59buB6E%kc z-`>a;R-kkWxI5f5O9n@379%s-91!za>AyTNm8Jw4vB1EtW7Q_#g|enabdU@ZPw+1z zy(b6ocVc}41J1ShnV_KRA<5dkOMk??2;>JE_g~YKQg)~KU}b{QBNz#i6-Wz=pCR6b zcTSh6EKw{6=ylg@g+kjcEO{Pw@NJmpHs!xzj-btY9l=!=XX-WKk+W}f0 zqW9xVoHWUQrh-p@abA40bvyh{&&6z4YpZ_r*ChWoRk=qrdx_$!D2c55l}6|xg)AJst`)?o~w8JPew~~`p z3inVG!nDa>*O9_?8CRf{Sv3bD*Iyu*qq#{=yw2bFaY(7JwG@?aFKBHHH0QtLxyH7v zh;t4|ph91U=BflPYs|Y$O`F`fzuwAwvt-owk!Yxiz!*v5{8S6Vfu`NN5^YmHW|2{2 z$QTLFnaLd&Mo)4jX2}W4CG_J`t)EOplLO~TZH*QYHT9oA$f`r9d+j-y6~RW1jlbNP zGs{Ky@a<5Z9O?@5X#6P^zKwuw81LY-N%hu-8js{-Ma&T|hnGhm52vnz&L2}LPP?qi zd(!PWPpQJm_mIfBI3&RZyMNsFd+b&Pr^Ogpq$2S_(kVeKp2AJk9MBXC5Frs(7@H_p z8>hGM&oc+vr%a3(JwdI2o=$;EzCuN$?x4OUn)R8mMFw`OSMZv~!}1EbLs-+|8=gmB zshPrm&=-;K%_QWpE;_L+f_&#UpJH$^RIf24^1Wr$BMeVftf1_q`dQ$eRzt}l z`?|3?BL@>A6(N4(-g?`XQZb_&y+#u+z5NAa?aD^!0}jbiO)n?=|l`QQ}jqHAFw!!i{it@oItqz5L)U(+^1LY9;0K6Th;vY zgsW^I?^mt$**{%hfVZnO6!@y3f-RO#oK?%>By34rF8L#^2zSJR5K{4vBK;@CP6P|H zS>&|q5+LR6F{IQ21G^-K&+7Ns5+fdQVaSpZ)*j8IG57WaHoRtz%knV|du-s%$8sFY z{X@>>gNXPdO<`e28PeKK1tvE>X{CZXO}=wFC&?d%+MVgqjY>y8a~(<3W5M8{ZuW}` ziNdyMC&q{p%~cNr)wSg|?FUgg+v1nYtE%v7)LYe1^XDxhN1OnT1ZDMIHz?$n#~lg+=VMD17ONq^De?D;-slRHk;P-?oc)wZ`oPOo~w zArOEKvkNWh2VY}ZY4p3vCM7iSOlaaI{*Mww*2^_AFC3kOZ2F3VIAD ze$|TMWTVqxjXG(Uo8(?R4Ze$aHZgpmVwME4ORL@ZsxfLXm%&qE;q%=`t&@byypaKb zj{?oGID!X7PuL`E=IsrBXQHftlFR|oKU*3Nt(0MCs5)RRao?I}%XKD~BL^#vJ0$S0 zcRJ_hJ}f?p5(nU{k)gj8hg$kGZzFAsgc*yRks3gd%6WYd`qhafpbkFMYCQ5Rt)Y># zKP54@Ff6#~{x!7bTwfgv0E+=HKYm6{mr#bDa*H|N(}pUW2?ruQMOQv#94+Ccj$=7! zvS))tx_D%Wau1T4X%jeHBLU#DG$3#%$XURrEZ8yQeD)!_|Buk}J{2Qf_Qd{BZbNgp zdH{l4ewnIr1I$NyE?<&Pk%6%OK@hk)lJg@4zKtHOmVeJ$ibck_@E2nHHt}4>)L`at zn*jU}7lZvI2_2NLcz6n`pRE&cxlXYD9SaLE7IVKt01&n2!3Sdj>!xHn&d2W=D_%V?`vAbL1@5{xxF#`v2nD?}DG(Of_ZjANz#&3WwxzvC zQdVq-PEJ)=-Agg^vxH10+S?=TSC!iPIGNvsU^&BLcub14eBfiJ*9G$1fzqimQb(+- z-hm^_`X-AZuQ=Q|ZbL`lM+d%cZVkC&+RM_hRQUAD4&ihz7Uevboom9ms6w$+ zM%#j?C%v8&ILZ?PNJzCQ;Mx5^@tF>#Nv~?&gK!W!mvT zEo~7;YP61^y)`=yDFgw;M6>cy@oe|yZZDtRDU4Vy?yIL|w(T-z^E@E@m)fJu0_osx zd9A#UaR^seKNxu*hd>8&&vs1+YqOzlE58@xaWruTmmc?r<0t-XY@JgZ z&f)`d%^}_~=u@>fwn=3v@G1dGX(O&v$M9|D&lZdyjwqDz8V2oCG# zq~wO^0P=N|Prn64S%=nD8c%w33rmctdh?eN!3mtDD5oJd?u#^}+%Iz4jsu3fd(nsa{Dk_?vhNv(>@0{VwYh}oVq0i*0{NnGRt%bw`+ zp|ux^d4d4nmQSvHD{HdBCu88ro|(#GnrVWD!3O(aQazHQ^Ru&pAjFp2Sw%JPs#|lI zqjh&k(XK}|DJ`?MKn84adv+1U^;qg_@tRpjlrW6@j{XS*IJb)_TnJWw-P!OJ!Zwx=!fErMWMxr2hE z{H`65n$-t{ShWC>^2Y!afC8YdS{^r<^`mNtb_nK4xvH!Kv*eUfVeRCiA21sPRnJgp z|K<;C#}vDta~euYk;FU{4?!@G+FTcD=<3#L;K>YVvkSi*Ewl@m3KKbx1nanK``J_~ zX_0Fb-?{F6mJ`9V`#Fp}{lL?8n72G^>bjgUIhkEK2pn&~-+xpSDEB=4KdS(qRK#Xf zu_iiZ?p*k&8t+oDSNciZX$e1m{6Xv`dLuYLV6~%s zu_#@~>8bza<*mo-HMZ@^RN(%J=odU?)s$=bOPCqrkfn}-!augDgv@G-Mp1Mu83aBu z)qBMR(j3GcmB$4Y(DeOfKhGIgXiN!_XPH(Pd43O`!J<^g$v zr?GTBR97vqB3kLPSUP0+i84Lr0*eg-d6%5j{YMwbe@eKYfg9FDdUDX^*)_LTyPUt?<64?xpGmjGLrCG&veAHs*nyFf4s(?M0olyN3isl|5KN zbIq(|XVgL*L(pvwr_$HEV(+v!iYfN*#>KX^gJN<*6hlfPglezs>vrB+5inoLacXpo ztxBkznc&uh)~`n0@;NWyf>@X|OOQZTYm_T<*VQV|gPj+xJ=0-M@fmJd6aD)SoEYGH z&Yk2)ZHa9Owcad>Zfpc&0$lZR*>3P@-J~PY?|FH)6lxKJ8W4Mj_u&T!-7al(l|Sv( z-M3Dq_jpm3$JlOR!k4Bi8KK66bL|(B%`KWp??L3x9(&mz8v0~e$wCkG|Izb~LG|G$ zy}EZ74?jZczHrdA`iGM4Z$F4x)mNaRYrDI?yvX3^p3}`1Z?`+=vlGc4Z}#oikdix} zo=LbkH)0joexGXn6plB@?%0y8gOO9CZPA(7)u4+B?OmO0owQB46Qw!%dOR*IrQDLv z*(hxMY@;L9QY`zIt;t7TrT*avnPD46H)W?kl;lbnWE%HE-CiYk5`F&H&tpZ`dj>4M zSi{)TZha4%t;jZnfDK(Vk%r-o?ZfBFn+J7#hLMB+gahF7PEEGjynqMUJw@~O z{cC#kS1D79efYOXL!!q|*|pI6*la>tI%GF`@sRusFtMvq{7WMJxA5A<^S zR}ddA&{ga&OJi$)Cb)O|tOI{*&*bn6IGzqXB>eZ0$_;Hh)5L(lOd3>!JH-uNXkGcS za6WzfZP${dte#HEvRn)|$cisOP1MZx#MfjE1H*Sdflq$20j^R`e(f7Ty`=52%x#|C%S)~x5jr;#>UE;3s8+S+Ghh7p8k`!V4cR2A~LZhdEuNcBHamM=J&r z)frWpWoBL6;&qDyh!7AV1&EW!OGra=%7I{7a-^Of<)IPgNsqog#7(m_6;-khqMFg& zZ$gH*i$_^?C(fJNpui%lL;u)RdmY1%eKF)zA`%3X-8=uE=2cYc8%K9}Ka5yZk;<^6Jo8Zui000Xv zZZejhyJDN?mUvE5b?%2Nlx)0~BnSvH6ReegFCL=Z)E|;<9y6nCYqhVcNDKR>(1F0i zq~KA!5E6tfDtk1k(`WwO$q*hE$YK>$1r_$%`0!;vnJOI)Y_qwC5uN_}reS?tzx zesl%`eX}n&A5CDzL+hW;SB7{>oYXP+Mg{Xh3(m~*ViNLW-@tcqHveo^3pEHB#2jIX zSdhLGRSj}R1jHi00;ply|q20K^pfZR-5KdD5YF8D;!n=v2){`HWVKFd@U zqS^A&9>2bIq%B=67c*{QGo%4|kd2DSrHih`3H<9X$}<*zeSZ~?{%Q-dubzQ#E^m)g zg_@Vl+}gOpUm||D#@M^nE{GCjaW}aw$r8Dri;E?804o<4dp^e4}%xZ}ab- z6ao|4xxMPMnW(qX2|FENpv#S3#KWU?kCdqu%no{nZznI!?GvM`{o>pJ?x<6QA0Ot1 zo_WCerP}x8c<$#f(gnx_rug7VH2Ci|ktSbkljRh}++>7W)fKo_y1uQ^HR^BU@}%dG zvnp1&tKG0wj4p<2G(~36l_R%2@pxM$I%v;dvIY`kJf51aVhB_1mgI~t{!zPNV1n~& zSA=dL26Y7}fT@_w98h;Zv0Pu-tlX2&gjj=g1XwU7-8|C0LMX7QvxB(76IZl_fA968 z^Oe1Y5@k`sauI)lAl>InkQ~^hQp^%wK2c-RJKC?*`d-)TI zakcC-J4`uKPIvC=Ud?*LKjR_goY5UZ=VNcR=2SH*))?3Q%Kd#|rS;sLS~j|dEl{+8 z*kkD>k%kK<7Sz%q;1U(+2$v(GQ}TuTqZT7Ob+!`{LBe}2_qDK;b3iPm>vvB zNc0IkwXHC=M1+1kDm?E>gXnQo@T=e-sUa<~g>ku6UksQS{*@^gt zk9e^Q0*NHZi44>`TcPkOxgAOKv))xsFKJwr;^(i-S$qa-y9r1xApRSyldo7Pp@z_9T z`|ij+PLW)13yCdPjW|{Mlmr0?LWX(_LKKQ(+M6gEjukB$2x1N4WiOu|D&^v=U$Ek= zO*UDM9UO4LrzAVm0Fyj^X&Y0k0I1l}C^Un*ft~TjuT+{ z0a#xq)HrkuEtVyACd&4wK?mhW5ev?kh)uTnCTKQA6&|R!fnfozTK5Id)=dW3{P%B{ z4V2WUJKKQ;%mTa4yE1o$%i}r__Ju|@W84GLfU8(htvX@II~y*xsCqlPD^fb|4QLBc zB4J((t4S!INN$%JWSj>Sb7@h}rj4NVo3vf;qd%;qyp+^eWd>6AYll!r?JJ;df~d`2 z8fQ4-WE^Ka3JF__(+hVSEvX}^j!mPqJQMrSQRyfAb}T2m=t(D+FqV|#M;&*=m&FMQ z#hgr7$Xh8=Yx=QJf5d^8Ia+?K06ds~=cTwWltn!U>@KgDqf7CGxr!ZNljm}0T?yc@ za2`Q^Ez>GPPKBXUObePL^UZgSN@w&Iy{@a@a;D0FW6dB_3bT}+0@P7RRkU5R5&G|F zT-^=%CjderCKHM8t`3dTA$7XWC`r8Qh566CSX)nO zIgN^y#S-n2Ae5-jjNXu}F>ZapTe0@I_>YP`RhL5OTI0s&vR*-qb8M9LK7d!g}zX;8wXhx^Ftg zC9pWJ^RP|+e8W_82y>`TZ>f*KyWh(6FlW8=z0R5MMwz1|V?pQaDsHm%95pxMi=-gY zEJ5`6ARY!(bv+6fUBr67sI(bv&IunE*iabQS%RTO;>M>(X55y)&2Z*c%Rx9nijr5H zy-XQq1*3nldr3H=JCC)TF(MtBO`PJzRTZy}+PD&yW19VXAi-wl!5Z zHSAnz<93P2`QzA1rZw0KHI5IeL>U`EP!gcrOM&=E+s3&bdB>CyC!$n!1l$z{H4 zEw?Bt=SJ+`la_LGpLLMvHZ^cB2wg#RM$U)UDmU-N6I(rG-b5>2`kNhq$lPcn*jgAI z1V<6=*F>;Ci-T?_paRy0lRwEq5=5K`VOjW5di|&Kz2Sy zJe#lc_}q&ODsleVP7dJf1K}*Ga{Ah06Ff@r%DoYWXz>uPp4~#ts5^m7?bEL~U!))? zKgQmXHOPnko#Ez!Fzeg0#t(DP;i@mAC?q;6wYi#+?OMUCAdxb)#u~LdtK$A_3oxND zk%60X1)L!|Zpks!RRRc2@^c8T&P4&*wY9)BJMwMQZVhdjC5JL6;D@&e`6oC(;K z5M52JLofVhT7fXnkB>RS(^oEjTZ}u=bdNT)mk-;_*0u)-cgy=n1#TZcE75f)_FHty zR^o*AHxx^QuL8X4v#v@IOmJeGqbx&fDOTf}6wD6e+UZIwPQn_XpH_;XIsvw05NpKA zJ9VSw)73f|DTqtr;Q8IOuL&C;2MhVE|Eh{=yh3zpG*zfXK*(IQE$&mUN`obO#@nO8Cl4kV6bW-t@^?HYWgNP61^slyEhco3&Zt0(duhH77e*xm9w zV1yISUp&tm$6p=c*&+dC(X)kuC4|X1TUJ)rf?Nf7C9idd>Jhd3H-~S0xaaJ{jfwP< zqy)q@v^X4)%e(nc>#G2JhbLvKIM2=m8r~s>{+{>JEftOQ)rFHSr)nu0tx|{*?2uae z{8{G|y!YP~HMp%RcW~t!FTPj~z_#;=oS7DKj2Es0$^I}`T%{qqeYH!Lxon>k+PkI* zVRI)*B#;mZ#l6>&x1h*$+l_QeNNjX8Vj|M?d~RK_b#+E05yMq`FlEsj#99{49Wu^~ zi3&CjL;VW*thV?#3pACs6Rp=v?$LXpjle;bHij0>76S@f{^p2LhkhE3g{Ieawkk`Y z+BYzOMMv^4`RKuiE+Nl#F>Qe$i5%Z@rSJs>r*}OK)U}`sbvZ!K9_{{+%7U)jmYXKI zN8AwXJIs@H5;z>5ESaTl$N)Y@tPE0#3z0{Df+eDwH>SOaIETqE<>gI)fJH>1Ew9Ci z_N6*T2grJ4W#)cEtx2Bh!XJ%Z`GCsr|4P9{qKSZNpbDWOpryGo{BRZxj_<--^mdKq zFXt6fx||dC5onz5aJ{TcK`P6a#u|7vMcwt7h7j-tL?3MmNC{<4IZ{p@NEq3L_v*x6 z)4eZ-*ZmVB%A!<~p5>4{mi^DXMy3~el9AmGOreeCv8|o#w#sxnYv2i3@=w=i?7l0Y-Q-2atyk%20nFhv1cMXEAc-Em?}vByX8wN$*|PW>$!bD9IlN*4PN;* zi3qic*ydkFb*6HYS3!Y5T0;J<@lpsf{uB)RW%5geI$pO$wUFrzvGWmc4NYt%90Gr% zyS@|)=(+Tr(QUpCnN4UjnUWes;N*<=MoRJY2GTb70Oade1A}E2Z^SO5?!%a{MT^4Y zM!|6dhL%RmawjNRtXrRp0;0~eOD5l)#9*WmZ zK$_JSNHV@2S*rczONdXQ^aA*lfRAih#*?q3>8Ft12hm3y4@h=jJ>i@*w1c%y%XDzpBJ=md;$A{{wH-!qb)kiQIt zmPKj0%PDw=qwAzvDo)flQH=BO#8GQ7Ev=vTP==fV*uU1swtN>Z@5HNE4*+rQ7Ki#1 z3MT7oxuZ#f=-OYh1Ic;+PeVS$hVfWp!lyr2`XOlzyOEMbBn1Q@dR$bbh9@TW!{fXC z-kU6FC+TI+*B^&7(R3EU98-b_hF?Hk=|V6<`Yl>#>nyq^S8?-poAKLv^F{>YqDL{C zMjL!dqCAOwnz3KO>pn4rFi%CGieAfr-c4s{lYkL7b9S^%7sybpN7S7>RN8cxAx&@0 zU)}$~N2Q?OE&#aT9nQ`6@1!_fQoDVq8p1dpX~k%`hGH!C2O5Dd z22{P+QTlen-c3J0sD zC2E?#E?gD9r5gChy1B=Lz=4U>+$EBnzD$w=NvN9TLb2d)XBdjyo2qh7y=QoCji zOA6_WXab7Fs~{vg0sX7Hs-Hg&PDcLTbL{cBy_2$?rg~woms!Ks+IT9n@4{-%8MjtM ztFm7`^6eqdX1?Kanps;ly0Wu8JiBi=dVW*{MYtycf1-OU5^Ud`a{twunb_-_3jiIa zymW=(i~y(PYxo9EmGycTqomE={=9U-^GCrkz;*4w$gNS-HZ5wKmHAl#g@Zue;3*v= z(--l)1*_vYh=2jLi5|6xfM=uF`_(kLulxs<41uq;InKLg&rHGJkuj zUT^SuPj}@cDusF16#>yU{B~D5UDz6)hw?=E;*)C#3F;6YA$~-+^Ma8d5qksoL)^AgJ#y&LFaq&Sj~+6^mQD~pRs^Z1 zJMrw{MA1#ujY8@L*n{*wU(cEPCv&}wzL{t-clWQ+8eb|!7wc{4Nj8;xErXf)wvOPy zK#fVUD68zR>868P9}{u~&Ffi=cbnj89TvaUm_KEirv`Dy%G{4fIJVE@>C>5+?{dAp zm!iqv-4H=QPVI_eQo$_aZbg*na+h^Hikra(ic&Y^-(~!1f2@-TEr`DWspMzs3U&lh zSxd2>1gei&{=!#vNV|3DXzql#Fn;^O-ENQs?uY6Itp4(#EHhkGctsn#B}V~y#FyI( z0R(<5V<1Lm$kNosn#KF%eOBPcyX{A>ftisxe*V8~)^H4gkR1L=VRrr?}s1*7% z=8h-6Y}tn5U|J^<7qnVeVGXgRk6EjUme>L$QCY-dVg})hoU7a&MjrBO#3w~6{Zgb0 zoh#leJF1T5RD^Z&*6di;v)I6P(d6mi^a#w&FY5!7<#WMvG{L0^Zr&d@0i0C4S;Fx$ zSNxP12dn;E`wv7(tnlV$eB0}1ed|&e96BA1p0oY^x$0$J7mE!pmD5>hfB0|o-rDb2 zrx>_MXR`-*C<`C8-KglL+B+L`$?A5RrbH+L%ea z9MNdORMr>34nLrN^*%m82;Z^o^^AgH*4}OwTVDv7`N_7X-*tjuZsqgSfaNcSj+`^| zR^M?IkL8(e5iYpKa+zLBP-!I@f85RtX?vfwHSWIrmw(eBA-UTV-IwGe9k)A!t_0zz zF(1_1(6a8}92Ap)rDqS!YyggHG~V%D!~Uk(hOqx#$T=A#-JW62xKJqweJ zyE5MO{Jf`M5+{`Pes4TdnDC*;s%BKD|IEK}$lsgG%Rf%5FHZLR%Rf+VK~OhOGKnYu zvFnYt`r0?MrX-gH?GT-@QazG4VkFe$wJ3NsPz14)eNlbx=&tYVj~t&fY{4UMj@0+} zEe~}y(@G3uPsKRx_-)_!3vMnU3B_8(f`{rMig{CAm91k>w-(qneMytmm7P2h5}Das zBQ}zH0cMY*tnmShdf79lQt&Mv-yHX4dT`)zP6}dc=HTQIU~~Z7f`MlHSN(UkQox`f1O1+F+@T~vFzOH z(xW7RJvyK3!~ror8DY5NlcW<6Z3;X4tsU$%2$YQJHfcIAc&SQp_n#|YNNe6a#g%}F z`4j**deTm2+0(;VT7ROPet-aLtiGgHX$amK{*Q~`?af{qY@oq-uLD}{(>cphDor@N zV;lzwHLl&uB`Lkum506J_g_Hz$Kl&$X?uMlzXWkNG0_ttb0W~qU>M?cmUa;ANYNqM zmd63uUn%gsFDqGQ3@LgpoC0zp8>>0Px?Gp9S#-$}fEniPnW|9lN3;AxVG0`AJ#enF zkIBN$`oId?uXo@%1E<$-(h}-^20Wz#N2Y8>|J{@k;YG=ht`b>71))k2aYqQjim6s@ zqSCE)&sJcI*wV90ff52~ zOl828b>-oRQV%lSQ^i96VO(K4%<9MvIZ`Le$<(Np^CxZ^TKF#mL@)_kohLO(f7)w#H{DE2%U?sRS6 zKb4Wz4BhPwQA`Mj!#Br@_WFiyIJHHfNMM-ab$&Y~gX`W-t# zm0TNwNjz3yA0&m@U$-0C8#st?&1PyjBjP?Qs-I&XO!F38@^IG0IPX_VTZntkRc%jh zo@|-}^Pqsg(M{#gN}{?yYzbEE*im_U>T&~E(ZiGI6v0o6cs?W(v?h*VAV(7kK|>Eu zUx}9{JxQW#z3wF3iSWv~638_&MUec$2wPGHWrb|jIT#d01tcX>J<}d zZtfO%MGw$R;|B}#(6tvI*y}#t?XYAHG_J1|rFRa++@rA^*GQ|`uX&SFvM2brdr_`} zwg_Z9jV)*LDm$}=KC_%Rrg6iG;gN!>893-&vE{J_oCE!WkXn$+T5W=mNI+u`f{{FA z8a_Fvj>09Q6-Iwh#U+pq&HzO?fT(-d&ZZmJab#?)?cx3wlhc7JFE5YlqyweNBiKsB z<_S#wc)T?>gTJ6Oron5KH0EQmlx#E2FZwIZZJRyxv=@|D$cyE2?5gx7_7x>6$8DnPn8lj6E)$i4=@Xs!fOYj zs?NmqMa`DkOTmbBq}8a6?~a7vUX7*nlT!uJL>|V#vs*lY)TD#JV#IsEU=UU9YxED+ zXSKgr7vp{oh_0~h+ImYS6Oxs3q7cRyTnB)w!1QcjJLqtbk|HO;MDa#foG>niCw|rc z=WUeTOQf2D==um$b=&11GZUwNV#%AQl#-I6k)GzBA!9?Mb(Oh#9tnPrqQ=Z78>=T~+RY|ggPz~ppw_q#TkI_<|9~#I>I4h%Zw-6+Tlb8N^IRIWRXDf! zL_`vC&Gs&ln}chwA|u|{dgpYO|H$s6jhYX)Zih`inG)SyVa67XZsP*DM#6UB;n(u( zL$n&ckxC1-ZXH^+{yWqf`b~3DL7BGo{jfPh?;t^}0sHn;F}wkrv^ zPJxqkzkpR{!tSc}7UeZK0TX*BDn*EVW}4rYZ)7(g=yG5+6u*LWy&)vu?yFd`SNqGx zfqubSdeoydbEvpBNrB9)vNl069>Hrml~&-N2Dzhx#%PHD}gr9Vg%llze~3@&nDo&wq3H%ecq@swr&C@f3IT~ zT_-h)m?OA$;Lf`KkgM5#{@lBp4DiHM%a1fHHICf5h^R1}B7ndF(iJUCMD%o{$5Bv_ z=f3xrWqAk%?32X)ydf+Gr>obCWE4%~S zt);(-R=2o%vF+WY>Wp-SkD}pllZ<>W27spT?nYvoWBs5JE@nsp^fF*02HFfW1m&H& z03rwh8ofEmaZBl*bZHC6sHynSyJ!v9Bm+eRTRHzA)zem-jTIl5=8VAI0DS0}jYxy;fOFF{4cL3Y{x7?*U>|BhZv z%FVBS>@tkl@|>h<%8&fggsFaJ$V6YFuO}$ybhv!sHjfLEH{pvX5>z~2$#CqUobIXp z`90~A;w~WyJnU}>A>*6rLbJNrrcFUvqrQ;6MX7Q`x^9Tebe|pNldu>)B0hX9bP90$ z&Zvt2;k$d}*vV&3(|OrT*3-(tRLY5FeOM`x#9Zv;^nFp)8&?mSmMc5CZ7Pjwb$%J6 z-Qi&6*?oAx&9eh3r;pU_>n2BgkX8x6-SaqvqyeIr<^QvwfZDVr;S41r_y@i^Y}$Ih zjm{4;NxZjFV6T*+7~UB$7&~@-&#zc5wJf$Xwq~XeSWsX!vW`*L1rRPEuKpb zBvEmV*hdt_B8f&f!sAo8L(4G7rApZH*8zE50?{PTE?=zOba1ZL7Cs zq-N5YBaxI$XRHo}{f*|#@YZf@W{?oTafX26vf<8?Cm|EvJ#7#TN;_7~#p=KqD5 zj=B{Q_=Ec_M)%x#Q3$^I6KxOJ)tWi!NsWA@fx)$|*8@k(ENLenJ>aBN4L8Y}Qlb(q zh#lt6Je*s8Muz>d=*M!74UNmo>|1;!z@Enrw!B0vXyNK^ zsCPH)g04oK*g;MS^{grFzNYNuer8UP$o<$7%ppp~X?^%jzxiD;YgqFI=-KfOQ!Wz23VY?>5OzZg}cnDWOOVey3e6 zq~&s;`Tev`EJz+?o>$ha+V$2i8_$phmjcB1lCS$u!7tm1D#rlLH4u7EhvmLtjoQyt z9~Vhtj9R3?)qIF~fOqV_8x)kM#=4Q8Cb4((b2-E|I9UGfaJZH#J$XeqV!$x5@5*XX zTdnEtmSKlUlzi(*KYq|UJa_Ur>D0=VQ-SNFIu?&I!gmgefP=>yTo>C%dE6?d*(i4> z$@o97r3}VK^u&3@orprAwweiJLukzYvq*lDSo;;1J42(2zFAjq^hHaS;zGbZDm=7b z@eu*!6%VFZKHTXD(~wTFYUDT{)`+Thq&*uuZo)TrGg4?pe4!jeZdxRLQM#*f+SDMg zTVDQ4LQ!03BzXYtua?^xxKA=)!Ypqwqwf|900BQ1H*&V_*PyU0 zRQ7JZmj^NCXW8PRR)an%C>!L))La(s0mvNYC<8oDB7u(ECQrkH_Q}>|kY1N)xXk2a zWl2m;vPHkCpc2FlxZ&JVd-DcpJh5StR(i@?iER{1H}%5g%Z0M7mEn0CS)CLj5^C~~ z2lObCk{>NoLf-a0oVAEfj_x3!fWM9JE}(rGtu96ricooOxwKq*jn-#rkeCF)PJ zI6<7M4|tv1B2$;ZP9Q--UTi6M0DN!b(iilQ%ew&BTXsCel%O;%pv_D-%lfMVB9%m& zYCl9#=O(Y*2;w}I{*x_#c@sz=h76O_+H>$J&2NIl!BP9fUDW8n&?Xx&TMo*P^dXBD z7tAm8ZQ++I~M#=k{~0*zFo~Y<}B#_&1}E3-O;kBX;g;I zV_zE%q>YG-d5!L_6+Ntn=02`)KbZ&**L%G19LdSPZbrRcg$$F0!`64;+5Gu1Gbl(`0lfqm=TT{{nDs&b>z0nUBoTWW??bfn!*ph;99L zD(p-A_G<}3iE~#gvTYM_FEXGsTmdWDpNBU2s~J&LC{xcS3oU69{}cc^((! zbJZ44zr-epURXZB2bX!W>Kp|ofM3poggONpE6g%+thom7?oq`Mzyk$j99E5gftfc( zZ*o#i7S$?%cfBNfzFBPObDi0J_k!t?<_&!w_3E+nAr~bifh=5Uew56~V^^tdvMM4gG$S zFm^FXRtkUAPzx#5JQb(2@#TKs#|DsJ9|UKh$02{|U;P2x*cGCjuPdDF zQCB>Keuyg|X{PY0+!X7K&XIUe`M_!n0|4lEp!L5^>W;}!G{ppyJ^;vhE!lH?={f&j zRhwb+WNpEjs^@ZWXm9nm587{2xXIvr_X=VIw_BB!Z8}tgY!GhjV#7-> z(W368mN>t-M!Zm92?k9yFsjSXt&MMgNVF*q<&|-I^=nz-E7;=EgEU_QpV!?QBP@wP zWSLqiB9gg$`3X#&ejF}{05IF?76X{9$r&w5G&QYyhk}>7Pj$`}aF#pq)R~zAL-)vR zX3+sKEx|sI@=J$!aZ zd7yQ)qJ4fccGQ#_hY)!;I#{8ozFb_mn#ZNw7Kxc7Zp@kR#|akVTk06qIC2jY<(D0b z79b9K(U^$O8s*97dhf7l51jraU0ijZywb0bYL+K|h{6uL0nMP0``g@yOGYP^CjNPf zqL>YAO9G3ofhU&({eV_8@r_R#u=f+uw*4O^0MEKnZd4o!XXjG@1s8=3DJrPG0jGW$ z$RivUhu~f2Nz*eJW59|a3gbAbY$6s;6(_TOIi?xxrx+=0-hdK?HZ_hr+%emwmP^8> zUscrj=VS>_X#t60z4ml$m5B7k@J*=I__UKLcsGTXO$gunv_9QVKl%_bRtEUUPud6n zk)KNbyt-RBhK8P%-F&7FRR8E(6!6RDn10+4;*X8iHdH9U9H$wPUFVO#UTLr1HE@K% zTv}fDfvYA)X*aVurLi9=&8;g8beoUhCO96c!lL!SojNGYhgB1H-*Sy>q{lVH>|GBL z#JCF-On}n=_efh8Z)jJ4*2uw!!(%>!fQvE;PLvdCQV}P4&_Rg@ ziO+--=;zp^vb1^6YfL+fNr1#!->Xn7o9C;J_Mc#^8=p2Hj?|SFJN?|Utj{kH*w~D8 zIKJmu3~eKHE}h%sM2S}+jt6!BT7+7TTd`!3zQCGvR|15-k7?mJaEM=otRsnbavtj{ z-35tP-ZHy))NSGZ=lW@U42!>Q1O$4lPi_K;_VLNI6(y~%s7&;qf22y)rn2!DO5Nho z3^8qmh*`VqD}p_0bP{ z@ajp|2g2T~-u7IkbZ!dOS#$54J#U22fwq&XAO>_?+h1@X(uJ|^dy|)wZk~2u>`MvT z!^zy~6IwKX>CbwGsV{l+gCOGE;-xkYOQTfD`rq2z!_o_3gM3b^Z=8D}l<(Xq{68RU zbZ^4&L*OoipC%#^P+?}XLXG_uu3{(J`PG{^L#Oh$_=q+FhsU3z8NAnu}=>N=EQ-7!|2AU z>{mJcx0nhPalFjG`iBSkb8sA97>E-PQ^W&(T^1T4LQfKy zx%JvdW#If5xQ!^$q`D03X;rvW>qo-@l7^u1CCN_uxhMP*)>J#L`v)qgJyW-#@GUfZ zt*o5JmjtrS)^7<&pYhomawSX>asY$Bhsbf26D zh0^O8X!rDyWH3G*%gKU{y=`CU=?;Q0k=EsTym%;gECB-Gk1721h=YEM0Zbgn_TG`-~?Nf z^1j+*c99KM7ZEp!x@^hx6Oak-lLGdwyFtKBtokQ1n*#2<9m^J4BVRbf2%CN;FxkeY#U^spk4&MPNlxG%$p<<-N`!afLscYc`L6Sll z*zKfA*otr^HE!2Fsa}AO_R5I6W*C1c*G;CAK5Xa$mPDy4X%_EyG;J!0jT64`yvH*) zb1JC=hj062k(=tb-Y)m5zRG^+=QFJAU~&B^Hu`#7s*?t+f1P*hMJHQzgfh6(tm+WF zLh2dri*>wIrPY-Td=a)XMFF%qozls%Urgi?jb(b1dxIPklp6^wew)=}7J30| zTyRllT!3MUfQ6at97`zO-}X4NjiLN+r{zQId#$nLm;HF}h7z!FQ1#jR`mLj4SIKDG z$H#0A`cuPL2)<&TvvV9u#3ms}HI$_KFLkJq5wmdy>^bRAdL+y@M6Bcy%GR#75eNug z1FAGRLnTAkM1Ca~JrO5!uwXcDA%8#b-DF!>i8x9~N9$Dw zD7-)r4mfzz2fBf0Y`oeEZ@KFm`^*maYpt>?O&zv)$sXrZqpLhx6INz$EEL_`{i-aVq0pqtc?=TGQ)#htQ6@&q{I2{(Nc-6r}+0LdzgIT_07O3kLm9rzp* z_AZ6Br)po@$ig4(gj}h=c?T1QA0BjoH=GI==Ku6bU@X*&nPaC|SHDZ;%O4FYtB|kQ zu#&z_jWu4Z{OT+Tun3up?zD4pfAH4cF|TR4aY;cqt4bOL=BtbTGlV&6Rw9qoFf>gY zUFrY>s0ys$C^gBXXoc?J8(|jJl9UogQb!5CSF1 zE}7{@r__ba^a67Rw}CWrjRr%|l@b3FR}!55*lsqUsG@q_^XhY$fd2Lat8ENzJ*JJ> zTUPALslO81#QIg;p+_6pU3SHbA_QY}r9zJXdz06Df!en7d57XUjpO}|B7@3G42&d^ z$de7C21}Hyn%eHS5}Ts6EcD@LuK1rm(>@0ACa$QiX)Q$F3zJU*bLv6ov!$`J?1JKx z`H7FpG1jaTjv-e#5^jCg%G5cY3~4M_4HAy0AvZgWV>0_*&}XZ?pkgr0x+`mWVehqW z)VDquuQ9qeBO0JqA=TEg4Y6qX*1>y_G#bkaJ3Ca+fu(ii=$k zzH_b^zNvbt8Rj1vno%^!llHC)v=VoN)8CXD)77#Ru%W_D_mZ#%ab?Nb*^DI#6_p#} zPC*%#JT%U$TqKm5FvUDcJ+Hd)1$+B3qKlMc2(&wy*%9ZbE235*6P174x?Noluz1-R zLpW)$zuLI%NMPfT-K=|$m)|ibG;6#8AgGk2#|QNhiJaFj659vCMgsPe&hUiw;Q|{D zh-0?*^$t)M+J1sl@IXK_=qP$1goGZ#I%sB#+oF5Y`=UG|9xUg4!_+nnJ7MF%WNW-$ z)&&eErC4~ZP>9aulxVe^WA#ee zVvK)2Q!yS%j1Zeu-7U`w5oAB=>eArWM8Tfd%}*p?#CA6^X(>+SykOhTuW-r9ot%E& z8VjrLq&(T$Aj`39yHxnalex7wjbGK+$dRg4f|g!XJg}`sEm~0EdA^?iasNbMV;iP2 zLZ|%sP$1D$P7p_ygQCt=E(1GAj&`5lM}uH7Qd;V}?9xCYQZ85}JAp;!7**DCL#JR~ zCF;M0qQJn331um_4Hhr2Nl*@xXNUx*T+5O)u#W(y1#9S;>oFu&5Ajf?s`kci%0eAv9SbwVGd~Ft;|`^5m2`kiXt0BB!bf z{B~Do%%MFB{k@&M+)9-*%y{nU8$E*zdZFt~%SEXtX3SOG$T;pMSn<4%3L`{p%&=m> zsv-`YRs9eA`@Z;5Ec^~{jW3r%my&RPr(CB}_19u!Hx!+oggn$zRgDC;S95%R=5C-^ zGbmm`fA@HC#K{x#cMki?vOuxv(%@W2?5K%VJaD%O4wC6$JQo841tFj9Btce<`X{x}^}c_#?xx~{5Z zNJu<&a!EIuRR+kx%i--t*WH0J@Q5$hJg%1i^!8LE1VPb@0(60cr?Sq(8sArwLO=7%YZqu6$v zQ02qL3EtpHh0R48FPg%&7vXE$UrQ4w5R=km%BJsfA(Ba)hpxt&e|UHgCr`%Y^H8M&(`YW1(HwIpUNOu?LSZBGV z{V2mIP;sl)BZ;?0_hhG!4z|12^Vkbd6&=$URX0O0hp$|p{g00_pdnb^PoLb@6XXMf z=C>*qhi=7PCqHoGn5vtYNIPV0AcqrOw^HSNYq@h2=oc#v`> zP*vtnWm{dXpEp11k$`{y_rJyFbB@`6-JDt2mg5PH7=Ji6EomtK)O?*@ofq0d-Tb68 z?ctWa;h%aXez**MKYF!&Kw12q0EOb@iRu9i>T|KokiC8tK}o}7Z(}36^U33>ZtbV_Uaja{_F||Ww8T>;(CEP!SnV!v?Ml5Y zF>Z7sJgFbg2dVbnMgOmMcD1Yc_?OIV>>By&=@C3saQ1@Vn_Ib6#-daO8O1WD&6cX2 zOy1h8B%!>1+PongmDZF4q}+J_nW$lA>NiF^0p;LpN0P8;FbNX`Xk9!=MA`H9F!-VdWo z+~di4XI~7?l8ubE#6P<+)y}3)n4=UF?ux2>$S zaW;2qY0CjME-HFHPoEC5GiqMD+V;18lNqV!97y9nul;RnCAwtmgQ8%Y)cfu*PZPa- z?{$NB9`6LbI^L0ET8ubl*EbS7t7tRf z_dFBjDn{LGqQA`#CTC{H(m|~iUUt19)6n$+jtp|8B8;yyO$ciBgU1=Nn5>sqA9s&M zm2GhDDyjOe9F3iq+{`r*dG5uJnUmArgn5+qie{2(Ml6QrL>E0$f6lzUHKj*0V(IjL zmPqBFgxlA+KXcZ>M3rT$&mm&3$E1ARjmpm$s6)cx=ICbZ?(M&@wtZfx&kJVZ!W%;q zcy;naIW)(S!KA|X!4QSZJJUo)9|L0}`X^yeO`dG$G`A^R@|gTm;O}zONoAMTOwtyH za4iOr!w5aF7=3XKJWo$+-fRjyWHX=pmEFfX+5-+}8M#Lu<6(0}Be;%TP-N9wfux`+Rq6 zKORg@Ce+^)WL3cqF+O46-MJrB`x?`t+xKWUBF;*Fs>&!b-|LWGkQ$w2W--)r_e4K# z`GB-0lsY!1Eg5T@O_sTLYs&0)hXolBix0@)hpOvt!c}#tysL)t^r}aM57+1(3_(y6 zcf~PLpI(x*Kix(B-zCr~*)k?SU-3C@8G~C;2jVH$hdubNyu&~OAX@Z-GnMM8&KpX_ z^%tWZSxM%=kl0Ek3o$d_6`QhaV;$w}ThAGL`ZSCnq3DkHO*pRxNBQ}0j+xh`F@i`u z&f6r7$PnIR;nnk>2}FW(Q90L+R29A>s8j24{_IR>E4+Eiz4jgZMpjyQXxxm+wd{pa zMJ8#em!BW!d9<5W8gE0qPh-s))1`6qIFs%F{6g`#Ah6G~f7q3?s%88#64f@Yu7H{( z6R_j(pPF_S&YW(>lC{y~I@V;^;GClB`O1Lgi9va7?Tz@hT3l70ST~H`ma#cPbn<+{ znbvo$EiAFhntwSKZuIGWIHV|u*J5B5-zPdS2$*lzbH0s*T)8fbpQLL~TkjW`i3i?rhMMN5DKm86HCe4@> z*Zt@cp~<~M#PF02#Tt4xWhTHsa6`gXS&AFS7&ZF1bKce|_Z5u%6ppjlh)qoN>z6$@Fky3uqz^)HL9r9bKv81E7{7XgiLZ%gF4Jx8IO@(j_Op2 zB6CykOBU%IK<%J>;Fr~eGQ=o==4BmuS@jg2KyhB;F-x%iizwFd8;!wCD4f;fK5|4y z26;lC-PE|Kr{OI!4R&Ml@%vxPJt&(yS$lJ)e=OU&9%imrIZaDCHcr{xbMSkl=?{*T zIj*8S1ULq9i?%!Ul8D7!e+_e&qx?6-Z}$iZ$NoL&>RDM=S;h3B*WYRcXkDVzSV7_K z61Mwko656W_z5Q=CvtPbg@1WBXB5O@I#9Q4L;MmMD$9P;i1VnB)*0Q%UzK*EeGO#Z z|2||A>gY-qiryT#QrK--EZ1yx9;P{_HT?%1{g`yotPu;kSq0#T{9RB5n>&H&Y>%^7 zH45>GEv%k;0N0ag%j#Pi)X7l+&l5R)q)N<)W4ghJ)_rc~Pmw5|E5`Te1^1emSB)#0$wdguXus8Jod{X2YqD z(2wquKU_0^HxkP_V=@Z-Wv)0Z2{#4PviT?8@@;#RV*4RVM4zt|@9^@l7erQ0(bGvnvdP%z;+lHGuVKdKC zK!4C#4llDl<@NE6?8VdcSStC>?)cx{zs%mudE!_wPF_#%c?3OI=3iyLj{f-1A5A2V zOJshPLBY!z_v)zNB*gjsADir+Rn&XaF%05H$YwU_?zx0YC6-RA zTSxVhZpS#OahX>hn>Ag?Yrl273^efuzt&{pV)J$~6$&zTqNcaG%i$gV?&xF=iZ}R@ z%LJ>Ma>%w%BZ9u(rT}WjVG@Ey3q~_{9fR}f?g?>w+tSa14+mv7aecn(7Y2XrwvZd> z)AyvfIqmMSG<3WX;nJXpo9)f6GyE8yVG;w*;K>3NqXK@Rzcmif5Aw$$%`N=vJe>By zwue^tN4C!-*4BWD*xD6}`S;>Dq2Z3CEBo#%mq~$3+5=|V&Hf?`m&eHO-J5&J;;eE% zsy2!yA`24roTzyw;fr=U4d&RyT~puh?5=aG#8w#gpf-X#I!4XdVg|O{&iklipAg-g^$qO&?z}{6gyHD4QJ>1UAsh8%j2n&OpA_URIk)K z$7zOZqYsw<5+6N(Me;S~FAV@aROyXdXij!T1QlykLY3jk9@UST>j)2lmt8dN(NmZE z7XVIRq9dD^oYJP}b;C-aa8@hp2p=>A9R`Y4wWL!7jXwZf(0K9f#58t;L{z9%Pe{Ps z<;b_X)@<=YK%Et%(i_^-(NucAe|m*C0J`CWg7)z&h$b$_VHBM0oZH-%=t{~DSO_h9 zI55R4Zf&pR(M2mcuCjkOLQfeV?PeIAjh)i?yfqCVf6q_UmNlX2`Ga9NcQduktlPt* zsBFWn$vH*5xYe3)yo`c6xnh5e3oH^-45OH_C7@XZ9-=0mhe_h|ChWb|k9R#5`*@u* zUuC)EnQnK7_ve^DMWdA;`=o-eSl^}Q^{SWYL4m7f$rpW|AE&L+_)Z}|q~3uG-foea zGsL3q$2i%uOh@GY5%cM3Ywpm2gZFBtL}WEgMgE4T!*C2e~-WtFK zB4z=kpA(NZ4IE4A>(CgjX~iA$r|an{1rV-ve+HJ3*?I&>*|g$c+e66kkPDt4Pclmj z;UG6g#*|n0g5PA?)FElHV@_AgX`vFX*}|S>?ruwItB^>gZ6eudB`ofxr6sj;BSi1# z`<0Dk%aXq~&vPT*LEpZ;@&r+d`+%R81|2hyE10D~vx$IkDz2M>YQoxrSFJZ~F;;P6 z>2x%YXsEf6xMO6&{w6Lh@Mr33>r`Lo8;m4yz~sjM!p zIcHtCne?<(HWw+ZxPMpfs)tG(T%3{DWh4oQe^li(8J~AaD6w?Je#iHt;Vp?c3e8J- z0X@;zsv8@R?C8V8c?B^WdCrBkRtkKZ4>egAcGX^sX5A$W%8%!`OxNnuFi3Nihzd1J z~3WQ)3xBvA7t) zZaKZ0WSZWFt*WKJ^faLR^adW4s5Oe1m*Iw0RAEyxA_Wr`YKi)D*4G8oU2bRh#S6bW z?vUm?sQ|h}+X-58er*k_=K=)dKuS)DflU@V4FTvPPUDY`h>m9sDd`hm0lmY@bcM?J z>O+EI_mV64s_Kx-y|;_2g@VRx#PTSr_<^HD*`Br=r?<8HLEfY10bXxB*%D^h4=+2^NNtFStd}AqhD}b+C0CTT;G!^w)cbAg{$hr<1wJZB<<;NbipigEzM$YiVqc z(^}u<9P^ExOm&xgcjb)AIK98FP~4mw=PfzYoB|$3@-r@9!iT3Jn{fr1?@8=4sE#VgN}EbL9+Ew>`eajmq$Ot4@crA zp2i1rM5zP!KMNq#PLFJjDBmaNaI*0$o^0g(zN}Imy}wk2K#c7W-hdR;|1s)SEkV;* z#LDwM*6$R0V#K=lSvNkM+vUa6u-skz+>ldn-J~w~nx)&_UOtrC{K%IQtx)j&fS8J+ zG1$;=zHRNV{v9qOTI;vf?liGhpf`fdYC6Jy^2f@HpND8xB{Vrq?tmi+xmO)MIFQ!l z<56y<^h1Jhf9gbheC91FWBVfSV%Xa_^22&UbkEzLa#b-!Lbk%%D!9M<=M_Bi90%RS zeU00{sU+l5Qjvd3hG-8Oy&z>&tjg6D-T`JKCe0dKHS}OY)%=ge>jn6-8AU{@VsVI< zJ`cLcDqkRw#))x^>l_JkC7u;~|KU>8kI~e+0zuxzP!PMJu(w(j6q4%&X7wP^_Kl46 zfRY)0z3hPouI|3s&51tE7kmso*c`3I*GL|#@W6FPI}9!$UFgq4MiCpGzl>L(6W{vT zXVksdamn?y)<(us!^xBEi%xvdsNIekKm}Ob)GzeZLJ-KeW!3rbU-_dyw)z8l&*>8~ z?tl?nPF5bFsnmjk!eZE%NA+MT4RGos+me<{I0#^FX{}^Jm3k|LDPdE39g#!n)IRsq z84V%Ceo%IRE^*Rgzr4z{87S4;uU;g@t2H5}IBC>;S;-G6Zi@x@v+x$uauUjBS3?Gq z{2)m}JlXIGEN)2I|1ajra!hirU;9!G;DFp@e>IlS0Y|2BFZ{i0`b6?}e)o9Q@XSlw zxRYAgf=18yNIah{e87?MCn@A^8k%7yd2D$+y0j)Cr}(SU^IEQg!QMQU(Y6$)VR{8G z=gtcdn_`se8C0zF{DNh(5Q9pJQfe@fYhSgoFYk=K|9o9D9g0A4+>9$!H2sMXOLkx` z)#Gu3qOoG3JfkhuyG;c%G)4uvSmu|CX!?fDV*Kt-z82w};SUby5gNU17cHkd4g29N@j z9z}Yi(UodqU9bk1c}@WVyf(yCs`eZ^1SuY8^SHTOTo&6xKBU;kb$pl?;~noJVirSb zNmM=14kZp+bjtbUY0#fHdRFU`xmaas%=hhp}LEmm;KdzN7S)T3G z$YmDU8f7ilnd{f?oXf!6lrKeF^{7M_S=+XRrl-ZfW5R-h0^F*qJS@C45w>jCZt4Lk zo8EwBPjNUNV_#HfO-t-htT*lpzpCYIdiWk>oH8HNcXu8V8WWRL^y55H&5X3& zba(rttARg3I5V>=xA&&{1I%=cI#8#^{fOEj<6h6+rCf7z*z&hK<&As6% z#fM~m>9@bASN@YkIB-eXTcY(?%cHCse)s(Kb#@yIvHeZ4E16q|H~{JwNpw5QkH6n! zkd{D0U~Yb&Z8jlc9G<*WNpymG*^Vi&9l}^*fcRT)0ibNq|c+@IEn-=?j7smRAlS~`^Hcy=LDPX{%t^n9k7V%%1$dRc@C?u?pvg^~gEY23n&P+!-t zv^RiHQ2MZSdU@8~x;h#_`zP-4(&Yd|1nmB>lX^|x7Rx3>M8|`1s8Tw+EEbHJbl6yh zWpM7#qtsGgR}CI;!~#~d%r)n%zP@5oR*KBSeSWRf)2l(8 zhI+wgEC|{4g~p{q%LvF<;vdVp&rM@EAE4Rg%l2!ybH*GANLb5~B`5Q*hh8L!*h&KV%RsXzjn!?``G*;gn z->n1SH&L3kKL~7r4`9pp-yD39Rn@i>^uF$v+>MN{?iv~}UG1(aw6PzTENRmYBYF`> z5={R{V{D#Eb}tT+PGm76H@Ude0t_7&~KyX&jMTDd(b9z$Ok%b z{0&9;=KF0UlB>RyfNG<6<*Ste`WI^{3HcBYI-D8a%pO}EUYj&3cYPC>X2&6!0+rV_ zTwu#q8Nik$R{0lr?9WWY{(i(|5T1lat6S>AQ04rYx{xpv;Nzw98Y2IG>t-(PA{K8; z>-x~02M}@7V?5yBp_p0jN!Zq)Zimm3-`^Et*vUaTU9~jTy5HadDeG`WA+-KCP`To( zWZz^dF6^90V~LrChmTfx7- zd3I#3#5~Fw@W@N6pe-=IOs(3>WGnec(RXZ`lq!0^L25#k!DCcuj}HMnNEwtuPCgDFuDZdEDD`d2f8vXATd ziY$!6z2)ikk!=(Ac>4BHWWjwiY)jq6(VKkUqQb-0C&r$rw{UPChJZ>A`RvftMP%C^xp~3VL6v;8M zPyX;zgIVhodF7c^3jz!0(J`|!K>wOd0Sl-Ke*O&~M$Ba|+RrcFe;O{+ww!lD6cO3w z^+X2^Mf)h*)wWZTBI2`{G+nV(D67?7-2^`rdQi~l-5u?&pabX|K&>=rzF<;bYec*< zz(+AjSSdhP({a0ci$IUqlF>{zQZN_SosDH8op zj;WX$p3US`m+*8i&Jm0^Fe}sbStGSW4HORY&2*e!2?{4~fy^W#WaUeTgYqn|%U_P3 zmCH^~a*^JtbzaXGj7sE__CZkQkD&8JvE<2@C259 zuJ4EjWp~6x3SUTCiNgq4an?Kc;O@%iH2Q!$kKkec{ARKwwQFq$FTwk&?rOi~v08VS*R!zqAQdH042 zO$j{CTAl@r)0lU+s|ss9?b~Z->JGr6rFI?W)c+J*zw*@0b9cgP%J3qj7?D(tma*0+ zrV=*AC)Omq`n*_yu3*E07+Lh9Gb^4%44FVdd%0V#v2J^3re}qxAhhZ7OF|i~grW~u z=MjUB&QGD(+#5rx+^uN;&I~p3$ZZ6|}>4kN#RezLHKw z9|n)R^X&P){Xdi%tv|`cksb^qsi(YDm)=o);SD<}SykB1pC9|($?2RfssOMr2Clu@ z@TM8(e@zk0`ps?+$G4+gk_n;IFRL1EWW^D~@PBl-Tv({f`6%AeB@>WIPysi(E4-^n zC7k$&-FkT4ZBR2A;paTNX>oaV`O$J{+eChMO?mDw3bRMJn-NW>tuNIVTdB>37(m@~ z9^@DHE1DsDm|h@PEo6Me&33p2bGu z3?p+{$An7>NbBra6sXT*4Yntp#a2Z#8X9x2phxZ8Q_za8H z9HsfjkB-5a)b)2Z@kp~Oq;)*|VMB9=f|hx9TAznl{}X(Y6iQM%cfIw0Ob;Xvvf>rS zlli7b1c%Mii3m_-bOmPCvT<(f&=MG881?0f{|@Iv{U)e;YRcc`lbJn=-;4n6frDqK z5=GX`{j8LUZBjD>P3`X+jG86Hx`YE6{ljBX1yL1sc+|r`t*gFm_56UqEn<+CkP(g% zS-;tLqI^++Aek`(ryVJ#l$PCrGnZ$bk3`&ZmP({oUg~~V@n=2G&mU1r>Xy8ah=C;Z z_{<0E1w53&^9mXlhM)hO0qOPAjLNJ_860HwB&R*@nnCoFug6v~6B~@S7&p)? zM&q$WbamnDGDzQK++ZtbWhn-O{hf9@B^_YFXh(N%2=4v^J;|wskTkS0@j-wahL+=3 zy$Yyt_`JNJpy{3jW)g!uLcwHb^rgD>qvi0FXET!Qv-@0qShY;C_-5KQnckekdvDm} zSN_WZhW8focG>#iy4)npz;~8$Yt}u}sq4uMe2;P@gB{jc*IMa*Q;vDl1Rak=bN+Ic zt+7p0G<}vN75rW%Z+SG}N7+3y{2Oq2x}#hst#8=}q2n)-R|iT(Zz4IR!D;>5B_M|` z%6DNy^16sV=Sp%6%Iz@ftNeTI8a-@Ts@>niA2k(`L4WF`VhXvqxv$&se!v7)cuDj% zHGpy=;GYY5c`@4O?!7A?;Y2%O$OPWcTEC!AV7&3!WU6;P!kg&_o`WYOTym#JcjPp_ zp!ks*%?w50Cc4X#m>oHynAf~JyS#F!WRE`VX=Ib4(@yy|xJZ-YLz>JBe>Q1JojIjB z@5`Wx7p*yyFMIpAI1)>*t}&5pXJmVs_&oxReHDCZ>+YoiSvHTJIX3I_b={!^Pi6r2fqKQ^02ffdee<+G*}Ln&FquT zYmYOZv-1bOH#|Y<%zEAv)zCZDyB>eMU$A{u7BSc(c;#sJ=FPjPe%4tBQq_g==#^&s zSKVtFjwjejYU~T#C(+nq7~RzzX42uo5|&cMlwp_(V&d)8tb_vUbEkcS-`Yn;{Kjtzjxeh>rU6Q2eNez<9RKaliqXhq>@|CZD+o z>12T=Bb?Kn=TvcXZ8QiO!y>L=lcPXmew_~5wWCv^lLj=Zk4oV9Lp<}sgsC{-(2Q^FOc-%j+!Mcj0?s4gCE08?$iwz{tWP$MKmDub}B+WxsHEvQ(t4 zH>X_agIdXV6PJSk2YnY!B)sD1C10q5&~SuGle6&)HQee~b3M|_>pwXKUzM$MXgV)< zdYdQr+)5Efp>x}ab$?EO&HK{--z;5V;DGN)h#c?i8bn5@ed)*)%x6C;XY1^QkMtO+ zyUimW-Pou(|8pQ$=a@x_hhlc~_^ek@{~EHImK0cf@@&?BP%zf@w@2VOl5wEy{+BWH zX$*w%VSPCeAx2Nisp1Qg_SYoma$iSW8dBNgBNV_jS?cFRMcEgI7yKved>q)eQ}0=l z2DzLIP^r;ogz(rUU)(KUtUd*M|>b$EA1T{9vkb$C0YmyGNf#n5e3 zY>uQ#r9a(>*Gs*(mO-;p=?~}ma57=|I&G+CNJy{j9-F88*ajqvcY{q$5Z_D`LJa!@ z;x`go%ZGoXS(9KNo>ubZ=ge+h2c?j^(QK1Y0@F^kF(TjqAFUCVJZ^wIHK}%T$&&zp z5?}~C%gfi&l2Bv~vL|HVy_jD#b+o5VvqA3bnTg!47et}9V7Qd0NLyytg}g}@lJ;-& z^p6tuk=t?8tJI}XAJYpNy5XqCT{obI63+`sTWY8S$)PerJyxxp16H7UrzM$Rz1bWd z<>sb%V)wWVi>LW19)sAG&dKQ~cENMUvzlpn*u!=WdaY+*r$Qa0$I(Xp{pEk+p4UvB zXEp2{Vy(IC?;&piCO-9#)`c&CVA)^ zOLR;Z`hRjW@P9@TbpJH-LB}#=ZmkcH6<2rr)OjXaL;Zq1&Ay1NCzNW|Bg-!cKZ8oipQh0v)kZje0UH^x!x zyiI0^)Qt6FDuwFEJx(@Uh)R1jd1)aK9Shj>S`R3L_BouK6G`O|wDeljwqLB;$?$MMW%aB61p}TRPg9f1lNi$b_^`@qQ4UcN{Ut4nqbL2p z;kxq%1`4)^(B4UxU8XzM=YF=6jO~LF*B9fY*KBO9NYN`z5I^r~b>2{JiI<-SQm5Ei zt$?Fmh9LN-kq$*c$x1e*SNTFVpM5p1=6sY}3+RIC;!fX0mWoE1FR!xyBbOY%g20q$ z`1dxP{N$kF+z9bXUDoSXD$V#P(qK#x^>uH}ZL?lIrq2ldpZo3y+Dv+eZ`Hr>?(XU( zXL{lar%vTY%5d0q?9b%>qUW>f{ZG0CQhlZ^!oBMkVAbrXm-P)XfoOlg_jaFo)hRJ+ z6U(tp6gjQD;{T}oMJ*f&BJc4YSdrh9?{vyoX3d4lP_H;?NFU74VaU1CI|W(%R(Nh{ zsalpNvENw&tghU8!D4XEmA&3su4LkxCGWP;#Rq=fc&e4FZ->m4ed7&%6BMOeb~V?4e_#Ba9kO34dk& z>n$-r_pvzA7n?70Xh0j)-*Wt7UEipYiw4QU!m8Pe1H8inxPzCT;e5nLjf4g(0<5<5I*=gAd!-$H$iV+FQf`YDoq$Vw#!%%BxFdmdT7sVAphrvoU>pXj82hTbBU__v_el zpl%RBeUykseL*12YkRq>zk(^%EJXpYaLEqjR-96~=1m-m?bIC$(C`6w=rrL2RI7AO zXSPbUIXHh^+Y2$nAt3B47zFl@EZ=(zFBXhq44JSATu=jY7z&~j%8LFRmr*yRL9JNU zB``cW!^u+YD2W(Bd=%IG1q0pJJ;N*&M_7J?IC8bt?BU<=SB@l_?zq3&L;n-wv2wQU zwkuqI-D{X5e>(LZR=4NmdyW5Za)7`*Fa(&uf6MN@I3%G~wFbFISb=T?m;-r*Ne&Z_ z=|eL*oiw8g;CIoIYL29J@>HJnZ1BM5+4h41MeXP1+j*Zo7SW|^>*zFv0UQRZod9y; z$}Qtj9L?8+B8DBmN`G3ioszIaC-YF@@H8Q|cyDpdyE{RAvAM@?wtpr)xHiK4Z%~_J z{-36ij{t;K#5Eq9Umd`?`NOm!|GWy&wWU;&#kP zuId&z@s3ZH8g4&GfvzgBM^RbrRCt?5T8_nYL?16Kv60ix4r=nJ+IuV>V7Q%pJm!4q zXg}W2*O0qyP=)3=Xe;$`x0!XyPvJ`=X-G1I_=2kV+CJ0&L&{8XVwIIo-Pc6?jZxTH z4%?+QwA=pW1{3lKxc@8<8OR=%4Z4FJFT&mb6t$z`H0)XXMiR(sk;3{}11_1v_bQ>} zefLuU&z)SO{sbNUx9bp-~2O@lgo?zGPXO|q`%^7%&wiA6MT9G#g` zKPehh`Jox!Z*Ron_HG!Yqvb;L7!TPa@7|6s`kVWn_$!e>t)QOc#9*w@6Gumt|J{q7 zdB@)vrM8UFdfa$Vr}5mqQpf@)9)l)OUnLIg*FPR5dP;smZ0nCSM?g4lUfp5YSzG`% z3i#_(A>%WUXAil@)Eas432b{9_{@tuC(!W!9tqx>5$cf5un4tT1# zSVIHO&@1-lhB50kS_0RHnH=SQ46IH6>2cOwMjC?Vnmm|qE_UPB1v%OD6&c3H9n_SV zuUEoJ-xGPAeVAQAYN>T4f5m7(uKoFi8Tdc7E&HXb%iu%kXxIk!_0HdO%u3!qy3wdF zkW^bG!8t|oeb2*IXkrr$$?LAo@IL09#ke^ub*-x#KR}LSj6J<5XcXDGNo#%ax7){o zQ8Wld+lNG=;K) zj5u-{H?)Z9&M%)CU{kp<(g-5}7KwsyBR+oSJTfLwKA!h%{RbWp2>zMS&6%}h<|;+T z#LL*g|ExV7sGU|{k6@hK2aH~&;B;-oJDX2DaRVX(%yz)Le9^Ph`*lR|fA{;N0VjsT zbs63BHirue%z5mx8~o$RR9izM)x(^RF^CJx;Ns(nIbn%V+eIFD{n%rL-T!J6oJ`f# zS~p#6#tx*qH%3u1IGJG)X-;8@6R@qtq($J1)G1vY-p3E_Ip&gxcEjVMD8;zjK zdfL*)cCQDZ>f&NP-!1WrTuY^xBwmN9@7z0gAP{bQ2<_HC7AsYOnFXC&lhKs2BSFf* zC@?YgQf1UOyJU-xt%58pM}0QSawDaw{E;D>KnfrD7sisb#Y-R8cD`A+K)2$05c*iR zVC*n(Z?yBkhy2{QwqqO@ru7*xf}=Iw!DwoqXtu_jLO@ z_Z3DV;w~}Mr{?GduW~3`Pjys?e9DtNX@4B4q_7&_%H_Y}#X+R1CjVM28`D?t@x@cx^mfR~tkC`h(9qz(I^1 z|3{KUX^;oWV>?C2w!PZ*d~BXeao2NWdJeZK2&)d>bp8d@B?!{D`3Xzk>jRe)0CiBF zY~#`8JPn07+>H~VA)kbGiHa+5xIW|(tu&F(Y{Q~Ldbd~xlX1frnwt!QR>R{+dY?pMP3lN#c{ws<8_-j#-jPq7 zCi4v2ul@K$6}MI*i~kAFCfDj{dM(%b-L$0X?zsJ1)3DUSJL-|vR7W4`frYb~Ya}}t z>cigV!@GS30Lfo9E%6+105U7SL{8kU7VVp`7uimPpxlrZPK?i=muab1BceG8(fCaY>f`QWT3H((x?nw(6)cQtGHV%d7U zoXLOm#^K&9OmsXhYS4|QFC~_6Z6Se_z2$wg@+4kuM4bohf-f>{WEFp0vv^4iImJBb zn7vR)6q=~A6@1;oNaQm?E7OD!souARq&SN9biD4w0WFd%@ow9gYxD{%-~Vtnsun4g*zV6 ziAlL0TF7qfwrm=hQ(!iUi+V!+s{lwefG%^$xp)~j8oTGp0ed*1d|1bA`DkcJ8BUaK zt~XpqDui{m9vsLe;E653_R+uRMBVoz^?OwDozyWiaZr4QE}P6sPQ;YF(C*xaTR0^2 z%4(`sLqpL5d@{S=&F7_R|G#Ye_GtikeF8)Q!Qjo|k|*2W(X&s(-da<@2%a(%l~Xmq z;@3N#q;NAmrOt3qT$2Do&E<$)t;xNL--iz)R{GXXx2mf3Op*4qiO?}Eua!GNhDt=Q z4DFS_R}2{Ggy0d4YKEt82az~cqeOlE4Ae#t2|u~u-~+bO2}_80ZY{dp!teGIf0ToFVyS}wcT1)Rjv{$@!O4smELa06I=*nd0EBpS0dzB z54+*AB%BswSvMDzlVG7*d@iu0Ea>Z!aBxF1mf1Ll9irZ~6{y1Dp|+%!aZhXGU9fZLv znqO$7mChO+r=P$bl6z)HJ-goiPs`;0u2HiZ3$P)pM>)dETA9wm{-X$j`DXS48O_I6QB&HWML> zf0>DabrlZcb{VeN8FWuUKF&`?_LsUrj<3vehN1DipSIxl?{7M#5?{XHaM&(Kq}bG( z_Y3+TS-cdg63dZ3xW*@AZpcUSgteZUVIgthvQ7o*0Txxh3@k+F%io@4!)D5`%%>aq- zV=k$Gj|+Tt<8a~}YZ2U>D$jR+VD8UTpdf2P1TS1OiO1UDdmQL0*#B10lEQ*|q|YsJ z@cM2ucSE$Se_1%i*9-Exu^oo3-i$7f+a^`lXW;_C!?a7C(I)#+OL_F%eo>=?toc3S z2e7}bwhs4$2m<~$AmUUWB1BusI!=4CCJoL$d1LH@>_lujR9DE9Jb1H2NUbnoz|A-; ziQ4tDiHp$J#GsKI{MmxHnGD`hexV`y=O*u0AV?r~x9c#AjT|cAS%$GeSc^JUp=nUF zfT|SeGRm}fV9Y|s);{8(J926k9ZigDbOeOZ=V7VIFHzRA1KQ}puq4aL+k>|zEJ=SK z{V68X@ZUR?G%~~g4ILJwkdAZQPE83LUM7Fo|7nPam!9izagJZ>VqD&-z)b?UQ^!k2 zxH^*URLi{?g0iiu4KvbaML|w_63LJtD=*zOu^7 z`LWTT0$vo0=}286P03z+#*^c`d+GXf12ZmYJ$|9|x4o#Vyq?UxRZKft!DDY%Kig&q ze-&Ybkde2eMK8T{0hZdit~05GhcjrOjr8!3_N++xP7b^tr$VB@X#Cij9|Gb<+LtjU zr7=I?h#FhrRn)q@EBx!E{aUetuL2yTk{H`~;e3QDQ#wj}sazt}#YF#Fd^#tsSXz+u zJiTjiPQ0CgIC_-Z2|Q1}y?UeoDiXb@_YAto4#b1pGr*gq^Hd1~|5_6Q@oTo=vmbet zpp`VQsc^M{bH&D?*!>3mI038o2 z=ZMp(-*zqD*ua}ZJNjx=T?M80TQ3slr!dqQuHWGsh5s4do}M&KtF8y? z8`Z{UON-(=2_qTb%jWd5PlB7ImaTh~n3Z|j@j)?6)c}3{JO?2AbPHar0AFc$P8#jYAlP+Z5UKfouxw+lzsXHP^@j{K`lWgj0pJK3fLivX+ESg*mjBb2G!(oy(L2i}l4> z-$#h``~$C^z)rVzt}r0c_0_%;2S(`-NBykMGH|;pbnFVe2P01~IL6{+ES-)f2;8LL z&AmYX3;_LI+=QwpXFon%1KLBe{s|>xklW#{>M&=8dh2j>xPl5e$3JTwhSQ?SLfr1s zCcycCgZ%1*Sseow-CQC;p)lSTBL4WUs}?Wv2zpj`G(tqIcsE6Ftm)S-^T+dgJ&(V7b>E-+zV7Q>=bZO-&ikB;&qevlbQ`{& zQe@E=(-g)@gA1HF@zQgFPz^*<=Dil;;x%NbJ(F0vhBhEpmzH@{iRl{0BMs@nMJirGH=P}UDLy6GYpsY|2LqB=n|joiZTKbx(p;Q3j-U-@!P zRu1+U^XW*USy=8TaW)TEu?zxqG_0t8RV<`ZXPsibIDc7^wL`L_=}+Q4S8^=tL5)BaZFs9w|`jxwN?CC zLGwj4s;akbn>92p?Q~V0&$)8KM9$3izQUy$H9HTV6MOX+Y4#$u1||w$tyM{|@LkPx zA0(=H+LoRDlwGWElg=WGSnrLrqp{`Z2!+O=H)ivnpL38g)4%vEmVGdoHwS}3Hx4rE zlS({^QAwg653|nZjy%6JS#lu}3GGA3a`tAG>axl;qEJ=erB^;{-`fBF5=*CBPK9^( zXZ5HeQGq3=HL9fR@2)QNo0^Skj_QxecnHGEGkI9#^ZMc}&-AkN+-V4XqcpVzuX$Ab zy>H)J1am*Td1kH?J1+2GK|Y-}}CQwDcLB zY^u(MY$j$W0pH6(J1RH2rHu<&MJ8+9vd&%TU`O7DO$A8yr0>;P3mr^Kx1=IvMS%IK z({bv@WJY>b3}r}I9ls<=OJ-aG@!Dx=7&SEH zbkFmyhvP3zwf?T)lw{JX;>z|Jev~dKkVHl=OSVKqFMHJk7E5sSbvoJQG|#e5exmKB z8yf~DpYV}b^39faEB7ea5_d^|<;F{^C(V@o;wC!c6Jiz@SI{4&D{%{oYmySi#pk#$ zTpCGKsP|@ST`m5DZZiE`Q({nc8fszyw9}zAR?(E_*ZS;QptvS~24fUt9nQ!8@GQQR z1tC(cmJ6fN?OQD#@7^N!UlW?D`HHks&CfC6Czs4+l>}_O#VpEvM37aew1zwcS8LBh zF~?pkSFLB`(ePNg-1Gv#!J$QcAANNsZi!yr#~Zj^>)ER_`?k>AH{ixeTD7QNq^qM_F0CI@n{QJkB|d#y{oxNoK2Ql zjlI>Sq<(ri$CCzvA|bmkh}2_Nj>W@-PTvp;t+``3)hT!Y#EOK}0+PuINz76GW^GWO z&Pj%w;>Ml$vl*k<=^2L(%;PXs&aCdzi#Y&=q8toBFFd0lH~!6RIq43&L6POBZ>sd}GRCnu~ZLOf#r2m6_ zY~d?>z3U@0`0rZzwOXcExPn@y7p3Qtn9dTcjg2HLKa|8p%1myBW)9y6`rtIGS+XZT z<0-Zo=ibSGl829wu%7bM?A5vdOe@~1$-mn_uhn9_>a+m_L{9zGa~jU3GASbE7mo*7o;Sdbp!QGZl)BP=wqXB` zK;L4yg;oUh5(Dcxv;tf*MZfsH?oVBnmc&vtmbLPr5g6_vzx9bI!J0=h;KoyT;%&~{ z^Qfw4B_zIQaCsonbH88g8a~|>Ra;eEF+*Ei*~hFsbiH)4>RRT|yv3z8AcGdeSmhW= zN-t}(`fS6f)o?EBA*CwS{vH5Y5&@p$VZqn-rgEYLrI)sgY|*K)ZG=E_gH>ALiE>6z zF4O%)eyDcUWXW(`ZOb})hdp+?sq`1`qwjbKtc&T&v$M5EH7LSJor>p(<9vK?2Na%B zUCXRAD9y18vF9eRQ*{-aEsA-i^;PER_imUd9jzvB)|}o(a99}{u$s)I0>z%^7Db-5 z>)8IGPc|`T4*j|KzHWE1Zlut%vct~)KxWq(b&Tep)X9*16E&UxkojB zBri)G)Tbx-s!k+b`;J;=t|kV6R4k{fo)+qe@tOFh5WIz!V$PRnTaPp>F~vWG>i(gg z-{8sP?224y1alr6b)MlnMGGkQ439v!!+;3J(K&?Y&FJa~1#xLITcZyho$qQ-Gpgx& zy^>IM1vy08T!;MW^Ep|bbC0htbpR!&&K z@l?~cZ*R!6$`(TnIn;Hlu0%TeJYFI`Ue$0fU4Cym)plWe!T4l=SD3O&5?5+h#kC7E zNZz)Ws0!7GhEhCA)%)+_Zb`|k;~mBRdGT{yd(STOtJF5wuQv5?F27iu$pJ&jlz zb3qA;B>JxI@2?JbelgRnUv9oaNMY#;iwEX60$yP zKV91No;W3yiex;h_fef!MDDlabBYU6)|BZfUr`nUBQfcFmD!(AfTtJA`%_dR6YUn#%TCCnpE3M)S)m4$foQFzQ*yRiVb6~;jS zM|9|}xcS|3R`LgT1=tlIKM?4c38b=`do_XY@zm8W-_E?dw(B-DE241A4mYhx!>6z~ zOG1+UL;m9P5{*Yt-8NZp2^yxTS&TKmn1fz66`+Obb1Gb?BbS_(H3jc*hzc`gKp|$3 z9n+fyH8tN}x_W}b(87~0QfBW7gl-@4gwvN~IrL9kggp9?0S|gkCndrNEcEE4`i#$HS;M##p)w zd)JKyEDajA|FYc~+k1~s#AN;h;<>yB)k|7qaUPE#jz`Hj6~VJ?X?^DzB7{FSe~w!Kljw3NV&H zqtyY@{ckS6a+_-27?S>a^h6b#Tf;PAnPblI%=>!f*Jm`twwO=#CgXml(GU{i$w5-M zggkuQD|-NvBzQ|0;0C1QhZ8s4I(r%d4&`AE!-vLEi_O9Cunq z0&dx!xBIoR=;;{G?Iu)oVQEk>+NU%2p{0V2h0zUg@{J)E%FMf&>?mY$*wt+tvvYal zq@M1Qt|wLYRUVriAAGtwzq$~@TAdR!ULnJ1ERfCNqF;z^P@w$a`>vN7Q8wwHq!Dp}5Y7;1| zU?pm+(HZl<&(0(m*DJqL=#kS3$k=!GQy}fd8#qXTa?t8VU6$L2PZ_Itp)5Cl`PhIm z^yP?lo|Q42e=?eul;LSQf0-1<`8D@j+kQhPAyp`8GlgqLqNxTj>+wTrC_apr0= zl%X#pPx9VF+6eZz`2nW>hvEf(aUO5&85BgkdWo=CPH2!>xt-h#8uZ~ym>4y+5pai{ z**4@GO+i((u%Ie0CGbTuNFOE-;v$U`mgEMs!q+XIwDn`pzSTMUTAVL+b+YkVEb?c| zfGYTBc9EcW*;}qZv=_yHn(mMN-D6c3zE*el;l7o0M-vB>lLP18siqDm5xaAaN=-g{ zoqf%Zj&zxceM{(LesGE5Cg+ot+{e{X{VMoaZ@h0=oL$C+GAlEzb-H5KKQ-Q}9xYOe z-hU}iNToP&A?appr)j6X2)}76DKf0&2rgs(+aTu+8<>1&iFEPK*n(`r?NVT(r#4;R_)tk}X z`b#~w{ppaY**7nVEN2{~%wj8Xk! zT55Ie6Rgn8g{lN)T|-}Be4GfG0e4}0Mk;8G_^~)&SNO@Y=3nolU2zfdA|_3Ud~U|Vl%Pur%Rg+w}G(hMb59h@W4>P4<*W3tc%Ln`c?{^?_#Ijj*+ z#+?Q)dULEe$uV6Ax(u|)gida+i@Chd=Q6^Hdec z>Wv;@QhyrF4fB+cdYs=4%62&$n-}bc_e{F8m~n*?7x>qGDe(G? zNUNQ>9!&c9?OFGo^Kzs21n5%;2}z}@YM6M)hdKEq@5RU8GOskp4h#%Lj97)R5>gzG zG%QxJvSO2P8I2@7?u%%iyv_$>=TyhO{d&GD;T%Ry%8bb$d*dn2&@om7Ni{gva#M{q z$5{n<@D4(8>PF6aZB#TCOm!ET1AZ>Qrnd?VtN9$v{O8XfCl{C3ng(&R#h8{hv8aT_beP1NAB>10aiEXZMM zyH7=Hywk1hET-!+fE9`g5CGIzY`fVgGMxvlTb#}n{THI9+HosEAPFds_pU= zUZm=d2Fu!yGjp=>uO6hxd(BfG4_Fj-9(&$1+A49ZfEUwp8lASIPbD@dS0_YyrG|)F zWVh0UrR7MqE)t=2))i6hn)c21(jQH4HJJj{(e~tJ$>Fv3(sRedBZ40OczKmxIwB&X@7go^?N#dqv3@x@xfd&qyQ|Y|`wVZ> z1u}cpalY8*&D8$avqhM+c?FylL4AH~CRkh^?6_g?VUSp}At3k$63jm4q=O?P^o@-d zO7aK@2v)yDP)tq6umP&thqiO>Zv|?!f7;!Qjg38Td-@r*^v3iVPgj2EDw1;J>2ZO? zAOrypx_VPtdsoD>&d$Wv#V8NxN`RHJT{rT{^gUNTGMyiDyfsVOlai6~CiDbD_%ZOT z*vn`z9i#T*ibsFDC6p_wd;1ZpP@T!Zqs?|!>hIz{Q50pfRUoDE8h^c!6~kM1Od68q zuloztJ$}p^>@8UBSB-R_x`&MfN)DB`1v{juIu(XPp#j$2gmj;Z?{VbN$&MY^FiOsoj{UQC@y!Q(a=g55_bn4#FetSzLf9YK4BMP^mt=r|(-p6@d$diN6T1rBt)i znb!8}2S|IJq=6J##hBcINm5B&d;jQ++f`0f1Fv{6uO|%`dir|wLmrl8+?$|7T|n?V zeI9&1M*Z)HcFxigON(F%Qsnulp@+9r&{LevixvnHF_AE!{v8Wkrc|ejsuG3A#uZ2y!Lfhb97LXUuemr*MyAsC zowUO$qY6b%MZFyV%-F#e+7N@SV5)xMbislGNUoGD-y_DsqvSFkEi&ZrXq$O__#)+B z+kQ6vk56=4+zYLme61Q#=Z%Ce>k4cITP76-lmH&!cj%1|t4TJSitQTuYhO7v9vKm9 zafnJ1Kxag#8%>5f3M5qbdRyghag@}-Sj`Fk<#t{HM}C>y%3j?-582P7U;cg-&Q2)bBQ>O`i7^6x5vS;EX}nNtT+uHX>muN)vCq zpzmH!b@Epalx8ELqi=MfYbK}c8#kDIcV1d}QTyQ(=<+~T5qhwfe;v9XYhv%Lk)NKlJ{VFbukd9x)}hg9uG2m5-GcwT)4zC$&U0UiXK^-bSL z&u^dWwxmsFEY8om-;7&%;L|wo(6Y?NiyPZpgRnqLR3^K)p=uWio!zknDh216Hs)&j zc8&nugtq=$TwN*#v_<4ofum!)MdUb09`JWr7UP$f74K??4^E~KLI9;u7(#J(-Rvp2 z3mLzvmZDm)8`LFSc_kz8MY`b+zIvEqoN@A zQIubVGy`M)odR#0j50SY^!sFKi%EY)UVAJQReLr*@LQMyg8i1)K9r-NH28BmRD73MC-eg5oosmlUg6l7sn4tm& zH_A|~54wniKG?Ag{H-dT6%X&hTC{`1&QSqw$`(!GzOmPJ|=?yy^ zB%IxLmr!;7+k0pYqC@|8$_otAOZ60$Fhs~X1^ZSNr2-N(hybw_PP-!@k>3P? ze=Io?)T-OC{VIn~*9kK$ZvF}*3kju1@*0ZR>S}Jgw~O<%{ap@Dae;wDFQy)_022>E z(N$8gg#fgY6g>Q}0YIQJi6i#P4jHCQ0Q>Ww=BYv#Bhr7D)9zBoU0JwLn*1}^GWx;J z@oz^8pt2D43se@7&&frkZ?qQ#!bG4k!`3%GARJ_pCe1D<->D%+) zz9~98^P{k+t)pXX;Q$82vb=$EncI@HUmA?{Y|lHYDCwiobrCi4Cb& zoO&dC=Tlr(bMv{q-zy>NCMI{kM$BONO)3y^aq)jY>{98#w_Q5MINU*aBU@4DN=zh` zpCAB)YtNWIe*8GxQCeEM4Isw-=cgNE3tT92i9K;!o{tzhl$f!tfCtES!KGH4$yoH! z>otX&edqII^bHKcc9-iYKJQNae6Sf@dzqm2@{Xlil#=|c9w)Ga{~5~}r=;%MqU)*s zV};7#{Syd8QIcimiaBDXAXFD3VcH3Hmvv1MeeCvR4EFZzC;{{5+P{Jdc-Nx^(9g!4 zQY&OEmk*)T@s>kEfuuLSu_^-I^eXNKCDcge1Z|nh{v7w53+fpzHOHo=in{K$L=dj^ zxCPUUJ$F$yGaF5lFaK{WIh__dL;0VLOCpO-ToyA+-*on!WV<{)3gmoGT=UnFcQg2V zT9hn{XNyL3>MWKTcb8w4)=c<^QBXvGbS41u0@u(rLqlpLuc`|3Mm32&mJ%<9R6^(G zNB(B}souLRG5-Ne*g|QMxq7kFm;LO><$b4VVG6)$xjQ9G(v@mrI_Cna0HGd`?i$?j zH~;Le6xP)-|NI=%v#T$q-Fq)lzUack^& z?k5Lh^Os(y;^N}mpWyQWmos9Z=M?Z?wR|98TUfZ(Mx*y(p@Rlhxmr?iW5<{W`bI<}qwl$b)7^Hyj#F0W zdAwfC$7<;Hv^&NheJ&>===6?0Eq?+>hS3+cdWnen~C$Wvly^~4Z&;H$!qFK~@Ome0fTsG1- zIv69Dd)uS^nbyDO+rj}XV8l)JAM-z)yA5I5;Jiyyk1iiN2~HkLx^Kyl2Q5ubUMW{p z^(cv?s{LmNbvaK7kZO%QFkxHy^^{g1u>th*-oG!#!WKP6{45h=1!sH~~cQ327*x?H+f69UR49nW)ZRV@s1dJ@@*rOgix)s+Dm6 z=@}$c(XAaEP~HG46Y0Yl$Kb%XEi_@HfS-k&3t6E=HH14AAVJn4+To9a^7pNV1ot=>hj?LYwtn}>?ch! zZ+Eum8<`-kqQQXmJX}yS;c$FVrM*to=GwBJR{!WAkBv%i#NnjG->a57o*`|3Ak`Yp zK*h~#x^aH$(fvcb-l+Cu*fo2rZ|{l6ckJN#ADnFo7E#86(a{%L`SzrRM6N_oG40%2gQ9l&cW zb9^j$^UD%ZP?g2f_Kf!uN4cv0EjCxdlcGoBjd@JjRwgdRJ)f0_YeI($@woN!Ej%!s zc*tGtBY=xz;n=sf_;QG$w%X=06ent@S+u2fkLyYhSfq!q{mKvnb zEXzb8rqqfPZk(Q^26GzOp~TnEd59`itXGLGOb1zA%FBOX-Wuy1bNJ0M_oOB1z#b$2 zAm;L*Cd*LK;a+a0CA~~b(N>cPY0i_J+$hC|PHZQEo_8qhb!!0q%o2`VV&n8k-&pCZ zt)C<|5+u~>K5Hz7GtlXe*{z;xVYt+@tphdWRQ+No;GAk_@!frlAUc~3+dJ`?*n;c2 z0!t~Jb3KxuRa{(3x#Z(e%n&OW4qaA^$5U{l-WzhJ(LMe01M7dY@THWG5XQ{hywZE& z(sQ`4NK2%*H3EqCi#;KsNUoatb z=Pmo4x6Uj>KTgrE3x`mDWQJsaB6!!fKrUzipko0YyLSS-r0Q#!5ESOOapS>ONo%U} zL2sFA@A9X*e0|;uh@-^|Vw98uIZ`#OX0nb^ei85{nz8v$$cGtf`dqI`1lw^F?2_?H zeI!cN2sIF5mufNoUh$yY^?bf29J@}ozT#pjogr^%eaBh)MHyXS-5Sgd|GRXdk+cIEnW*H3YJ^1tu^SfGK- z2fI{uMBzk=?m4DV(162D1Kek+{{kwB(vorzoWVu99jHP#vQn*LwXgU#GJpHyBh6c{8VtNmlGgDu%_4;^>}g?#fuc@rF0 zjRrI!Mk=B!qH=a}ez_nRxY|NEydMmhV(E9~!PS}k#YSCSUtcNn!GpEZ zY4f4%0GOz_M*UsbRJF$rm1E}8p<%;<#{XiuR>$^{_Z_v zJ4+8$g|nvhua$}&CqC5r|J$2*i^!WVE`F%Q6{=r7BF01pJBggYb&dN)MH;MhwZ-q` zcS#{g7xmG(y}22&wY6P^9^9Pr@|NHGI(r)a>@1*~U>UF^&Tm<7^4+~NQCvb|iXvni z_vOo%eV;y&-PVj!XA7d$O6SMQh+#wtB5*?XrrA7ZI;qel0D;6X!sE^DiOfahiHot2 zvrWm(O-6r!-n9lh3ueg)U#{h*{a{j3((|lLwZ-T~*aWay25nxf2Xw!>x%s-LaZ_M^ z)4y1mOLBO0H13X7u<7?RowxuqCztD1+68LH!$&Lw$k;;mgsG{i^-WDrc{|NSqv&&S z^cjg;<1g+B-)XYF@5W876v3;wY3b>jtV|!N;bawfijt&{1)Dys%`I+w1XT9(?J` zSMKcUs+3QE`}MB0CpYsGza&I3!5b6?UkYn3*S+mJ-cfYwC;BHZ(t<0|_-NJzi~5ddWPOtBNP0B@}(Q@7`KRzG3lw@|f|y{okMW zC!t7DRYe7liHQmQzsn&Z3dq5U41DGF;TfWiV2IgnO!)EODK^8_^pr>BjO#@3u1o!o z_gcQ7;@Fgw(D&~RS>yv5G*9Y9d}16M539@A=Q)eM{=xO|>gb?=*A zNGrP1A|vsTa}EyNTZ$S~tx@b1`WPIh+0{JFEDt!+UX$F0nxM~{Y? zm8uyT4c=W678aKBnooh#yz9Py|E>gCCtEwa4(H_f_*Yv?Dm?4w5CrF!hKi!m*cxn| zHE9_?GQ*NaPxT-sAt_B(HBj;J5N=|yH--)T=l%~LVikUW(}O6w_okch1U8ejH^f@HVc-)L#;vHzoS%MS~~;lKQXGMqgjVEIPVE-UJpjuvq0tqxw$ z;}8;}@bmNQ?Cw^91A9=%TKtMraubwy5*ApPT45NNL;x2FfJO|yS9Fqn^O8YxMj#PN zFrx&7gdU%~MWP3`RN+6wdY-}&tTK5H8j{?;2)g>aG`cdu#LqEPfVkUyJ}C6dfgsLQSBtCv-s zBFDN+Tvd(M<|nOfel0LFljNbb=Mh5&ukRd;-fQi3(Q>vQjQfHnm z!=XcDs`4fjZd^z3?We^o;lwS;`{16+i62f02&lx%1gOVLKkK&>F4Q(@eRG2}T`jVz z#fU7@f$uFUWcg_EwCwD24|c;3Jf9D??2Qi!I)JW!aa?94j~O}Y6jQNB`DYD+2C;c+ zpi)gyF{VAo!eU{fmo%DL>DvCdbRWF4uL=wyQBCHYGyA)Lyr(NY1^bVzxp@5Z>tj^@ zkZfEPZr=JP7VkEv)n<9VaJE@zq=^t3!lyT)azXam6D3W};Vto5-@nH_(34EK`b*pd z^N5QN7MT4hEKm$HmcV2jd$f+Onxs77qZ4D%14dNI%HCaz(j6EO;uce+-2 zx>gcus_ybhnB6j_tfzxFZ3Hp%l6qzw_R60kyP`Aqx$us=BaJB)RaLy_#)j+SWo2b) zm8Fm+y}qY$Z~SCmd1de2;V={Y3-&u*B(=&AUX}H=RSvX)>!jCP%eJ`1>7i{|*OB8H zhTn&j%SQIZG#%fR+VAMTOWGk{@5UsQCx#EYu54oz;U5_A^K9CLI(7K?zk@8lT&~Zi3vT?< zMV8hn&5Y^8*;sbZ;26<^XZ{Y9{);>PHUED3C>^~6f`t>^ZiKRs3m)p_*(kNms=R)G zi;#Rp-%Rqock|qHuc$nCBF?aAEsYzWmX%M#jb=q?JUuh(>}2 z|170Np%6P;58tln{S0@5D)Cyj?F$PFJ1mCHJ}hX?p46ZjBk1_a7O-IojZ$%a^Om=ZHM*ZkaFJs_*j#Fo+N@6J)qZk*`; zKYTXMGK1hKt&To_r`@@sJ=-ik0Ta5E#e*Dhiv1N|5hO|SVNoNsr_)0Ll!Gy{DAd{> zWhMjU-#rbF{%ykj?>ObPEgyxtKlw4JiKCkB;`owhBZT4}AB6hED1T=YM{(^-s{hYf zQD-?&C{^^fx1#E!1zXdO>@eg&as4Fy3w7EHFw6CR16kJZqYy-7{`H@JVO4(%VC4g( zP1n9lKLPq>`S*bIzpu}q*xqmG_(^_kmFhw9V$xTHyx?KmzBo-O7O{sB=WLK5F}KL9Td{Gwl-nssaN~< z@28VfxPHA+sK-@h6#Lg?r=Xj-L3Un$wcn=f5P| zck52Ln9E-v&b_reE{0+DQ5VYj7?O8eOSFJBqZI#OuZk5~NTli7e+&ooTMMW$i z-;tVL>dR!&UEquq42|Wgii(tT$6+-tUe*$7bxr9q&-n*sTFgjIjgNCt4-3edIz8(g zI@nn>vGn>Ni$y{oHGJOb8C+Vs4OmmbE-iSz$;ru7{X)_Pm@WMRX0r~5&zZ&mh+_*2 zDhEO;2j;MghBg-}+`k2xbQyAE#^jR-$`V6{$;tSzc1M+HM>C%hI%3UKtYdc{Rh#Ve zOns++P9{*j>E$mjW9>N&fM{X?T~@gcZn6}r&%LE4n^|uHk z+4{1Q`oR=fuXdL4GxvqwZRmOzYT-%W(m1Ow5*k+)z(zcCtIR2+hh$F(;nf**UMXcV zFa;(y(vTETi~(woGCWmmkiP80onG6ZAdvRTMX-%91SF4f4k@_B#%Xl$GEuf6e|T zS&!3{{+?~2CXzm=s56LIuf6l%^qK}GTlUv?<_cK5N0Fn;!f)Ac6Y9aQkM{qy6ZFKf z0zG{?%?$MPQZVXLQ|T}Fy5;t#;hy!iTz#kBEmG3b6*n_X>tCVGn_p{XR6;bMfk9=q zSptDOEaLDQqGSJ)A93>;;=QcO-M{kMF*b1z2eg&eJ0zP zFHQ|e@fF|gl|)Vr9}i~Nv#k;|?|hfR$gAD|kLWe+BH5eQ#g$S;vNi31Cs;rl!UrA3 z+9fM7%0T0M^o}5LQ&CAwsy8w*q4a+Vl)ZkP=bYG?GfF+K({e8;ETT?|fUncem29F; zwUkwDo*U5gyYgSy{;{MIFISmml`Jx!rip|(Se;FKs%t-t$Ptzb%=TG1|K zFfyn)jh8TpS8J-i|C4N_hgR8xv)1`sY!)0TejY7=cfj-5(dQ%5YV9Jb2E#NOg^PF$ zf0`1JtDf)b#tgtV%7_%QAFr(3wY*U{An@Zx?5uZu1Cr9Ne;Y5PjN#YbvZ5> z)8cuihEFN#&3N>HNVDun7`xcixK6@YVbEf1$}2otxlxcC6Iq=7s3>S|sbH)oI{3nR zYJ}vxCeNav_FPMns=Yz{z-qpF{qkisH~#0c<2pnk>5zBhKlys_-78A*XsMWXS8S(b zon6=CPO%w+3W>RduW^%zD=Kfc?Fxj3G`h|OZ=wQ1rU$IF`h)VYEJ>K3&e z{sc~ip3QAp3kwH{A=4H}Y0;P+X|f+}&5&#eU=>O1@9+P)1uZW>XSNvkqwDRVf5YvbZ?pe5+;)Yw zXr>;+&&kUxw|DPeq=6DJ*)8OVq|7%VxD6cCm6P_+#(#EhNR$(?){Z!T2dT{-p~xX9 z+~&JntjJ#;3I97BOaBsI%P!Q~Sr6UveM3FD=0+R%9(?Gpk?^lM*WcpV2Fx#*_rIe& z)I%EM)Pw&Cq5KuT$8R0D-5I#;l2PE4&Eui@iMzhv`)&nAMIHAp;Xp6lt{C}tU@LUu z|BK4ST(B9n=XU2PGwMOWFE!|IAM9VKj{iolY78P5JWW``X#2yR5vYZ))=BlO z0I0pay|UM?O_7s|~H6a92EYA!IZ054a(jf!tbN zUY^gBCl{=-SFb*@v9ioB>oCX+e%vMk%qN=Zqj#Ke3naNe?Ii;oXi(T)8lu1~BSVP+-( zp&Op-uxpbvt}g#x+AlN9wm`r6&%CxO6yN7KlV&&dxn_spbZ*iSdi-z4qvo+Wr9#)4m5z*WR5BlGteLv-8#U=M(i$h;XZ!+N{hZr z+Q*L{vkgCNo*nPf5j1anZf8tG_7@qih2m|$I@`i6B&2cZ(4mpeye!ZI_>&q^+OkXt zO7YU>Jw=7ZvRf3rszzQiB8hW}<68@J0YdsA5k}gwlGsz!2kv9)BhI?tzRA9^pVsI@3~5F}!` z$O{>*i%*|EZGUs);n2Mt^QLn*KW%?@@XW_(BP3%k9gB6`bPsvR?Wg|;#5Kg8Gn@BZ z;N&-Pa;K-y1-%F9a*jDi!?B)f27sW(7;~<+rb%OZYLOg|1WBgC<@~^>`>BuNl4YkT2&y7O4 z{tK%8@4)V_cl~cwv5%{T4_BAxEjr)bLxLM16QkEX=3oK*c&R^7;dYL8Q?k-=x$tx7 zCW6MiV+!)7cMia?VxTgV66<|LSyS_63+49h+osTfX%{$UKsT58kOwhQ;I}gP{$ePWMcY)N2M8}5}ESDaj znnESDJqM$ur3L)g%S0PMn1PiwEytpTi-RK}4udo^ijTBd8(>&*X>nCVy66H-3^rmN zZ+7Fxja03y#Iphdh28#$dQiIDFkf+8v~&5g5TQwV_+&*{C_cgmY1m2DCs+t97BD9( zupPH}g7@0GmL66b9c_DMQj-=&91#)RXy_5V5<4u<*N9rPvwvEcDNLMt5)d%y9QE$q zDGAk^{+TnKs&EPwB{u=G=71*m#&AOa$_&rDg?wX15BKU_rMJEe8J>e-KtfVd=)13( zMwPTtyNA-Udbol7z`4g9;M}*3252lloe~i#rx>NF`@$f`?)%mBNcm7ULTbA!IWGnD z#>BW~?>TxX~GbYB|j*te3_($RtAle zog8Z5uB5DtdnFJ`6?b2G2e?9={YC2xGoJ`9>0PEPBijX=iA7$fBKqrjh9;pvYz=aq zwA)g%kMUbV(S<{r)6}6(#nH5y52sF?kUn!}?a&fw;k$3}^XD-L!<3E$d`v>{<%;)i z_qLPpicmDOItuU-K|pGPX%h!pouG!0%lN&zmxB1> zRX{)J;-jagr<;mBJnYH!QerN1LF{?Dd;WlZnizdr{qAlU%0ea^h`)U(5a z711D?r#`Rbvq+zrxkwr(ogZnj{^oo?K_);TATW?Q89V#VhH70ILd-U*)PTfImXsVp zMkxy=3zGrYa3>VjOrFEg@(~x{W=w^!>%vK~#r}9%h3NQlHjk?Y?s=D?D{OhRBVdF= z%!P(^DBE_(K@Pv7u)5sCW(IY@$KPK`y6h&e9!Yb3sYM39RR`W8?64AuogE=?U>M2q z0phhlqoKf@M%Oo{2&krM-+ms=i;5B5F1;Q7Pta^8g9Ek9^rhEyE zR<6P8W#1^#7MpF%0u8@mVbY_eI;um^aoiU>6vY4przt>Uyrnh6LxrgC2K8BDpuZoq z9AXe&{G^k@c9dc)sJWh8Ms;yJ(=;cGUI_?o&uz0zNVl!z&8@tdaY|TNyGBAn$D-Ml zE2$y&_3K-3N{EnW=WcwAcTZG^G+{+H|e`i$VksR zpPsFpNS9p$FRAyj#A3WI>e7?#BW@t|P9%ZwgS96`lqxTbz0uAw$ByT> z&CC?YL=RT|;kkV%&(sakVyzq=B(#BCI}j?=g8YecoU8ZTwG>J7Jx*2oNxMMRqh~xE3cT)3&q|s7nJMfriA?)6oaiwis9kcUgskG<-c;#PR;n{8&eE z-Dq1)<$%qXF$aR%vPDzz`cWX&Xl3>0dveihe!E!~l)lffzmph zH-uMna0p@X-<)IFraj*^F9g+1F4}#Gt#E0?b*wE*M?peNk6JkY)_vt(F%0a8>&wMG z3Rgtkma^;X@s0khJ;`PbCrE2c%s(|AdoQv}TWsIxKF^pNpCFaU_GBAxhtUOT$DS?~ zVv(?d#M-;!+Lh5!o@YnIWLKBz>wWpFTV}{v*~V2|y!r*TSzjORxOIS8M~_&vMACP< zf!EHpiU);C2X{NSWA@uak0u5U-IT>?CbRAWDL`Bz-ep`1UB63c1wG%h9>Z?7zUGEq zB9pPbUe4I{)dA?t0!`S+v6V7Pcp8;SNB=NksJ6ogUN{ zxw)ui8OOmd(P;{B^w7$L)nyMwNlCeynQ+1VV#P8t<)==bl)>2z*B-mbvb!lwr$(R{ zK*%RRA$h?VI3D7sQN@-Auuaq8UcF+U9TT}E)MZD&0=7|FJdUAZo|{tNFmxcVs6(&) zh^*fGzEQ^z_nx^n`PTHsRm~1RUwCaB z{gLOD15;mXs;ld>%@i;=Q;JOs2nZN$ z(7szl^Vb^uTF~f%T+|@I3zB}8C8$Y>J0qfJ6EA9v1ei3i$QJ!;W%$j`^)?FNF= z>}tfWh;}%Ae|io{>hZj!fVlX0`#LF3Ja-k+07aZvX-JZY|L|}*CFXs42rscdS&9A{ zu9;{+U^2KK$_`ZEx`NoOUwuP#170j2V(JFcBU|@uB2kNqIeEv>v*d;=&zObkn>2D& z(zJ7OOqCwn*eU?1=Wp)3h4_$Wg=v!4}KGJ3>~k|s3Mtb;Mi?Eq+87Ehx?ZQr%=^r3ih zNR)*M1?XUXLEyQpfC6{K%M;g~RK;8zIPWpn)t!-*w>m{=8kt_pcPz#zDb)dU^rx2o zgp%BhL}Lm_sY#y}5p;(aj_HyI0x`2PoyDEX13mH;B-PA8bL73ZmX_-sV~!5x1EgrKNf?Z0@Z|DJn?_S;RMt4So$v3y@g52Iaglw!|J8ox|`n^+w5B zoIYUrbWMvX&@#0`Y~)CDT9R><^2(~}CY1XpWV%!aq_}2gI9|bgZAA(uosSTZVNbc& zcLLuqS$+8YC_iZ+qyu^xM=)e%B+6l0hU@wc?7h+O;<)#0fQMhhMDM&=b>z;~G3#41 zbuw&Q`#3kDxVFQi8u~%E&nYUZ3&SuC-Fpc|>+TvUl*VKw&=rZ=|eZJz+71LaqM{^L#%g0Is3=;o*b8)PhLz5R^3$bA18JYPnX0k`0h4 zDJxqxqaHklYJ2)Ikf%E}V-Cv&)$0Z-&{i?1GN6tgxH}0ttSek~KUm~syb@W(uY3nog;`Z_V*pt*n&DmUmS37tf$zZuk4u$ zs9De*YK}}nCn{mx-lSoWwoEPVVH?{FmgXq0T&N`GYVN-1a{hzsz0UtbO~1F~$bjm}2|6;GQ<$`g?E%*k{c`pR>! zel;@ivK?;oU5+r`1p8sIwBNy4;622aT+NTSGiV0SruO!B(~~v^IK8Y(gCc)4)|E~S z*!n`}=sr|alNam6^H}#ZsUL?yta~7og2b+Zo%pa#(6sg_uU6(&V4j3K(BsmKuI-zg zJR?rr&%`v~Et@L-Pv>0>`f=Ws<%>cG0;E8n1Jh(M9R24(JpOGuRE&k=pgQcGqOok~7lYQDGyA)}E z7IK)w%&-z#IZD`mB~t-uX}6E$;xjWdktGCO1cQUZ8T{f!?ANc?(KwJdM%%Iw*#vBV z;!lDR5hS_d-|cI!;ouD;?nkqk;W|(qgn<55`aAXrjQ)=vnrUB?sDpTR$7ht zCqP;e4Iroy689dLHU-^zjfW2PEuJay_3q&{Y5)B0f87W(X!b55)a$s+Zc4L`h)d!# z)@31RJ=LJ$u6us0WR>KBy0rMH6HG6ti6)~^K@IE@i=Lw%^DIT^jpBtDub?O}Wf_;Y zWe#xPX`51ubxTSROM)h3(X9vNIYn0}0$r}+hq3?<_EruQN|FWSGEXkbL~ry=e_#NY zn43Rhg+-dAa{oKD^z<4xZ{At$S%0Pk%MH<@UASe9pd|4XJQV>h} zU}Sux5t<_lz-sJpEFw^*A(UKV-BtCcZM&Of*xX_u<+0LFz>gYh-t;pEdU4SiJJ;r6 zPVJ`JFU^=i_Y>GuQRxVq>iMg!S7P6|!%Fflpa|Ni|uO!!Gg@HA7fSeS#LL*zE$#+$=RHn{|0%Vg>{$dC56^X= zn~KWHtmyiu`-{n+p6*8&>??ue7jbs5+)^?!7#ui*n27WDtQNl1)Py#f&qlu8hDw~A zb+?Q4^tgeX30OX*jM=&#w3id!CCAVJv{FcgT$}KwaG<|@`Qqc_QxC&zUNlmU-B$%G zDQ=C)dx?Q36=ZS7)sgAXm|HYP1EDWxSg+WO3QB#qzT)Mr1L}MH=qSrY7O7|l#_fOE zZMSCZ2YJqJ*AdtG^1O;W@-SHCh1XDQsbs;v!1v53c+T1S`mO8Ge6rD4`&7 z?;Bw*?m+(>oOqwC+K{c)505|6^ZahZ01z(t^)HP2<%n(qvX(9X7uJeU#jTRN{xPkC zPPrq`HXpj~*fbp>7$7{By_W3Uy`n~FHlO&GZx2K|D12#Ij!t6DDS?q+AM)h{OfFz9 zMDgWcBOJT(*d-)Pz~rs?+QM-TjwU7SQlzICKkCxAJ#ftg;Wq;rp5utm(p7scvof3; zfFV6U{n^qibFz;0n69aNz!8%pCr=6~&W_x}lfmbxk|>&J5*-x_y20>6`j8Doso=Jy z+Y%29?Po$$Dz8ovS9#WE)T6T-5*xF`+V$FS(Zy)5L=z*{&@L+tKo|~mZcn#dgpf6@ zlN-wO8zRrAzAo|hs1H;SPX%EQ3+!fQAe6cB?Hw1nmlMfFYgmGD6;>VxFSG>kd8B?o z9{TJ{IFJQFWP7qVzOi7D0e^=Ne#U(MbxU2mH?lBxpml9qao%)E>LqACWV%C-5N`GH zFU7WV&QmIQrT9;uu6B8>Kl!;BlMO8K@VQ}qq(cOgx`EizaEYv3;bkDj_pygjD2ZbS zo$sI9g$gc@lRBiG@!bt=Qgj7uw_KJTJ9g|KjiiF;%B%BM*~%&%;kz6IINFYFsG}D) zkY}LNmp@W0I6bcMb=dHt*ohPBxS+s5l^bvb4gEa0yf-i)t+vuahi9nx4ykLLwnH3s z)CqBDM)q}1^xeG>6dld<$sBB)l0dV+v4HS}!O2Hnh!wODZ>r=kKF|-&*ZV(&C&;{c zU40jj0+GH+J}ImAEHLaR=Pfmpb}=`bRRopK&bH`)C8h*dt((4F%GAJth9lLJ3X*ehcnkKA zbf}YC@`GgiU=j{YHvfU1bV6WY;QrBlTv45!ov?mZ$Sxxqpx7<&K3@Yd_>zolr~Vm;DJ>mcLvD{d0faeK(3glm|NQd<_!ry!uB8zIG25z3_)d3$7RVE%v9}#| zK%ckeXFo@z6?=-h+gzIoSfrm@<&T{QKg^j{BQ;vSPz)U-Ppv?WXguN#4b^kddV`@g zgRyfvmVw#E>P*R;VR&;%kY10y5jXuYI9=-(g?sR2l%!DP>M>QrsAQhZ~c?T}$8 za{~p(A`dH{M0)?9@e!_^{IhgcQ{` z7&@)xd1?!cY_e~a@6PM8f_qegUWJ(L(Xm~6(;KJBn=qGv7GECFy$-Gj2^0_O2)5EK z`32ASGcqQwthn-OzPUs#nD!TT7){3T<*~OKc@@x0tSG{Ti`Ev(CCtBndJ2NNBQjs8 z3Kztrq@;Y{?uS5Rq`=98C;Wetx{P`0^Je_xIn1;jt3DFT~TCnU#gu*g-7VntwHVuipxfh=`7L z!Q|um`ufIfGZn!*?W%6eb4@T5f+kcC=@35fo$0^8vKuT%(r7dqENzY8EZ8Bf|5D85 zEp~o9EuUJXgR@&-ou`0~)f{16lwv_FRPXIA=&S*?Nfg+AVMuw#)D>4?_KGBz8ojg>{C zU$8?1d1<%z6?KB(XNJciZ{D1ix_WgOR)A|Jzz5hJ`lNj1^G|0qaO*2W60noNyUd@J z0DeFW&Z=n9I8DD$TMRn`)JXX`_f9c3*NOW@D^r-&S)vDs2ea&GFf%j);feL-BVEe{ z0qi;V3~SGssVvZJWntvbHFLi=M91m{*i3-c9{5xtsR4ohn4T>iYHKo3b?ri|{Toc` zfpHRr4|GNi9PBzX%4%zbcMn6k07cbDI@X11dif5sq7K4QFu;I;s?;6=ts5j+ec0aI z6r?%DPkbM5<%(Z78lnC4u&VgmX+^6xu02E zn=j^u9SN|N5q=l1EfGMrc?B&Ln#GpmUVaxci-1w`aq?G}j}S`1FJBImIcYnJJ&J!m zew=PKFb<}KFrx`tgQeJ#k(t@ofQttB#efQ{0z0Zqv>`;nj0XE>&QF#XZU-YfX=SRC z1D&Rm+dM;V2o4PFTP*R9L*5Q36E=Ce@N8fMhKZ_cEq_@Xy*9kFXxFKygVj+4sHa7N zYP8@%j*Mi1!XX107#X91JQ6%5 zz(+N7Z-y)Z8M(&e9s3vKEKUv%DWt^!6&eYtp`s_+Z!J>*x4_;f@Pt8aH?kc0U74CZ z_=fA}tvWS1y)%D8L?Z}PMF0r|7VO8`)rbXCKG1%#0(gc%;|h2b16S6=k`sJqAV~ zFnXVDy%lgTCr9~HX(_oi<5uM0h!^MjPJN4~K)5=1&)d6}`R8-J9 zJUx~K<%sm{0o^%p&DI3}EM(P*Y-Mofvr=t4eY4*rBO?RZ=>#q7YgpVafsq%Hk*{H} zYyq!pMW&brgg_M-c&{z@tWQO&@kg?rfu^#>V{MvE5Y_>-E?NhzV?>i0JicelAz$BR zOt6V)U;tE*_NH2_MO&uP2ynI`-<(i=YzlIJcRJrG!4o%tw`>P`Js#2xX+faRL!30Q z_YrxGm=GQ%0NcDvvS+IC`&Nh^s~AXZ1RmMRfD-kff*`hGJ=e+7%jswBz`~2s(6eXg z?Z8BPtmSokth+%OYM7o+14u`oDr*x^^*d>CW>gt+RGpCo8z+!Glt|P7 zx5z%BfPYO%&zC=`fLDcee5i-7ahj*~goPsmA}oFPvC^pprd|rT=h@RM$ z@5s%~?Mtt#5A&}m_vJC6F6W^rT=h!J-PNVp%jsO_pag=`wZ{7~fXxVaCF)>n&B6W( z$-oA(DxfXk6$1l90nf+D5+sVZF3C6s=b5zG@i4 zIAsNcF)uGKGL)+6>lmnDMQ+Q$IV!&~Fn+8)atn5B;3KTm-%_zOC2-Po*fKlE(B&kB*MYE*=V12-{Tq3cVlMVTj~#(c&O8x*o0r zlTT6#>;pm;9=SBV$HGJKcqIZmF$X`pFm}fDO(iX~@fdE6bk&x0y&Sh?hbq`41PyI6 z3^k~W3WDxJ;2!~x@Ml|zAP9F(aq%MMwafH0&{83S;g^%qsy#(+O>lW&i#iZW^F1EL zh$zV_;y{!C_4!PvZ4DVV-r}h)mcYcIV01PlAAlTEslkgOa%auP>}sOx@mMc} z5h!Jc%qcCny{0!eu7E!V(cWM`Tv@F9>JsR?0_%~yWgu{6Ln*B3LpT$>GZhG?2O6q} zyz&m|?0N~wiiXcO^pnA%9jJmOB!Ded4V(cZwXqKoCTG)=akBuf4i6Rqjs8tU=LP~*4|O0OUe6XFzM-L^ zNyt6-xG$$6c>tgboB#AYmKvfT=xJXQb`gAo)OJvx!%yFMpv%X^Hd^@RUY~T~WEnFO zVZqVS#7B>45C{ikizXYK3E?+L2RAE_KrS#P!s~7}Ses(1DSXDY|ACU{>dJHt6u7O9 zl(~AO`~VB=0>CiuA_;0VPk!`fb5BNGH^{!P^t{2t{yEB#u;UOIHVF~u5Ml+u`TfH5 z$2NNXB!%dZD&szVE&U5ny1U)x^^k1`;TGw=n%oadyy>Ao4qwrmMs~eI6KYq=pg{x8 z1qaYAuw^PjMRR_J?FG_~n0zhIM}Ur_lH&5~e=bl)gUH^${qF{UY?l6`{}|zyk1l)U XiJo4S6vj0mEf!iz{%X3U;e-DJlSf`T literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/html/_images/week34_93_0.png b/doc/LectureNotes/_build/html/_images/week34_93_0.png new file mode 100644 index 0000000000000000000000000000000000000000..49c6e7364cdd160e78496a6386300a5075666900 GIT binary patch literal 24893 zcmdSBby$^M*Dtyh5Tz_S1ZkwZLy>NfZt3oBR1lC_NOyNgH!2D&x;sQlKtQ_T%p0HY zec$iwz5m$zkF&3HUC(nZUF(i{&pF5V#Td6Cit-YeXoP4m7z|TN@})8icDo7&yLEB* z4*2HlEgE6)!Rsuh;jCh3>g;CdXabWnbhfv)bGEj4P3CIi=wxAM%g)5f#L7Ts?(A&u z#K+8R^REM#>>SOQCk^*nz(wxaOKLj7VD}B7|8Ert=Uc#F21QaYpTBZX*_!k4{-twuXHByrtSqHcUceYL`sQ^QX?o6S3FivoAsIXnG(0-x(F&c``UzQIzs-|REy@%P7Vpq=9unJ_C{Isw{~40gaT1e$@p3KAtE^vOzqfvF zjUrMYD_17oa<0*3W2!3O^XPXWp1!wIVq&6F=YsQy+;po?(`sKbUL8K1ZhC-Ts-~vK zk%u^m#W2vec~{(JXI`)-fLf_2Z2UxFDZ5PUZmz(-3_$H3IBl`rO>%hDbd0ZY&sH&^yZSVW;)Htkz~GZz+PGY3wQ0yVUdn5k87U6l=-{ zcO~J_FEI*24MP*Q4J)|sUObgjX2BnQho@JtHB(2=&aSf4yf1Iu+Z{_^<8$sRczv<& zx!tfUT80Umew$p$h zw{2TK0lVohXsy!(y$-(P=}W?W4~zBJ=iC;?!0H?>$0%%+*3Azmaaz67)-DcXZq)!c z@yR|fC@dV>@jd2m+3)8EOX1j{bG94lSJK1O#K2~{(CU-KZWb}`ckRPVjD_;%IVJ(y z%en<$uN;Pq%1Y(?f(KFKrP}$37#gd|3PY`G^Jp&X$&8|+XD!!P=iS}imiq(v7Yx{0~aq;@Tn!auNLg@+~9v+QbwKL>SgX0aTQO0** zGwy4tZXce%xkF%<;t<4YH!raCIRJ?vN|Pxg>adFxpbtN9?=(%EJ?x22^e z2>TDgq`s5*e9i^C-ySRAZ;qF;gD>YjHyMovGoFi^n3!bmv|a4=gfF%S6G1cl5bYgw z7EQYMFus6$#V7C6y{>vj?OOadcQD!Xz@&89jE#+TXqu&(==h$lh}+oMEIZt}|HPU> ztGZw;y=_s*Y#?oJV{)p>j0l`!>8x#Hl6y0zVHsvl*WZ&vzt$L-nIk{!pL%!X#_L(i zrtl0+R+&W_;sh*M!(U5riumNVGv9UGm%=+z|0^r(g8|Fr zuTKL!OBmT18K2!3-{C>ywQD`?;JG{)2?u}3lM}Y1#b%rLJ-0FHih3#zLeb&ZU}o2Z zTB()^1`$`U`^jQ3MV;WqpZ1QfXo~P~B5rZpWMRM7n_(VK=tq9@nX#!O0P(@O-MCp+ z(AGI)+r+r%-2lzBS-Q_S3r711NDY?Q*w`<({n~>tZ|w}r1@EovF*QdM?ON+0Xy(G08fhK% z2?z+zP6zzTl z!rhos#grs2qe^nNPCUeX8wJ_bUdP+i3o~tgelF|7RC{)jo!I~iG)es~9CAjVCJhfO zyMSpLqUNMa!M2^Mq=H{iYFP=JSy&WOt{;pj*gZ0{u_*;J%_s!`u2`*1r_k^E*l%JP z5yJ^i>=G{pOU>;u3)YA8?D!r{3R!;DVb4jm#mGSik)1gLP6bWsXt}#C}% z*woay-#Zq6etcoW5Z!4xPNSp)z_+rplEi8h6vv>Mm6Svn{zOpg`}gngulqyA>dd-+ zmxpI_JHGh?!q+-*=grc(+7mB2dU_RF>`=-|%peQ`^74t|%F0UWg~yvt#3Fl&?C6t& zN)jZw79)8UAYJph|H)V#NG}ITPs44_B|eTxmvF?#o14H!!|-z322Nhu7J+TU`3X$% z-udNpot=J;+O8zX#%S+eH2w%gC!F*B^W*v9?phxz-Jd=VK6&K@R+^O+A+~^9mdIK^RHCrcuTlm2iQNqp!z__YNBAb39Uj?@X zpE`jj3d^o-3*`W5i}Lcs;m*+R1+nP4`yI~(q$$Lf>*lP}z?iPvT&__BuQ=F{^2EhEsY52R}*Er zE>kA)7JW&aN!)g6N858mr{)4mS+N|y7uxinMAGajSR_MP>*wo98B%n%7aKy{qwJ=QGFkG4Dy$J7tO z9^V8kYg~r8g;Hu`uBmRj_1qR>ZqMlHM?ki41~6HqUar^O(_^)}+jhE|Os#*unrufN zA_Q~F?gjZ2VsD;DQ)bz|ee2HMOS|7*y_r%<=C+F%4VPYwt>?5FHwWpA(brl}L!)40 z!_sE1v1Vv1nagI_rfv>*eSA<4thud@G!7a%I+awq&yi;K%za{>3@{-f7wGAc{_}lZj^{6!feX0%VPvKHRS5FwXb-d_e|&cCeleu|VZ#Lk8d( zZOWQ#w~w(g?zo(9)mdGh9l~=upLofFY!i*mWL6KZ<$|D!f7yKxuuJs5M0N={;2xZY zUy&~XBEfUI63o%i(3YL~4tG0Axf?I5283b;*22D_ex3z1HQ8gokGsYg@rWSkVMA$7 z&P#MWrcn_3gkv()p%a^9Akh$nEEeXc4Y&H7JA-5EPJGmL9Nx+EEFue@{CHDTS{jS3 zhw1$z0ChB9fwW<6^`ye0`y(A8uVaocm{qHt`4;$Bsyl5LE6gj74dnT@KHwQA)b)LE zEO1=Fq)IBsnXBu%#0GvbFgA*(y!>q@JhDC4bQmmr@sy2wbE15r-1m?^6{H|=87e4h zgaE2{p*1Y*O#GgzO7=!xNOD-}SWj{4A0oXz>68q&CwnjmeyQ=;U+oGii=|iJ0H_n* ztZ7le-#;)AkF7`L3zoXZb!)oIhZ!J%ZJoAJXN1ksE+B{Qx*jXBo?A5&)s|z=q!!_^ z5>Gr{f?N-w>rh@gs)!}ehn*`iq^R-na{Y2Z0}B8T1&ABfv!V8$dDph0d)3RLIrMdlo~a zwoJFFNWaY&8VjY=NWKD?nzrlX>jOZVpp6e@a-am_B-5wN(mce=TaL7rg%pG@odtn> zfq$isZ@b*GQ`OTe%Z}iaW{Q`Ip_Us~Ro5N+S+wnUIpMdS>NY>rpUOX0s$Ex*91G=a z?gej^;{`vzoPpm6n)DK@vIcQzg9AO62T=i#+y_nKb@QtF`DD0f?M0W@di(mu7OsxC zr51&%6a;qPLiEaIYg!fD4J7bJurDd>@If*#9Y`w`3&SncuCv`3$_{QlZrND}*r5j4 zy-NpynK{VPjez}AL=J)MJLfuM1H{{lFej2No*B(5Q_ksnmv&5U^Y!6eLLU2!vW8Vc zNNN_Vmybib0q{?-zE$~rZrde5f^hI^od6Cw3^FW5{s_U8w+rB47PIvZp1(4K$&cF{ zSGv1)9Kg2o1RIucE)r~_^s56JMvV#)z(~M<6cQ1m5TB4bTkt*Kl5|6=n<*6`%h+A& z+>Fz6zpju0MOb+|z#p)- zkMZ#n1rs|U3r%wz!{9)IiYRkZRslHk4yAbB98Sn7ji+NDu-$Sz|0?+QXK`9~7_1th zZP$_u^ij(Y08$$jz^G$lzQV|cAUSSR>2+)iqMSzgz79~AH~aJKa0C!c6=D)t6RTL_ z^LEXNg;K+w-d@>h@w>2)I^RprInT|~=x`yL6#GDsT!ja(x&jFsQA;$d#DS6n^qr8? z^5Zs8kcV3{++&NkDhNg>t3XhU)jKRHlR&-*!&^3_|4{!u942=lx^uKOvvPbs2B-<7 zpZONwK2|0n1jottV$)XvKp6v>qXMfuwEqgr}$YFfjh1h`Lams~n zo4}LEG?3UqG_#Cl0$b$sH`(Z3`9wA`l{{G~7kqG91-{Bh0`Ljjdj5ZGrhv2mPdu3a zzn#{gLIQ*3q!@OoXfmxuY@r>l(_f!>BNioHUcvnFqk3;waPItk*PY@KsYAs5(iR=7 zOR4X&YB`yM>Rt3L*z4zsw_r1SOnL<+1S3RLmGI^1`sbSLc|A?;hN0w0uy3Kpxc-6! zL2PF86cV)9hdmLBk_7i>=?TQ)NJy}16yRWK@4brS){GW}d$oId#Hmx#k;6p1p!-sR z;W&_>!Xz<)-BpXdF~VK81cMcV75yA{v9ymv95wPYK}5dnVrdVK{ScNR2!e7lrad$1 z+zeqa-iARapKrP273Et6GjIgHnRO4hMQG8QP`QY_ZT1WjZ$nYqxqy%9tI$2+x=YrbI*X^?X3cgD84h40j7Zr2FOFaw}z9$hXxbjS2Tq-)&HGsk;s(j0}zeV z{f_}qO9Eq=U?iWHWn+7{XB$-fuJ+SaNFbAx;p;2JZ;(!$K|0Ro!vNoCytRN&xy^e> z0C*J>6D!05z9I~F5txE-X8?+jkWAutEAtl8Z`$cxVUx3sK(+R;tfuM<@pb0zf&0qA^}^z1DJwUC`oGWdnCfyqSj>oIDT*a)bRp+FLI?8R1!r znC$qH1|gA%G``Py2M5zAO|3%(4P*{SJql>PZSmuCT<7v~w=fg4)Wa{KH%D6X5D%C8 zF}4#3lCY-8kIf&GCASTd-<-mSr;{KSujp`JOd3Od@%K4@U%A(P-d9)Q~of5aBQ)0EK*MqVA} zw)?`QFJYcGI+k*x1G77cRd@b7x(YVI@3)WxU!7`53!VS=g;Lw^yYD{ker|yqQu4A) zVnQ~(IlF?1M-0vs#lII*_hvDRds{x0LCX|(;#gXnt4%~gAf7nll>`cm4l1AE{?Kh!Ol04Qwpt46viWk0x2VA{ z4m^1d3mZ(5u^R&9A;GEhs6GP8-x-=wYTUn= zMK+g^mHY)_l0XO{*7n!`EVcumVMM`q?=fJ|tboLbNS}7D(eseye^-9<52WGaz!HK% z)`e_mKx~G9To;uZ=HF?~hU_+t36N1C9h_tu8ao9Q zgRT~Kj;#Q+@Bwo=Ha$HZ(6Pyg5QWJ-xg_$1Z-~Vc*X_HV+)M);ySGG&637#6Se#a6 z8FlIv0ZL@8_NNwq`0(LXF!E<=28XY4?N9ymo)=}qU@8vyK@EI1sBfNM0$*&KkUqJuOf{4_GW z!`){F9iRZ_K8anPj42fzHro36!gXsu|Be9UHxy!_itw>xtmnk!xNIhoI{ge8j@RFd z7J?;2(RE`0Db%cIx4Zyc4oFcp#$DCF{KC)g^&_>ph0?QiZY6hPe7eSTON-=W;abH% z^5k*E#0^%#3t}}@D68=F+%s_*wMKu{&NaDN0h8FKG2MVAN5miRjV?emc)_Ww%{gD5 zLq~17M=RCCatKF<9_E?%0jaA6(CIq#;yj#telQ4=h}_5Q7MT){Av&z24-j?8Rl4zM z*YnfW3Ve||fL!75JotqzaffWq;VUjoBbkodzeXJZ8>z~q|IF4aSJylHHjfbrc7UZZ zrd53Zf{bY2{5Llj1Nog^N0szMl5ykH<~1~#hyP=2F<&q5^J6GVFT~762Lx3HuVY0DOTxHyDLe^0f!1I}po{T!ZxBuii0PQYugb42vRCFS^wL&v z773?o9eI`q=*`WXZ6v^Ufc1rBKJ@ND3b8=GPbpnuQ~lbuF7`Eq9UR$=^cw!E-=@Xs zS6#K~#KFP)>=+F_GAq+ZF2P!zFq>FgZ~(S8Q7ItN_2V@ySv42Z57bO=B_x>*-3HM% zYO~isVn>2ht<+d*a0~w0GF#P4|9r-rb7HTj%KW)RmMv7Y9)afj13hy*pgzuC63~T* zDgwvgSmg${=mxcjN{?(UtqWJwLDb9auhsFdt_~)l$BEM8B}+PQ;wQif^cVVE*rUGE zYdEu;3$ypvZE=i<7q=XbO}l8*q|y2us-ADR5tQ`aW})Jm|IF~HCJ-gzg%43_BdOl# z;^o~>I(0R!HcdNiyw)%z18M8#fW(u(?EeG|+JaVf!U%wD-nR~-1bH~o7bLZqqD@x7 zY-B^ixnTy3;Pl6tatLc(UEQpoF#nE6vwU}856nuz*ST_$uKm1g=0jOxA)zQR!dYanJ^%{@vMpy9z=8gt>>tr=k;8B}0YF57VIXf7F zDg{;Pm$Hzqw-SL3zHQGG@io)|6AXLAJbFBtBdC@fMi?P@q#+2>`dN?uRTijAN<&3{ z$ne|*z!dWSK5P&-=&6EOB43}MSXyXheYrnOgaO__P%?+p2)V)+r~8NL*Qe>8peW`7 zYTK4y0>7m+pu@~{oz^m)elz>$$ow>|a20$3e0bpW9*(Lgtm|K&#bdM&Mv(AB9`FY6 z&c;CVz{4G$TQElRb%6tr2&Bxn{8c>?WS#QYzH29Q`Oi6Yl4!Mf)^OR)GqkgeR+%Y; zJ@ty0j+jp*+>ug4SQl(2 zDxU4!<-GJEkk2oinPj~YIu30MwHUrYnHdqffj))}kmYUFMPUB5A%_A7uTqmEklCOF zNMi}~!#_$HU~Qg$Eb27;RL7wWKSr{7m&Uu8{!T$Ht~GjLWMI{_<&}nD^K*9Admz zqI^Q>VYE@QSiu(_D1{NXr9eF=9dKqxI}OqXh_+6jKQRWkDhRVBK`{Ib`soae0JW77 zJeQ$TibOFTj*04~OrF%F7Gy&zVi6?p%cD7It=^z{u>fi0pO88(y2)aXH3 z^@}BRBqofDkxPV;!{59hECQGrB;pR9j-MW-&HqBX4YLvLW8jPUd64s(N&xde!GJuO zqVU}v|5|9EiYaLV#QGLCeQuc=BI*xd=BUAjIRmkuii*?%L=J?JCE-C@ZOFL!O*cp| z>DijRxk>*{{p-j&=1aalH>>?|9rAPCbG!?RQmf1ni`F4N;H&+YQhnQ^l#+%GF8`d> zIrBYA%^*y<`z_Hc0E}T-a$z;!hToCnj=n&8ubJ{#p~_Z~DiNjRL#%zz`_53y3Uxcn z%v;|m=R@)nBTLPY@S~*rep^bMg}8{-m*`iSMcp`bZA`ov6h5wr_O(@lIqjPBeVplWUw4wEv|3}a_UD)|X^tmTFT-~j>*Pp_kz%o2`h3;?@NSX$`syx|9#$on^`Tt6y53u4H>@1qrX5q)EFoi6oT%dR=H!*zHmWBBNHq{(K% zQ#~zHTjVI#RPGZ}l*lQ@{RT(}6{x-gN_=hRpgIbg5(Xg&0!-&zJ`UK>#<^wv(Hnjh!~L7t)Gg&A0o z@C~jFKJ}Q_XHuAA^N^8Q*4xb_4iJek6gSO# zqY}fk#GO$GVmIsFrP=L>K9y%{GrD?&BWTWR_n+ZZ{1F#oUp$1IDG&U2@6cA4yAM-c zqb1N*u|6s(lGP2+9HZ_abSq`IhfN&){5rWM9u^C5@Ib@H>CxwS&Ufi-LPh2} zLRAGiJwOZ5nBeg&MP%nl-cDcl20Wv{K)lxtO-?|YkVjFguI!YRCau7B?UKxvK8`MN z2T7F`bJI0LTKaR|9sZ@8VMqxa5v~6W%Qa6I=S_7CC(ZvY<4YSGD;0d7n`ERg=3_c@ zszZ=-2Z7@U68gvq`pBG@{S|bL1l|(^epgP18)GtcbFSLSZFP-jQGw50Mf3RMdkRZR zn0nnpJ^oQuGkqCH4S37x-oC-4G_m_zOw6ivh_zb2?ax2`D%pr=njvqGixjwF8;;k4 z;cDiX!J-kg>sgDtwDxD_C}?3gw9nKGdU-%Iq3KSOVe3-gEBdOo(>}Xp$A+6Ofz9(b zUmlbSQqELAc-^W(Z@+DfqRsdH_bfu0c#Q^Up9DW9zYNJy%SP47{othjs^rI^4L2Cy zyBCgrjaq5Lk372L5TCr6O{Gwsn?#l>o!`6=qs^TZJTWiP+ehsD$U$Kmp^&O1%C!=$I=&CG}Rc zb>nHf0vt&4E#Zo1VaZ8)uxYEPWV0G| z&|*{F@&aWuNmHZPw=w*^fnn(_38t+?-EPAUDAO-x*%u_yR{5V}i1vpQEi=+b=Q|e) znB+Qc%|3K7|8eqoS$Na>6pf}UU7vR8SN3CEwJiA5$9Mu`Q&UUOOHhc_zBfJmZC+SN zSW4-Z#ka+%bU8hkYH+!gzgP${Lh8IxDsk8Msv2PDnO zlO^dCOdpcgTD-mwgDuZ~x}2#DfSWkl%;UT0{CRup&;&`$adeDsKO#B#BL!EAx%rsm zOZXJWP-YB25PXHaptqu~nR3|SbBnDFxXscTLSKn}p&VU4vob@gxClMf~7UD2V&UiLnzbf!@JKaZHVP=4rXJ~whQHW0I z=MvWAi>HA1<|S6!fywa4ArYV*@QOX$s*FRl>{0LrexYuEkvO+JzAD4G2X})nV-W*LS-;S{0 zuMO@=>sZrvTtPJ_e-L%PgtPGE)gQfFV$}Ow1g7pSd+9>dm4-cjXZVFG`Pj(BaS^LIQWUt-DPj{_eaP7q>^) zvt+c!3hlP-^Q2fXFSlVo`$1V~bXuS|odcQZ-n(V3ea(97jOv$wKJ&IsEf^;(IzMCa zepx?YWf*;7w4c-4yWDAI1l|7RFE3_PAbddx6lv?li?m^r#Rw!Qf@UV16WTLBM;R`v zPJAT?K3Y!~-!qh3*|&hwBr6*}&3T&huszDEy-{+?zpnaa|6EfR7G{cLkvGbsO6|od z3R|TYC9g&`Wa2(hU-uq748~cYan<~Spp!UeS#s$c zH4zKsJ~+ho(_ntlO3}9qqb!@ReAK{Y7s|)PLq^kVSezkw6_{1+ZaZy(ME$R!BqSTH z_1m2{$1z!r${aQYUPi>XWza&4TcX&u6oY5{NKIy|_Erou7<~&cnBMQ7+{y)34t?}` zz#iWV1AE-(O#H0*3@sFULzs+epe?5)Zrc9uEw+ab!PCYW`NayXm_%c0Z*!)Lli+Q5 z$QqB0>1*T|Y=_V(%))E5EITn1ZdO&YGbSiXCjw~I;rNW522>V}A)vp`)m{0R( zsd{c;+}AJ0F!O^fSjd}gM8{%XzwG#(^W)3$Dr%>i(TKk@UucBCv3F8Fc9@vIJX^)@ z=%glJW?4`wC+y;~S8}452G0fCgS3@-Y^<>7VfzOr-DOgCy(_Rk0Ak?_Det*I0(o## zGp13?vK=!4tNp#Q_iH6p)fzP9tcmRZ=F8+WLj}<0O|Ayltq*4|?d!5lxG!p#j(b}6 zvj6N>(g~+ibA2shY3KIQ=QLAK~&38UcJW3@5OI!lEU%4E-cJUC$ccz&V_SCRAEV+I zEh>stJ}~9fJS?HJP&bQ#wu^u4)bgxNPP_(mHNKh$l>+Tin_&a(K6)~72o=TIt64$F85aty+e2yO(}l-7HqRqj#|nqn?Q_>s54ZYE z0xND|#2KAqKBt?+Nq!}v0}KE9!sNKCddE9I8$YdG`6PQ|L%3~lgQyCZdtCBtyVzfW zImyv>j$C2A(*Cx#8u`RHqf#e|qvHy5$^+Z@SbfTDw;e3u+QhOo)7eWq)xMSnfrHP? zvAmZO4t@e~O^t{|Cf~xSlGM7w2Zy&4V(2G~+LXqQ`)gOc^NYtD36Q;x)%x1oYy?Bz zv$EPAFmlV#RI)FO{<6cRo+$2}y`uP?Rl zp&xEAuE`0jj=6BtKd`m5^dG1mtQmMIv7>9qYNa*JX%*)wO6p#1>XGo7!5QKAo8H$}&Ct>+ zx<>Ze7gN`y_14ynJm;&saD25Dli$oR>yY7`IGbJ`!pGz+TW-f~je|>?y<9NBu%yRp9#N^bLtr6Q;G`G$A@LrG7Ww-#&O7!rpgB)e=KLQf5w=T5h&c1K3ykACj*w|P;{c{NrYpwlw z<r z)>|9&_nP`cO#Wbuj+d(w>oaC?6dFV!=bJMh()EyUJC=z0=JrD_ z_E9L1?wN37p#`f);iD%Qi{tpDz?QAoz5yGb^;z?tmIebv^e4*b?(%pHJGKs$1f52l zUv^wRxs^e|H(7*umPB7{|F@LTr#li72ZV!9D|PdRmKM2!SYgWfF*kuJPbyjpmjK{3 z!b}R{rgh7m=o-WTk}0+6gJ1HzcxobGU+(>wCjW}Up}L=C`D{bRLpmM-B3;2FB*p7k zoP-dcjR9N{4CO?CK*B}AwnT&wyr1Py8MYGj-dQNKdk=7L%+2-?tILgItBL*sQ6!r5 zM?&Pa85&m?w+RFzKqeD8qey(`0*aP6Ce|qdEB2Ca4|(dWvC5mr>v|l0S{OI#KYxDQ z4(Rgty%jtJ&uRjt1lr_%1Vf3%mAA>9Z?Mk7+@bA$IC}*aR)~2prJIkv?KR)<+L*Q$ z`NV2CNJ!`R--Fr5!=?j9@7wRlWPDKSfo%Eio|V^27|OYCLR1;@mo+xt79q=sPoQ>| z=`Pq#KGUW?K63%5M{eRV5fzksVIJ}eF-6*;RfoJ08%jbSe~rJv>T8@0O4lIefg(g; zn0*q9_GX$pm0D>k+-QQ^M55m<%U2_kAIl>`OCj7g)k~6ZX)hgO$htU zCQ(Ri!MF}(RR#8cBG;R6c8jmFe!qXFYH+$qN36>iNN0)fZ{4r;>~$!y(<52<9iQJU zn^tME7Z=xcIYpz?`Ofp`BeSp$+YjcXj31vLkFa@Mmc45I#5Mx%Ad#*qVx96FjVdEG z4!At0)bd;l_f)BeqRZ|_Bq2dWCm!=-SIgN_ffL@wyzbQn#0h}-Qv0*4^84^SUygVf zhG1ZL7tSsj`?h9;?`PCq?_xPSz57o-uC;WZ*&+d&((od}paL|-D@dtm4NEXw;+E8W)FiQVW{Q1Qwi&`sUsFul3D^DAIZD1S4rV4+ee_saQD#5X1x`{|r{Q~kAsoq8jYc6! zxL03(^ffo1Tby67`zd-m2`b>+Qco1JMCBroor!*Uggc9{q4(^9mJLr}9_^eY>MU5K z8y$luVwkp|(j+504kqn#JHvOGuYcZ`YaKMhcs;or9f5SEl$|S|3MaH=qb!<%4OrzM zH4=PwByZ9`9_v!l>h4J7Fw2|kuO5-kFi_l9P%BjNC~Q=V>!ziP&d$r5TFl&@bNMA) zHu8nYF1pcKN-u>j>em{{97YsF^=jxX<~v?Lews0xCfiM0e3zo7oG?oVNbmWqd*#9u zP`;p%!q4xc_tx;}+z*D>BDM<${fW<;Jvcwq*U(r=6{Ss>dihX%0uuOh*q0~uC%KAf z@^{%*ytyx)oV+4)1-d$3ze}YNf|RUR{XXYOSWZ04pk5HXkER;0PYY^)1ey}LWCl)> zmUdg%0_Ae2XX0N6&2UAeb2r|Ssbr<9ZDNR)9a>RzwtMyfBcO$TWAHPbD7lYbc-9M- z0YSZBNn1*)2543^uz}5qfh)Y<+>pfgw-Y2)OGy%*lFZc&B@KkKM>i;$UAc@p1jO7y z=s(!z`Z2%^AF482wz94B@oE(pt-F=s0-Vr5O<%E?$ANwE1=(NiilY!z` zUJim}Y-9-L;Wp44rD3*6BAdV{M3xT5k(0ms0;#h<74Z9ccy6B%b6{us13iyYyLcSF zGJQCO90gV|mO@$)sI2Rb9vKJaG)k@Qi$AIrXG5oyPAV~Xd=cftd-OXxzd7gX{p6#D zwllVkdV>WckbQlC=sqNkniUyIG%VQ9;!fJC+Kd{t*u72C?3PzpP9Z`b-TPGwLv(vj zd0ixScCHc!-)LTUdA%}AsxQrl=SUr)qK7+Qq1q@5l!w?~<NmZ69ZAP%nGWm!DaZ4a7D` zX^d)rMxVlCn2-vixF#Nw1%QUX)%cvp_CNcJ9Jt>+4rKUs<0b0pb19M^hu&Ygi<|_c zjB$5*JSiQ~T?|1v#|GsGEcEp7fAMlW@dhBN+t?-S>J6Cb^XS{{64Fl;&~A$t z5C7)7O?ex^AUn{_X#}qKO7QsfTe&&)ree_9<3LchDw^q=(5`c{ZTxe-qPMG5ud8Me zB0$OzzciRh-Tpy7b(vJKwsl72)9q~e_~v z*|nF4j^fJ20%4GMT6?kXg9|>zi8*7`sftX;j}Q!a^j4Cd!=@;7k_>{Nk}h=Etk$`u z|0lA3jCp1wGS#r_@HD9_h!MhZRaLoQUSa^_Syvg5RskyicU++V(g4GLA14iOFWbWV z^tipE!O=o%SQxhuQw^47N0RjS4j%dm=47~GSIto|xRMoYEg$1-_kv22NabY+mOjIFMg3Q`0e?2(B zNdM6exAt&w$0k~E=>acTz7uM?3tfSAO-A33jMYYCctJi%)vW99m7 zZ-t)V(QD|1K9bppCvy5dO=agH1yrrq-1LnW07QW{iOdhkfzI@(Wyy#1LIJR8rd99t z$&)q*ZZWndmrb=f(6mp;V&D(nCQ*J86NWSVyy#KS+oNXi4tbc8#^mn#2a3+tmc5Vh zK;*~8c3qw({mZ?>2JUW-qI_AYdI{q@1u}oGFsgFMf~UVV;jB`8osPww`a|SY^e^U}N6P_?}!SVY3eZm)ep?c=O-*e;zryalrdqe(yMi zLE=+=aVoYbR4(=yTVR(`rvG_S;`-s1ZgZ3Z*`r?B{qP7Y#_}ScZ4?vgjN|~&jCH3c zo>{??jb3y4_4iDyaQoe`=ajb+w> z%zcjzK9cz7WH@zQZ^2?ue5gd0d*N3F*tjnb6XlAHYayfl%TX$9wTw2#s1u98=lq#L zmhf4c82aeSmDloFe++p-OtC1t8AAB<0$V(e?{BBnwD1C#LBm?OA+6QHAMxEMbP4>5 zV>{&^uC4ZeJa1UgW^x0dBprvS`6#T79m!&PC#{w$c8nV)+s8qwMw|14 zL|EnUEop)&4zH#=&BE<4Av*#6GIdjoCCgSF=IcQvn7wSmVS-VGTs(X7We({QRBo>$i&d6c6 z-&ulWy83m&kn&o;#4O%G`Kz1a5H}xrDSXQLYbKWY{bC`z*EZ{qYE<=Z6iho&t#o|1 zisnY3|KT3MjcYRO&$B76C($PD3Nj~K$>o^97t9>^Xy(9&Acw! z*(5+i)T~y4Tmoty&$^c(Qe+2ubWwsLZ~S7opLN0yJUD)tFjmS=@ z-7YVLBR_n?h&j8zLszV$T0`EpN$VZziDxR5Qv%okgk_Xgcpe}8WUbp5%^2G3X zXD?R!0#Rq33oX;rH96WG`TaG7irTZayzo8mPvZT$f1Op!lc>PJ*QfjF#>U5zjur2@ z^%g)w#*9oBT(cWPmO)h__Um-XNwb6PJT~A8m9Nr(Z(8d|;-d58Tg=xG;Qe}^im^rT zQX`^o!+L{aHR&tb?{V1hl!m(Pv9j&%cbpz_0e$xF@#Me#R*8VDdmGN%_pANBX`Du` zr!GrITi;Z!YVp)ew|f6+Y^D4Fwea>g=PRX}*^5Mg4qB6L)t5aiL;D?{obo|A22kK_ z)9GN8a}s91=Sua?85J6tzMVZ=u0<-YK_7~=v`lWnlE@Opq&d|W*8jfgklfb#Sc&&& zJ2t<$b-+q9WLDqzp^Hw1!U8}WeUKq9Q4>rL@nHs-lvR=>?TV|tURiao;mdy;g!kl9 zWCM*HBQ90{m85q`3b*;tm9q|_fV7fwlsiz#`OZHugh;`ZLpIO4WthbTUFFy?cXOuIji4(=S+)dJGW#TDQpWdwUGD-zQtke<;g!$!tnf9*u$ zZ=Au2r%V$xQbdSZZg)bO#22f^w7a^G*dIcTcGUz|iODW2L$GAa;GK4UxDO3n+29*3 zVoA!}yn0(I3b{O?_V^Fh-`9y2kt^`Q%SCB%S3kracOk|C?hC28M#`=>QC#>osQY3A zXkv_ByOid|BC%xa{o}bpD#+z6U=5r=3Waz+7ir6ai%(%~Z_5*0e`uKey=3M9!?3bb zzA9%YXy=`cT^asX74txs@E&BuI$Q>on(Vz&>(bPQ+)`J17E5PP?(jR%D3pl@OzWkT zIKRM%-F8j{o+Le3V%~)uIQaA2%tX}H6F(%#kNEPH0bcW(aMNl^ zn5kSea)g?A_;wB>;FRE><^V54)YKz2f-36PX>bQ*zkc#qS+7!p~D)!Qj;QUTDF8T$m)tfH+YIk^!26#eD>kIjrE(y zpHukVytTO$8%w6%Qn zn5I$i(5zn)0U>WC6!zD^gj`2pV6tqf=SJ(8YO%@@J|5q_{r*1aoOuqlPN*n7p(C*e zb?pdR)a`ek2T{Vp-U20U=7E7Um_P47do(~vpNBB6<%lb1I7t{*>Bf(n#wcjB($q>ZS^iwgR#9$NL28Y5;u{6J>jm zB&6H>T8cT4;g}o`UM39rtgn&5>)SR#FE%X&4+P+h+#4pG&t9Py^+e`X#c?XLW7 zCAJuCefIZljMq-suo2)P`rNbz@XfB(fHL=gw+p?$syzpl!%i99raUqOG6$fY#BH=! z{$}c8@KOY2HR$3r#yxt7h5*vyPmvW6Qcw^r{4GA(Pyc}czg_?kBseK^6hlxZsCmBJ zYJK`2?RX!F{gts3un-Kx@sO;bzwrtiWz+T^@g8r7^Kq~kPu8L0EU)zN6nQxsp(ukl z>*jwVG~mns^y7X*sTHF`>wRyGyWS2Wr$TSO(mq{X!6HPb$rV_QJf|OT3}^^H4|w|{o2y4f?U9}bqb~Kt#0-cQ3;|YUt{*+XWGGgLQ=s&M z_nPc5SQ_dYVl)ER@?g3lutodPq3)4ffZJJhm+N=6bzme*gF`|EC6t4n>c zwSlFnTEOFY^O6|L*k)4$H>a{Y&d9KxC83(C+xFe^qKKZb(^KNeg)o9OBI(J)$kW<* zU_x`lH7R4WSN%Z_0$mPNj^$T88>Fp@;GeHp!xIA#s31h;7$7_LeLzJi4>VF)xy)Xn zTxXavM93a@MsRZueP2I*-FPbgl6?Lf=njP!7{{?8QI3Ckv)8OnU>gs4D6yo7hZtN0 zxEpa$RW)Zx1T>1=1bknWTp#{n1mkN8S?3GAVf|)lC4YC&Z0JoQ#4v!m^X~pLTKeeo z?EA~WX9Sf4(=K5lW)LqGlAy7Ls;(hD!nNT4spQP#pO%0UIW^8jc7-aUTuIjm68&ul=_rcAj^RRaLGFa6TW~ z!`WN;)8oQyYo$JMU%Lh@^Auzq$K@XO{Rw3dn@QFyEtFb^GlH;5xCoU7#do6X`Lpx8 z=bU@h&05-86Xv|zAh&($5{||V4yc*Zz1EGaQiJrS?7x*7oQy(23qA=?Ix#4aN7~Nq zU8eWt+c##hkwunCt=O3!&cJ+q#e44{PVv!i-&JsTp;RPm|3hXXDy=ZhU>RUD8SvE5 zvXPKb7p2P9oQsRNNhwrdCO)zxD=09>*>n~}5(*B3*=>*$0kiv;s)G+LZzykhdF#Nk z8hkXD-1teX`3382B?g>bp`gI{dA0y#I+r_83Bn7@f0ZCyTAp}SGdcLJp?2&+zvBZL zI=MvFsdwVpzV|}x$0y#2HGUacBM-w&;M$drT`*TAr60I0wnMDmlwS<|Q?*--tf(SU zRKMPAJbruFklLTZDI~|$HL&&nG_vbUJjO^uEYG#J#GO&(vd%@^!~L_M=Q$9Rsa=2gxclm9@v~bcltf`_szoF60i1RhlOJH9uZ=~q+J*~8>KwDT{OFpRhDXy*FQul)IFOf!V zdr3q2PuMg28T(~W9pZ($>Twp9DKBP|o-TNC23e(j0iyNZh6gi|C4U)O^J^UJQAkNQ27`6%JX zmpl(D$TNSBKcv!Wq$7RsCY$YgxYCJuI=T`zGdpH5Z3~2~HHFJ1LYmsK@wBJ+U|hEg zTmG=(l$PP{JP{s((231d4GFk4)7~!}k@i(_l@g*x?6^hzEv(A6AAadW91 z%Ns?OnlACRx&Z%=2xCz1HF_oF1`P~!b{TQSfwooBa}k7qW5> zer-yI#9#(@ED7og(FPgSIE>BprRkqr8bJc;GE_3CtHt-gf8cP=&kcMACv4y!$=nx{ zN_a}uA+l=xpe%=Rn7i0(EEK$rxqi44LcrsQv?$0jVldfhg?w=?Z~S)_p1>v97TvM7 zWAJ<$;1=50&%K#c3QN^PR>1;z#26!!R+YC;sDvrtR`{6Bbf>i-hhaDvFK6d+_^G|92euV8Y&w|7ui+G~( z{UHcbn@1p`Oeno&7};NUxungaSsA{-|#C((wDEM>)22ZTJpCT6Jn476|W}@ zGTF0mJVh{#x<8Un*HBoN5MbV)8uX{}Z6kY{Or8*p9kfnyT6LHGMFddluKdlG=Et!!u*1wn75IcM^`uO$Iz(*6IT9{XB>+_26JXYF{etmL*~gI zr0>k%>Fg8NJow5yek){X>*OIgH)pT2{Qp#fE5%Vz;Cd!7K%DYhR?Ct#*sMP1 zBbCpIb2iU3IC0Th0+o((ly_d#YrY3kaq3Ux&VCBpoF0O$h!d#KvyoCx)!+;pkdK%; z8-QG+z$+VEn?}8)%6zTa*pok<&pPr$Z{DoFDKcchOT^@BbGRkR=!XR1`$FNSfwEROI86A>-TjWic>f|4!=6Okx2~_L$zxe^mDcl?1yd^cIXs7-PFKSCVMPL{|+%xjq1NVDNme|jn=kFe4&WkwQvjSv8aSa=@p?+9|tKQu|HJy5-#A0p7t@OKeZKC2~6YnpQ zId+SqqoAXsq(wRNJ@9LM zI6U-h64?eITy4QzUu}v3=2T0^4+D3~=@L*vph_5|t2!Jh7?lg}1wG-0KxqJcqzY4Y zdb(u@qFL}Lok03gOnxk=d*MF{Sp5fjuB7%FnbYp0Uf-+Og@Re$LrY+l$DpCLkI~^U}JzZLQ+)psCU=(5L3wDZDvTbR`L3 zoC<^$zqHKETD5C`>a9i-hPoyqWFbB4K5a5MiEXz<1bd1fZMpTY8{KwOpiHZ7l8FSqCp4H_Z#0F-eFi|T(RjKyC7|iTTA*}d( z8yL6*tSDl7t}{hSLc&n4{JZWSx(4j|p5zd2q32M8uQoYOZ^>|5U0FuRUW4MH)WY}P zw~-_esFCK|Mf5ZZJjsKNi5aFz>dNYRgW4;7+WfBvob26`#u37>q@)A^>cq0H_K*QV zYwqgmiW(5qxM3EhwQ!M|@>`9xL!O?Zt-1MQr?HqUdgkZn5vL1Z zv%~sgZ*>^rFXB5~Gw2lUmFtSS21Ix^O3NEwff&<-2vYjA&ex!)ccYOrWc!?Uk0y+d zD-Nq}>GESX0%y!n_q^T`bk5<%S2jf10)DN>o4%ud1sBNK+1V_0D6kD~BLpYlUWfl^ zIM8xy9XsOJNNC{l9ef8Ck;v13wmSn@qlk0a@rTR3m07~7b<$yP(;MYRZ&e4_rzY{C1_~M!X@MUY6`PX*@mFZ42rrE8Sjb*^}%&)|CY`20Cyxk00|3 zCC!wRg}6gZ504%^5ifs)c+jyDV+S;BB#|Y4a-=!IJ;5YAhY+v_@k3_D&4BPDq3*&` za{TfuqcGHmlQ=+Y0?18u6gX(KQ7{2CI81oj_rCVZ+|O4eurzv=&;r1N=?_P@h%4$< zo5Kur;;BUdXEdMzjYUk)NC~h3(p)>}OrqyfX;(&l{AWiytwG}Jito&&?9PVS``tCM z4j!SZdq+AbP+rp9fk)V)y*l4?rzWg5%$kf_P+VusuN_dXHQygoTfz1A1JJ*o0Ypw^ zcn)k@&Y9dNGJp2;9X*o<;<9oca%|vmNDF8V(Fs3gx#L2?L=KS@_8ux^8 zfTP;(sQ&WW*KKyJJMv2B zYhQ1yvs|+Y8p0&%v1YIkIfx_9mE9c4rQiYV&47+H!{Y?Q_6(LN`5j-=B4NpURap3b z7J+!JvQN8$3qnkXS;PS7Lg* zMPm<`PvEeb94NU{dgC6f#7%(c1Trwt{%#|@9>7@`Li8}Qh6q>)fV~r|OzqX9BAl2;M;Bhq@qrBO<*Z}iI(w9`V2ePj z&EnpKIerh22l5F(BxL~2xODh=wh?ws>JHE|IErxc%6tIE`UQk12NL6>r2xY7qiZpz zKzNCXxdEWtDI{E9sF4eJjHWXJK%#*>{0dXqim)&15%3X_lacqL42vRwryPKv@)0;3 z;6+&o?Z(6-f h-~Rb@|EHI-Ovp8z;!3Lda2~Z_tZ%CKQr9u){{Rvf{Eh$s literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/html/_sources/exercisesweek34.ipynb b/doc/LectureNotes/_build/html/_sources/exercisesweek34.ipynb new file mode 100644 index 000000000..48f94ceda --- /dev/null +++ b/doc/LectureNotes/_build/html/_sources/exercisesweek34.ipynb @@ -0,0 +1,302 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "241c4a61", + "metadata": { + "editable": true + }, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "a33815b9", + "metadata": { + "editable": true + }, + "source": [ + "# Exercises week 34\n", + "**FYS-STK3155/4155**\n", + "\n", + "Date: **August 21-25, 2023**" + ] + }, + { + "cell_type": "markdown", + "id": "2990585c", + "metadata": { + "editable": true + }, + "source": [ + "## Exercises\n", + "\n", + "Here are three possible exercises for week 34" + ] + }, + { + "cell_type": "markdown", + "id": "be056de0", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 1: Setting up various Python environments\n", + "\n", + "The first exercise here is of a mere technical art. We want you to have \n", + "* git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). \n", + "\n", + "* Install various Python packages\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "IPython/Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow \n", + "\n", + "For **Tensorflow**, we recommend following the instructions in the text of \n", + "[Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do)\n", + "\n", + "We will come back to **tensorflow** later. \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for Python2.7)\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment." + ] + }, + { + "cell_type": "markdown", + "id": "f0f4ffae", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 2: making your own data and exploring scikit-learn\n", + "\n", + "We will generate our own dataset for a function $y(x)$ where $x \\in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\\cal {N}(0,1)$.\n", + "The following simple Python instructions define our $x$ and $y$ values (with 100 data points)." + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "93d6a2b2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "x = np.random.rand(100,1)\n", + "y = 2.0+5*x*x+0.1*np.random.randn(100,1)" + ] + }, + { + "cell_type": "markdown", + "id": "45392145", + "metadata": { + "editable": true + }, + "source": [ + "1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. \n", + "\n", + "2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. \n", + "\n", + "3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as" + ] + }, + { + "cell_type": "markdown", + "id": "3eed315b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "26038071", + "metadata": { + "editable": true + }, + "source": [ + "and the $R^2$ score function.\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "4c750d55", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "30b4731e", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "49afe51d", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "81f16b80", + "metadata": { + "editable": true + }, + "source": [ + "You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. \n", + "Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits." + ] + }, + { + "cell_type": "markdown", + "id": "cb71cb1c", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 3: Split data in test and training data\n", + "\n", + "In this exercise we want you to to compute the MSE for the training\n", + "data and the test data as function of the complexity of a polynomial,\n", + "that is the degree of a given polynomial.\n", + "\n", + "The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf).\n", + "\n", + "Our data is defined by $x\\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis." + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "b8c919e5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "np.random.seed()\n", + "n = 100\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)" + ] + }, + { + "cell_type": "markdown", + "id": "50629e14", + "metadata": { + "editable": true + }, + "source": [ + "where $y$ is the function we want to fit with a given polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "0578f08c", + "metadata": { + "editable": true + }, + "source": [ + "**a)**\n", + "Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data." + ] + }, + { + "cell_type": "markdown", + "id": "125ec9a1", + "metadata": { + "editable": true + }, + "source": [ + "**b)**\n", + "Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**)\n", + "and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "e5bb2036", + "metadata": { + "editable": true + }, + "source": [ + "**c)**\n", + "Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?" + ] + } + ], + "metadata": {}, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/doc/LectureNotes/_build/html/_sources/week34.ipynb b/doc/LectureNotes/_build/html/_sources/week34.ipynb new file mode 100644 index 000000000..83d9cbf62 --- /dev/null +++ b/doc/LectureNotes/_build/html/_sources/week34.ipynb @@ -0,0 +1,4962 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "390f1bd7", + "metadata": { + "editable": true + }, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "2c8cea05", + "metadata": { + "editable": true + }, + "source": [ + "# Week 34: Introduction to the course, Logistics and Practicalities\n", + "**Morten Hjorth-Jensen**, Department of Physics and Center for Computing in Science Education, University of Oslo, Norway and Department of Physics and Astronomy and Facility for Rare Isotope Beams, Michigan State University, USA\n", + "\n", + "Date: **Week 34, August 21-25, 2023**" + ] + }, + { + "cell_type": "markdown", + "id": "725240be", + "metadata": { + "editable": true + }, + "source": [ + "## Overview of first week\n", + "\n", + "1. The sessions on Tuesdays and Wednesdays last four hours for each group (four groups in total) and will include lectures in a flipped mode (promoting active learning) and work on exercices and projects.\n", + "\n", + "2. The sessions will begin with lectures, discussions, questions and answers about the material to be covered every week. Videos and teaching material will be announced in due time.\n", + "\n", + "3. There are four groups:\n", + "\n", + " * Tuesdays 815am-12pm and 1215pm-4pm\n", + "\n", + " * Wednesdays 815am-12pm and 1215pm-4pm.\n", + "\n", + "4. On Thursdays we have a regular lecture. These lectures start at 1215pm and end at 2pm and serve the aims of giving an overview over various topics. These lectures will also be recorded.\n", + "\n", + "The labs are also available till 6pm Tuesdays and Wednesdays. Videos and learning material with reading suggestions will be made available before each week starts." + ] + }, + { + "cell_type": "markdown", + "id": "c18f9a3c", + "metadata": { + "editable": true + }, + "source": [ + "## Schedule first week\n", + "\n", + " * August 22: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.\n", + "\n", + " * August 23: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.\n", + "\n", + " * August 24: Lecture: Linear regression, examples and theory" + ] + }, + { + "cell_type": "markdown", + "id": "769cbfc6", + "metadata": { + "editable": true + }, + "source": [ + "## Lectures and ComputerLab\n", + "\n", + " * The sessions on Tuesdays and Wednesdays last four hours and will include partly lectures in a flipped mode (promoting active learning) and work on exercices and projects.\n", + "\n", + " * Thursdays: regular lectures (12.15pm-2pm) \n", + "\n", + " * Weekly reading assignments and videos needed to solve projects and exercises.\n", + "\n", + " * Weekly exercises. You can hand in exercises if you want and get an extra score, see below.\n", + "\n", + " * Detailed lecture notes, exercises, all programs presented, projects etc can be found at the homepage of the course.\n", + "\n", + " * Weekly plans and all other information are on the official website. This info will also be conveyed via weekly emails.\n", + "\n", + " * No final exam, three projects that are graded and have to be approved." + ] + }, + { + "cell_type": "markdown", + "id": "5040ec52", + "metadata": { + "editable": true + }, + "source": [ + "## Communication channels\n", + "\n", + "* Chat and communications via \n", + "\n", + "* **Discord** channel will be added asap" + ] + }, + { + "cell_type": "markdown", + "id": "867e5431", + "metadata": { + "editable": true + }, + "source": [ + "## Course Format\n", + "\n", + " * Three compulsory projects. Electronic reports only using [Canvas](https://www.uio.no/english/services/it/education/canvas/) to hand in projects and [git](https://git-scm.com/) as version control software and [GitHub](https://github.com/) for repository (or [GitLab](https://about.gitlab.com/)) of all your material.\n", + "\n", + " * Evaluation and grading: The three projects are graded and each counts 1/3 of the final mark. No final written or oral exam.\n", + "\n", + "a. For the last project each group/participant submits a proposal or works with suggested (by us) proposals for the project.\n", + "\n", + "b. If possible, we would like to organize the last project as a workshop where each group presents this to all other participants of the course\n", + "\n", + "c. Based on feedback etc, each group finalizes the report and submits for grading. \n", + "\n", + " * Python is the default programming language, but feel free to use C/C++, Julia and/or Fortran or other programming languages. All source codes discussed during the lectures can be found at the webpage and [github address](https://github.com/CompPhysics/MachineLearning/tree/master/doc/Programs) of the course." + ] + }, + { + "cell_type": "markdown", + "id": "61ff07a3", + "metadata": { + "editable": true + }, + "source": [ + "## Teachers\n", + "\n", + "* Morten Hjorth-Jensen, morten.hjorth-jensen@fys.uio.no\n", + "\n", + " * **Phone**: +47-48257387\n", + "\n", + " * **Office**: Department of Physics, University of Oslo, Eastern wing, room FØ470 \n", + "\n", + " * **Office hours**: *Anytime*! Individual or group office hours can be arranged either in person or via zoom. Feel free to send an email for planning. \n", + "\n", + "* Ida Torkjellsdatter Storehaug, i.t.storehaug@fys.uio.no\n", + "\n", + "* Fahimeh Najafi, fahimeh.najafi@fys.uio.no\n", + "\n", + "* Mia-Katrin Ose Kvalsund, m.k.o.kvalsund@fys.uio.no\n", + "\n", + "* Karl Henrik Fredly, k.h.fredly@fys.uio.no\n", + "\n", + "* Adam Jakobsen, adam.jakobsen@fys.uio.no\n", + "\n", + "* Daniel Haas Beccatini Lima, d.h.b.lima@fys.uio.no" + ] + }, + { + "cell_type": "markdown", + "id": "afa7d17c", + "metadata": { + "editable": true + }, + "source": [ + "## Deadlines for projects (tentative)\n", + "\n", + "1. Project 1: October 9 (available September 4) graded with feedback)\n", + "\n", + "2. Project 2: November 6 (available October 6, graded with feedback)\n", + "\n", + "3. Project 3: December 11 (available November 10, graded with feedback)\n", + "\n", + "Extra Credit (not mandatory), weekly exercise assignments, 10 in total (due Friday same week), 10% additional score. The extra credit assignments are due each Friday and can be uploaed to **Canvas** in your preferred format (although we prefer jupyter-notebooks). First assignment is for week 35. Each weekly exercise set counts 1%." + ] + }, + { + "cell_type": "markdown", + "id": "37a46b20", + "metadata": { + "editable": true + }, + "source": [ + "## Grading\n", + "\n", + "Grades are awarded on a scale from A to F, where A is the best grade and F is a fail. There are three projects which are graded and each project counts 1/3 of the final grade. The total score is thus the average from all three projects.\n", + "\n", + "The final number of points is based on the average of all projects and the grade follows the following table:\n", + "\n", + " * 92-100 points: A\n", + "\n", + " * 77-91 points: B\n", + "\n", + " * 58-76 points: C\n", + "\n", + " * 46-57 points: D\n", + "\n", + " * 40-45 points: E\n", + "\n", + " * 0-39 points: F-failed\n", + "\n", + "In addition you can get an extra 10% score for weekly assignments (10 in total and due each Friday). Each weekly assignment counts 1%." + ] + }, + { + "cell_type": "markdown", + "id": "40b50d78", + "metadata": { + "editable": true + }, + "source": [ + "## Reading material\n", + "\n", + "The lecture notes are collected as a jupyter-book at .\n", + "\n", + "In addition to the lecture notes, we recommend the books of Bishop, Hastie et al, Murphy and Goodfellow et al. We will follow these texts closely and the weekly reading assignments refer to these texts. The text by Hastie et al is also widely used in the Machine Learning community. Finally, we also recommend the hands-on text by Geron, see next slide for links." + ] + }, + { + "cell_type": "markdown", + "id": "1ac7f5fa", + "metadata": { + "editable": true + }, + "source": [ + "## Textbooks\n", + "\n", + "* [Goodfellow, Bengio, and Courville (GBC), Deep Learning](https://www.deeplearningbook.org/)\n", + "\n", + "* Christopher M. Bishop (CB), Pattern Recognition and Machine Learning\n", + "\n", + "* [Hastie, Tibshirani, and Friedman (HTF), The Elements of Statistical Learning, Springer](https://www.springer.com/gp/book/9780387848570).\n", + "\n", + "* [Aurelien Geron (AG), Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/). This text is very useful since it contains many code examples and hands-on applications of all algorithms discussed in this course.\n", + "\n", + "* [Kevin Murphy (KM), Probabilistic Machine Learning, an Introduction](https://probml.github.io/pml-book/book1.html)" + ] + }, + { + "cell_type": "markdown", + "id": "c74dc137", + "metadata": { + "editable": true + }, + "source": [ + "## Reading suggestions week 34\n", + "\n", + "This week: Refresh linear algebra, GBC chapters 1 and 2. HTF chapters 2 and 3. Install scikit-learn. See lecture notes for week 34 at (these notes)." + ] + }, + { + "cell_type": "markdown", + "id": "6af0d013", + "metadata": { + "editable": true + }, + "source": [ + "## Prerequisites\n", + "\n", + "Basic knowledge in programming and mathematics, with an emphasis on\n", + "linear algebra. Knowledge of Python or/and C++ as programming\n", + "languages is strongly recommended and experience with Jupiter notebook\n", + "is recommended. Required courses are the equivalents to the University\n", + "of Oslo mathematics courses MAT1100, MAT1110, MAT1120 and at least one\n", + "of the corresponding computing and programming courses INF1000/INF1110\n", + "or MAT-INF1100/MAT-INF1100L/BIOS1100/KJM-INF1100. Most universities\n", + "offer nowadays a basic programming course (often compulsory) where\n", + "Python is the recurring programming language." + ] + }, + { + "cell_type": "markdown", + "id": "628fb74c", + "metadata": { + "editable": true + }, + "source": [ + "## Topics covered in this course: Statistical analysis and optimization of data\n", + "\n", + "The course has two central parts\n", + "\n", + "1. Statistical analysis and optimization of data\n", + "\n", + "2. Machine learning\n", + "\n", + "These topics will be scattered thorughout the course and may not necessarily be taught separately. Rather, we will often take an approach (during the lectures and project/exercise sessions) where say elements from statistical data analysis are mixed with specific Machine Learning algorithms." + ] + }, + { + "cell_type": "markdown", + "id": "bcfe0b24", + "metadata": { + "editable": true + }, + "source": [ + "## Statistical analysis and optimization of data\n", + "\n", + "We plan to cover the following topics:\n", + "* Basic concepts, expectation values, variance, covariance, correlation functions and errors;\n", + "\n", + "* Simpler models, binomial distribution, the Poisson distribution, simple and multivariate normal distributions;\n", + "\n", + "* Central elements of Bayesian statistics and modeling;\n", + "\n", + "* Gradient methods for data optimization;\n", + "\n", + "* Monte Carlo methods, Markov chains, Gibbs sampling and Metropolis-Hastings sampling (tentative);\n", + "\n", + "* Estimation of errors and resampling techniques such as the cross-validation, blocking, bootstrapping and jackknife methods;\n", + "\n", + "* Principal Component Analysis (PCA) and its mathematical foundation;" + ] + }, + { + "cell_type": "markdown", + "id": "f153abef", + "metadata": { + "editable": true + }, + "source": [ + "## Machine Learning\n", + "\n", + "* Pre deep-learning revolution (2008 approx)\n", + "\n", + " * Linear Regression and Logistic Regression, classification and regression problems;\n", + "\n", + " * Bayesian linear and logistic regression, kernel regression;\n", + "\n", + " * Decisions trees, Random Forests, Bagging and Boosting methods;\n", + "\n", + " * Support vector machines (only survey);\n", + "\n", + " * Unsupervised learning and dimensionality reduction, from PCA to clustering; \n", + "\n", + "* Deep learning \n", + "\n", + " * Neural networks and deep learning;\n", + "\n", + " * Convolutional neural networks;\n", + "\n", + " * Recurrent neural networks;\n", + "\n", + " * Autoencoders\n", + "\n", + " * Generative methods with an emphasis on Boltzmann Machines, Variational Autoencoders and Generalized Adversarial Networks;\n", + "\n", + "Hands-on demonstrations, exercises and projects aim at deepening your understanding of these topics." + ] + }, + { + "cell_type": "markdown", + "id": "a6b0c4b8", + "metadata": { + "editable": true + }, + "source": [ + "## Extremely useful tools, strongly recommended\n", + "\n", + "**and discussed at the lab sessions.**\n", + "\n", + " * GIT for version control, and GitHub or GitLab as repositories, highly recommended. This will be discussed during the first exercise session\n", + "\n", + " * Anaconda and other Python environments, see intro slides and links to programming resources at " + ] + }, + { + "cell_type": "markdown", + "id": "48eb8817", + "metadata": { + "editable": true + }, + "source": [ + "## Other courses on Data science and Machine Learning at UiO\n", + "\n", + "* [FYS5419 Quantum Computing and Quantum Machine Learning](https://www.uio.no/studier/emner/matnat/fys/FYS5419/index-eng.html)\n", + "\n", + "* [FYS5429 Advanced Machine Learning for the Physical Sciences](https://www.uio.no/studier/emner/matnat/fys/FYS5429/index-eng.html)\n", + "\n", + "* [STK2100 Machine learning and statistical methods for prediction and classification](http://www.uio.no/studier/emner/matnat/math/STK2100/index-eng.html). \n", + "\n", + "* [IN3050/4050 Introduction to Artificial Intelligence and Machine Learning](https://www.uio.no/studier/emner/matnat/ifi/IN3050/index-eng.html). Introductory course in machine learning and AI with an algorithmic approach. \n", + "\n", + "* [STK-INF3000/4000 Selected Topics in Data Science](http://www.uio.no/studier/emner/matnat/math/STK-INF3000/index-eng.html). The course provides insight into selected contemporary relevant topics within Data Science. \n", + "\n", + "* [IN4080 Natural Language Processing](https://www.uio.no/studier/emner/matnat/ifi/IN4080/index.html). Probabilistic and machine learning techniques applied to natural language processing." + ] + }, + { + "cell_type": "markdown", + "id": "e8889321", + "metadata": { + "editable": true + }, + "source": [ + "## Other courses on Data science and Machine Learning at UiO, contn\n", + "\n", + "* [STK-IN4300 Statistical learning methods in Data Science](https://www.uio.no/studier/emner/matnat/math/STK-IN4300/index-eng.html). An advanced introduction to statistical and machine learning. For students with a good mathematics and statistics background.\n", + "\n", + "* [IN3310/4310 Deep Learnig for Image Analysis](https://www.uio.no/studier/emner/matnat/ifi/IN4310/index.html)\n", + "\n", + "* [STK4051 Computational Statistics](https://www.uio.no/studier/emner/matnat/math/STK4051/index-eng.html)\n", + "\n", + "* [STK4021 Applied Bayesian Analysis and Numerical Methods](https://www.uio.no/studier/emner/matnat/math/STK4021/index-eng.html)" + ] + }, + { + "cell_type": "markdown", + "id": "4ad1f253", + "metadata": { + "editable": true + }, + "source": [ + "## Learning outcomes\n", + "\n", + "This course aims at giving you insights and knowledge about many of\n", + "the central algorithms used in Data Analysis and Machine Learning.\n", + "The course is project based and through various numerical projects,\n", + "normally three, you will be exposed to fundamental research problems\n", + "in these fields, with the aim to reproduce state of the art scientific\n", + "results. Both supervised and unsupervised methods will be covered. The\n", + "emphasis is on a frequentist approach, although we will try to link it\n", + "with a Bayesian approach as well. You will learn to develop and\n", + "structure large codes for studying different cases where Machine\n", + "Learning is applied to, get acquainted with computing facilities and\n", + "learn to handle large scientific projects. A good scientific and\n", + "ethical conduct is emphasized throughout the course. More\n", + "specifically, after this course you will\n", + "\n", + "* Learn about basic data analysis, statistical analysis, Bayesian statistics, Monte Carlo sampling, data optimization and machine learning;\n", + "\n", + "* Be capable of extending the acquired knowledge to other systems and cases;\n", + "\n", + "* Have an understanding of central algorithms used in data analysis and machine learning;\n", + "\n", + "* Understand linear methods for regression and classification, from ordinary least squares, via Lasso and Ridge to Logistic regression;\n", + "\n", + "* Learn about neural networks and deep learning methods for supervised and unsupervised learning. Emphasis on feed forward neural networks, convolutional and recurrent neural networks; \n", + "\n", + "* Learn about about decision trees, random forests, bagging and boosting methods;\n", + "\n", + "* Learn about support vector machines and kernel transformations;\n", + "\n", + "* Reduction of data sets, from PCA to clustering;\n", + "\n", + "* Generative models\n", + "\n", + "* Work on numerical projects to illustrate the theory. The projects play a central role and you are expected to know modern programming languages like Python or C++ and/or Fortran (Fortran2003 or later) or Julia or other." + ] + }, + { + "cell_type": "markdown", + "id": "ed3eddab", + "metadata": { + "editable": true + }, + "source": [ + "## Introduction\n", + "\n", + "Our emphasis throughout this series of lectures \n", + "is on understanding the mathematical aspects of\n", + "different algorithms used in the fields of data analysis and machine learning. \n", + "\n", + "However, where possible we will emphasize the\n", + "importance of using available software. We start thus with a hands-on\n", + "and top-down approach to machine learning. The aim is thus to start with\n", + "relevant data or data we have produced \n", + "and use these to introduce statistical data analysis\n", + "concepts and machine learning algorithms before we delve into the\n", + "algorithms themselves. The examples we will use in the beginning, start with simple\n", + "polynomials with random noise added. We will use the Python\n", + "software package [Scikit-Learn](http://scikit-learn.org/stable/) and\n", + "introduce various machine learning algorithms to make fits of\n", + "the data and predictions. We move thereafter to more interesting\n", + "cases such as data from say experiments (below we will look at experimental nuclear binding energies as an example).\n", + "These are examples where we can easily set up the data and\n", + "then use machine learning algorithms included in for example\n", + "**Scikit-Learn**. \n", + "\n", + "These examples will serve us the purpose of getting\n", + "started. Furthermore, they allow us to catch more than two birds with\n", + "a stone. They will allow us to bring in some programming specific\n", + "topics and tools as well as showing the power of various Python \n", + "libraries for machine learning and statistical data analysis. \n", + "\n", + "Although we have projects where you write your own codes, we will also focus on two\n", + "specific Python packages for Machine Learning, Scikit-Learn and\n", + "Tensorflow with Keras (see below for links etc). Moreover, the examples we\n", + "introduce will serve as inputs to many of our discussions later, as\n", + "well as allowing you to set up models and produce your own data and\n", + "get started with programming." + ] + }, + { + "cell_type": "markdown", + "id": "cd6f4f41", + "metadata": { + "editable": true + }, + "source": [ + "## AI/ML and some statements you may have heard (and what do they mean?)\n", + "\n", + "1. Fei-Fei Li on ImageNet: **map out the entire world of objects** ([The data that transformed AI research](https://cacm.acm.org/news/219702-the-data-that-transformed-ai-research-and-possibly-the-world/fulltext))\n", + "\n", + "2. Russell and Norvig in their popular textbook: **relevant to any intellectual task; it is truly a universal field** ([Artificial Intelligence, A modern approach](http://aima.cs.berkeley.edu/))\n", + "\n", + "3. Woody Bledsoe puts it more bluntly: **in the long run, AI is the only science** (quoted in Pamilla McCorduck, [Machines who think](https://www.pamelamccorduck.com/machines-who-think))\n", + "\n", + "If you wish to have a critical read on AI/ML from a societal point of view, see [Kate Crawford's recent text Atlas of AI](https://www.katecrawford.net/)\n", + "\n", + "**Here: with AI/ML we intend a collection of machine learning methods with an emphasis on statistical learning and data analysis**" + ] + }, + { + "cell_type": "markdown", + "id": "9bc6760a", + "metadata": { + "editable": true + }, + "source": [ + "## What is Machine Learning?\n", + "\n", + "Statistics, data science and machine learning form important fields of\n", + "research in modern science. They describe how to learn and make\n", + "predictions from data, as well as allowing us to extract important\n", + "correlations about physical process and the underlying laws of motion\n", + "in large data sets. The latter, big data sets, appear frequently in\n", + "essentially all disciplines, from the traditional Science, Technology,\n", + "Mathematics and Engineering fields to Life Science, Law, education\n", + "research, the Humanities and the Social Sciences. \n", + "\n", + "It has become more\n", + "and more common to see research projects on big data in for example\n", + "the Social Sciences where extracting patterns from complicated survey\n", + "data is one of many research directions. Having a solid grasp of data\n", + "analysis and machine learning is thus becoming central to scientific\n", + "computing in many fields, and competences and skills within the fields\n", + "of machine learning and scientific computing are nowadays strongly\n", + "requested by many potential employers. The latter cannot be\n", + "overstated, familiarity with machine learning has almost become a\n", + "prerequisite for many of the most exciting employment opportunities,\n", + "whether they are in bioinformatics, life science, physics or finance,\n", + "in the private or the public sector. This author has had several\n", + "students or met students who have been hired recently based on their\n", + "skills and competences in scientific computing and data science, often\n", + "with marginal knowledge of machine learning.\n", + "\n", + "Machine learning is a subfield of computer science, and is closely\n", + "related to computational statistics. It evolved from the study of\n", + "pattern recognition in artificial intelligence (AI) research, and has\n", + "made contributions to AI tasks like computer vision, natural language\n", + "processing and speech recognition. Many of the methods we will study are also \n", + "strongly rooted in basic mathematics and physics research. \n", + "\n", + "Ideally, machine learning represents the science of giving computers\n", + "the ability to learn without being explicitly programmed. The idea is\n", + "that there exist generic algorithms which can be used to find patterns\n", + "in a broad class of data sets without having to write code\n", + "specifically for each problem. The algorithm will build its own logic\n", + "based on the data. You should however always keep in mind that\n", + "machines and algorithms are to a large extent developed by humans. The\n", + "insights and knowledge we have about a specific system, play a central\n", + "role when we develop a specific machine learning algorithm. \n", + "\n", + "Machine learning is an extremely rich field, in spite of its young\n", + "age. The increases we have seen during the last three decades in\n", + "computational capabilities have been followed by developments of\n", + "methods and techniques for analyzing and handling large date sets,\n", + "relying heavily on statistics, computer science and mathematics. The\n", + "field is rather new and developing rapidly. Popular software packages\n", + "written in Python for machine learning like\n", + "[Scikit-learn](http://scikit-learn.org/stable/),\n", + "[Tensorflow](https://www.tensorflow.org/),\n", + "[PyTorch](http://pytorch.org/) and [Keras](https://keras.io/), all\n", + "freely available at their respective GitHub sites, encompass\n", + "communities of developers in the thousands or more. And the number of\n", + "code developers and contributors keeps increasing. Not all the\n", + "algorithms and methods can be given a rigorous mathematical\n", + "justification, opening up thereby large rooms for experimenting and\n", + "trial and error and thereby exciting new developments. However, a\n", + "solid command of linear algebra, multivariate theory, probability\n", + "theory, statistical data analysis, understanding errors and Monte\n", + "Carlo methods are central elements in a proper understanding of many\n", + "of algorithms and methods we will discuss." + ] + }, + { + "cell_type": "markdown", + "id": "17c6257a", + "metadata": { + "editable": true + }, + "source": [ + "## Types of Machine Learning\n", + "\n", + "The approaches to machine learning are many, but are often split into\n", + "two main categories. In *supervised learning* we know the answer to a\n", + "problem, and let the computer deduce the logic behind it. On the other\n", + "hand, *unsupervised learning* is a method for finding patterns and\n", + "relationship in data sets without any prior knowledge of the system.\n", + "Some authours also operate with a third category, namely\n", + "*reinforcement learning*. This is a paradigm of learning inspired by\n", + "behavioral psychology, where learning is achieved by trial-and-error,\n", + "solely from rewards and punishment.\n", + "\n", + "Another way to categorize machine learning tasks is to consider the\n", + "desired output of a system. Some of the most common tasks are:\n", + "\n", + " * Classification: Outputs are divided into two or more classes. The goal is to produce a model that assigns inputs into one of these classes. An example is to identify digits based on pictures of hand-written ones. Classification is typically supervised learning.\n", + "\n", + " * Regression: Finding a functional relationship between an input data set and a reference data set. The goal is to construct a function that maps input data to continuous output values.\n", + "\n", + " * Clustering: Data are divided into groups with certain common traits, without knowing the different groups beforehand. It is thus a form of unsupervised learning." + ] + }, + { + "cell_type": "markdown", + "id": "9f4e80cf", + "metadata": { + "editable": true + }, + "source": [ + "## Essential elements of ML\n", + "\n", + "The methods we cover have three main topics in common, irrespective of\n", + "whether we deal with supervised or unsupervised learning.\n", + "* The first ingredient is normally our data set (which can be subdivided into training, validation and test data). Many find the most difficult part of using Machine Learning to be the set up of your data in a meaningful way. \n", + "\n", + "* The second item is a model which is normally a function of some parameters. The model reflects our knowledge of the system (or lack thereof). As an example, if we know that our data show a behavior similar to what would be predicted by a polynomial, fitting our data to a polynomial of some degree would then determin our model. \n", + "\n", + "* The last ingredient is a so-called **cost/loss** function (or error or risk function) which allows us to present an estimate on how good our model is in reproducing the data it is supposed to train." + ] + }, + { + "cell_type": "markdown", + "id": "828f3ad9", + "metadata": { + "editable": true + }, + "source": [ + "## An optimization/minimization problem\n", + "\n", + "At the heart of basically all Machine Learning algorithms we will encounter so-called minimization or optimization algorithms. A large family of such methods are so-called **gradient methods**." + ] + }, + { + "cell_type": "markdown", + "id": "8b12df05", + "metadata": { + "editable": true + }, + "source": [ + "## A Frequentist approach to data analysis\n", + "\n", + "When you hear phrases like **predictions and estimations** and\n", + "**correlations and causations**, what do you think of? May be you think\n", + "of the difference between classifying new data points and generating\n", + "new data points.\n", + "Or perhaps you consider that correlations represent some kind of symmetric statements like\n", + "if $A$ is correlated with $B$, then $B$ is correlated with\n", + "$A$. Causation on the other hand is directional, that is if $A$ causes $B$, $B$ does not\n", + "necessarily cause $A$.\n", + "\n", + "These concepts are in some sense the difference between machine\n", + "learning and statistics. In machine learning and prediction based\n", + "tasks, we are often interested in developing algorithms that are\n", + "capable of learning patterns from given data in an automated fashion,\n", + "and then using these learned patterns to make predictions or\n", + "assessments of newly given data. In many cases, our primary concern\n", + "is the quality of the predictions or assessments, and we are less\n", + "concerned about the underlying patterns that were learned in order\n", + "to make these predictions.\n", + "\n", + "In machine learning we normally use [a so-called frequentist approach](https://en.wikipedia.org/wiki/Frequentist_inference),\n", + "where the aim is to make predictions and find correlations. We focus\n", + "less on for example extracting a probability distribution function (PDF). The PDF can be\n", + "used in turn to make estimations and find causations such as given $A$\n", + "what is the likelihood of finding $B$." + ] + }, + { + "cell_type": "markdown", + "id": "6656791e", + "metadata": { + "editable": true + }, + "source": [ + "## What is a good model?\n", + "\n", + "In science and engineering we often end up in situations where we want to infer (or learn) a\n", + "quantitative model $M$ for a given set of sample points $\\boldsymbol{X} \\in [x_1, x_2,\\dots x_N]$.\n", + "\n", + "As we will see repeatedely in these lectures, we could try to fit these data points to a model given by a\n", + "straight line, or if we wish to be more sophisticated to a more complex\n", + "function.\n", + "\n", + "The reason for inferring such a model is that it\n", + "serves many useful purposes. On the one hand, the model can reveal information\n", + "encoded in the data or underlying mechanisms from which the data were generated. For instance, we could discover important\n", + "corelations that relate interesting physics interpretations.\n", + "\n", + "In addition, it can simplify the representation of the given data set and help\n", + "us in making predictions about future data samples.\n", + "\n", + "A first important consideration to keep in mind is that inferring the *correct* model\n", + "for a given data set is an elusive, if not impossible, task. The fundamental difficulty\n", + "is that if we are not specific about what we mean by a *correct* model, there\n", + "could easily be many different models that fit the given data set *equally well*." + ] + }, + { + "cell_type": "markdown", + "id": "85f146c9", + "metadata": { + "editable": true + }, + "source": [ + "## What is a good model? Can we define it?\n", + "\n", + "The central question is this: what leads us to say that a model is correct or\n", + "optimal for a given data set? To make the model inference problem well posed, i.e.,\n", + "to guarantee that there is a unique optimal model for the given data, we need to\n", + "impose additional assumptions or restrictions on the class of models considered. To\n", + "this end, we should not be looking for just any model that can describe the data.\n", + "Instead, we should look for a **model** $M$ that is the best among a restricted class\n", + "of models. In addition, to make the model inference problem computationally\n", + "tractable, we need to specify how restricted the class of models needs to be. A\n", + "common strategy is to start \n", + "with the simplest possible class of models that is just necessary to describe the data\n", + "or solve the problem at hand. More precisely, the model class should be rich enough\n", + "to contain at least one model that can fit the data to a desired accuracy and yet be\n", + "restricted enough that it is relatively simple to find the best model for the given data.\n", + "\n", + "Thus, the most popular strategy is to start from the\n", + "simplest class of models and increase the complexity of the models only when the\n", + "simpler models become inadequate. For instance, if we work with a regression problem to fit a set of sample points, one\n", + "may first try the simplest class of models, namely linear models, followed obviously by more complex models.\n", + "\n", + "How to evaluate which model fits best the data is something we will come back to over and over again in these sets of lectures." + ] + }, + { + "cell_type": "markdown", + "id": "51f73840", + "metadata": { + "editable": true + }, + "source": [ + "## Software and needed installations\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Julia, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we strongly recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn mglearn sympy pandas pillow \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for pyhton2.7)\n", + "\n", + "etc etc." + ] + }, + { + "cell_type": "markdown", + "id": "404f8680", + "metadata": { + "editable": true + }, + "source": [ + "## Python installers\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "Furthermore, [Google's Colab](https://colab.research.google.com/notebooks/welcome.ipynb) is a free Jupyter notebook environment that requires \n", + "no setup and runs entirely in the cloud. Try it out!" + ] + }, + { + "cell_type": "markdown", + "id": "1b74192a", + "metadata": { + "editable": true + }, + "source": [ + "## Useful Python libraries\n", + "Here we list several useful Python libraries we strongly recommend (if you use anaconda many of these are already there)\n", + "\n", + "* [NumPy](https://www.numpy.org/) is a highly popular library for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays\n", + "\n", + "* [The pandas](https://pandas.pydata.org/) library provides high-performance, easy-to-use data structures and data analysis tools \n", + "\n", + "* [Xarray](http://xarray.pydata.org/en/stable/) is a Python package that makes working with labelled multi-dimensional arrays simple, efficient, and fun!\n", + "\n", + "* [Scipy](https://www.scipy.org/) (pronounced “Sigh Pie”) is a Python-based ecosystem of open-source software for mathematics, science, and engineering. \n", + "\n", + "* [Matplotlib](https://matplotlib.org/) is a Python 2D plotting library which produces publication quality figures in a variety of hardcopy formats and interactive environments across platforms.\n", + "\n", + "* [Autograd](https://github.com/HIPS/autograd) can automatically differentiate native Python and Numpy code. It can handle a large subset of Python's features, including loops, ifs, recursion and closures, and it can even take derivatives of derivatives of derivatives\n", + "\n", + "* [JAX](https://jax.readthedocs.io/en/latest/index.html) has now more or less replaced **Autograd**. JAX is Autograd and XLA, brought together for high-performance numerical computing and machine learning research. It provides composable transformations of Python+NumPy programs: differentiate, vectorize, parallelize, Just-In-Time compile to GPU/TPU, and more.\n", + "\n", + "* [SymPy](https://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. \n", + "\n", + "* [scikit-learn](https://scikit-learn.org/stable/) has simple and efficient tools for machine learning, data mining and data analysis\n", + "\n", + "* [TensorFlow](https://www.tensorflow.org/) is a Python library for fast numerical computing created and released by Google\n", + "\n", + "* [Keras](https://keras.io/) is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano\n", + "\n", + "* And many more such as [pytorch](https://pytorch.org/), [Theano](https://pypi.org/project/Theano/) etc" + ] + }, + { + "cell_type": "markdown", + "id": "32b6b09c", + "metadata": { + "editable": true + }, + "source": [ + "## Installing R, C++, cython or Julia\n", + "\n", + "You will also find it convenient to utilize **R**. We will mainly\n", + "use Python during our lectures and in various projects and exercises.\n", + "Those of you\n", + "already familiar with **R** should feel free to continue using **R**, keeping\n", + "however an eye on the parallel Python set ups. Similarly, if you are a\n", + "Python afecionado, feel free to explore **R** as well. Jupyter(Julia, Python and R) /Ipython\n", + "notebook allows you to run **R** codes and **Julia** codes interactively in your\n", + "browser. The software library **R** is really tailored for statistical data analysis\n", + "and allows for an easy usage of the tools and algorithms we will discuss in these\n", + "lectures.\n", + "\n", + "To install **R** with Jupyter notebook \n", + "[follow the link here](https://mpacer.org/maths/r-kernel-for-ipython-notebook)" + ] + }, + { + "cell_type": "markdown", + "id": "0c8344b1", + "metadata": { + "editable": true + }, + "source": [ + "## Installing R, C++, cython, Numba etc\n", + "\n", + "For the C++ aficionados, Jupyter/IPython notebook allows you also to\n", + "install C++ and run codes written in this language interactively in\n", + "the browser. Since we will emphasize writing many of the algorithms\n", + "yourself, you can thus opt for either Python or C++ (or Fortran or other compiled languages) as programming\n", + "languages.\n", + "\n", + "To add more entropy, **cython** can also be used when running your\n", + "notebooks. It means that Python with the jupyter notebook\n", + "setup allows you to integrate widely popular softwares and tools for\n", + "scientific computing. Similarly, the \n", + "[Numba Python package](https://numba.pydata.org/) delivers increased performance\n", + "capabilities with minimal rewrites of your codes. With its\n", + "versatility, including symbolic operations, Python offers a unique\n", + "computational environment. Your jupyter notebook can easily be\n", + "converted into a nicely rendered **PDF** file or a Latex file for\n", + "further processing. For example, convert to latex as" + ] + }, + { + "cell_type": "markdown", + "id": "40a41b04", + "metadata": { + "editable": true + }, + "source": [ + " pycod jupyter nbconvert filename.ipynb --to latex \n" + ] + }, + { + "cell_type": "markdown", + "id": "367b0b81", + "metadata": { + "editable": true + }, + "source": [ + "And to add more versatility, the Python package [SymPy](http://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. It aims to become a full-featured computer algebra system (CAS) and is entirely written in Python. \n", + "\n", + "Finally, we recommend strongly using Autograd or JAX for automatic differentiation." + ] + }, + { + "cell_type": "markdown", + "id": "e16a079a", + "metadata": { + "editable": true + }, + "source": [ + "## Numpy examples and Important Matrix and vector handling packages\n", + "\n", + "There are several central software libraries for linear algebra and eigenvalue problems. Several of the more\n", + "popular ones have been wrapped into ofter software packages like those from the widely used text **Numerical Recipes**. The original source codes in many of the available packages are often taken from the widely used\n", + "software package LAPACK, which follows two other popular packages\n", + "developed in the 1970s, namely EISPACK and LINPACK. We describe them shortly here.\n", + "\n", + " * LINPACK: package for linear equations and least square problems.\n", + "\n", + " * LAPACK:package for solving symmetric, unsymmetric and generalized eigenvalue problems. From LAPACK's website it is possible to download for free all source codes from this library. Both C/C++ and Fortran versions are available.\n", + "\n", + " * BLAS (I, II and III): (Basic Linear Algebra Subprograms) are routines that provide standard building blocks for performing basic vector and matrix operations. Blas I is vector operations, II vector-matrix operations and III matrix-matrix operations. Highly parallelized and efficient codes, all available for download from ." + ] + }, + { + "cell_type": "markdown", + "id": "244cc56a", + "metadata": { + "editable": true + }, + "source": [ + "## Numpy and arrays\n", + "[Numpy](http://www.numpy.org/) provides an easy way to handle arrays in Python. The standard way to import this library is as" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "1d8845bb", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np" + ] + }, + { + "cell_type": "markdown", + "id": "a339aed7", + "metadata": { + "editable": true + }, + "source": [ + "Here follows a simple example where we set up an array of ten elements, all determined by random numbers drawn according to the normal distribution," + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "ce4b3b4e", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "n = 10\n", + "x = np.random.normal(size=n)\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "307170c5", + "metadata": { + "editable": true + }, + "source": [ + "We defined a vector $x$ with $n=10$ elements with its values given by the Normal distribution $N(0,1)$.\n", + "Another alternative is to declare a vector as follows" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "ea4fd74f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "x = np.array([1, 2, 3])\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "6428f55d", + "metadata": { + "editable": true + }, + "source": [ + "Here we have defined a vector with three elements, with $x_0=1$, $x_1=2$ and $x_2=3$. Note that both Python and C++\n", + "start numbering array elements from $0$ and on. This means that a vector with $n$ elements has a sequence of entities $x_0, x_1, x_2, \\dots, x_{n-1}$. We could also let (recommended) Numpy to compute the logarithms of a specific array as" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "84e4a8fc", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4, 7, 8]))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "39bd657f", + "metadata": { + "editable": true + }, + "source": [ + "In the last example we used Numpy's unary function $np.log$. This function is\n", + "highly tuned to compute array elements since the code is vectorized\n", + "and does not require looping. We normaly recommend that you use the\n", + "Numpy intrinsic functions instead of the corresponding **log** function\n", + "from Python's **math** module. The looping is done explicitely by the\n", + "**np.log** function. The alternative, and slower way to compute the\n", + "logarithms of a vector would be to write" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "e5a5fe48", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "from math import log\n", + "x = np.array([4, 7, 8])\n", + "for i in range(0, len(x)):\n", + " x[i] = log(x[i])\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "d74eb874", + "metadata": { + "editable": true + }, + "source": [ + "We note that our code is much longer already and we need to import the **log** function from the **math** module. \n", + "The attentive reader will also notice that the output is $[1, 1, 2]$. Python interprets automagically our numbers as integers (like the **automatic** keyword in C++). To change this we could define our array elements to be double precision numbers as" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "1d657e01", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4, 7, 8], dtype = np.float64))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "7f82621e", + "metadata": { + "editable": true + }, + "source": [ + "or simply write them as double precision numbers (Python uses 64 bits as default for floating point type variables), that is" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "c4a5e02b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4.0, 7.0, 8.0]))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "3af71bb0", + "metadata": { + "editable": true + }, + "source": [ + "To check the number of bytes (remember that one byte contains eight bits for double precision variables), you can use simple use the **itemsize** functionality (the array $x$ is actually an object which inherits the functionalities defined in Numpy) as" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "id": "5cf0baf0", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4.0, 7.0, 8.0]))\n", + "print(x.itemsize)" + ] + }, + { + "cell_type": "markdown", + "id": "a70a4db0", + "metadata": { + "editable": true + }, + "source": [ + "## Matrices in Python\n", + "\n", + "Having defined vectors, we are now ready to try out matrices. We can\n", + "define a $3 \\times 3 $ real matrix $\\boldsymbol{A}$ as (recall that we user\n", + "lowercase letters for vectors and uppercase letters for matrices)" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "262e8f46", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "0f43d9cd", + "metadata": { + "editable": true + }, + "source": [ + "If we use the **shape** function we would get $(3, 3)$ as output, that is verifying that our matrix is a $3\\times 3$ matrix. We can slice the matrix and print for example the first column (Python organized matrix elements in a row-major order, see below) as" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "f12d591b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "# print the first column, row-major order and elements start with 0\n", + "print(A[:,0])" + ] + }, + { + "cell_type": "markdown", + "id": "7c1cc7ed", + "metadata": { + "editable": true + }, + "source": [ + "We can continue this was by printing out other columns or rows. The example here prints out the second column" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "1f365f00", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "# print the first column, row-major order and elements start with 0\n", + "print(A[1,:])" + ] + }, + { + "cell_type": "markdown", + "id": "d5cf7244", + "metadata": { + "editable": true + }, + "source": [ + "Numpy contains many other functionalities that allow us to slice, subdivide etc etc arrays. We strongly recommend that you look up the [Numpy website for more details](http://www.numpy.org/). Useful functions when defining a matrix are the **np.zeros** function which declares a matrix of a given dimension and sets all elements to zero" + ] + }, + { + "cell_type": "code", + "execution_count": 12, + "id": "907260e4", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to zero\n", + "A = np.zeros( (n, n) )\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "375e0579", + "metadata": { + "editable": true + }, + "source": [ + "or initializing all elements to" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "79e19787", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to one\n", + "A = np.ones( (n, n) )\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "be6c16b2", + "metadata": { + "editable": true + }, + "source": [ + "or as unitarily distributed random numbers (see the material on random number generators in the statistics part)" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "901ecfe2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to random numbers with x \\in [0, 1]\n", + "A = np.random.rand(n, n)\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "be629f7b", + "metadata": { + "editable": true + }, + "source": [ + "As we will see throughout these lectures, there are several extremely useful functionalities in Numpy.\n", + "As an example, consider the discussion of the covariance matrix. Suppose we have defined three vectors\n", + "$\\boldsymbol{x}, \\boldsymbol{y}, \\boldsymbol{z}$ with $n$ elements each. The covariance matrix is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "980cf441", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\Sigma} = \\begin{bmatrix} \\sigma_{xx} & \\sigma_{xy} & \\sigma_{xz} \\\\\n", + " \\sigma_{yx} & \\sigma_{yy} & \\sigma_{yz} \\\\\n", + " \\sigma_{zx} & \\sigma_{zy} & \\sigma_{zz} \n", + " \\end{bmatrix},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "d2ffe1f6", + "metadata": { + "editable": true + }, + "source": [ + "where for example" + ] + }, + { + "cell_type": "markdown", + "id": "4f4efdd4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma_{xy} =\\frac{1}{n} \\sum_{i=0}^{n-1}(x_i- \\overline{x})(y_i- \\overline{y}).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ee572849", + "metadata": { + "editable": true + }, + "source": [ + "The Numpy function **np.cov** calculates the covariance elements using the factor $1/(n-1)$ instead of $1/n$ since it assumes we do not have the exact mean values. \n", + "The following simple function uses the **np.vstack** function which takes each vector of dimension $1\\times n$ and produces a $3\\times n$ matrix $\\boldsymbol{W}$" + ] + }, + { + "cell_type": "markdown", + "id": "df0a5acc", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{W} = \\begin{bmatrix} x_0 & x_1 & x_2 & \\dots & x_{n-2} & x_{n-1} \\\\\n", + " y_0 & y_1 & y_2 & \\dots & y_{n-2} & y_{n-1} \\\\\n", + "\t\t\t z_0 & z_1 & z_2 & \\dots & z_{n-2} & z_{n-1} \\\\\n", + " \\end{bmatrix},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7bc72a77", + "metadata": { + "editable": true + }, + "source": [ + "which in turn is converted into into the $3\\times 3$ covariance matrix\n", + "$\\boldsymbol{\\Sigma}$ via the Numpy function **np.cov()**. We note that we can also calculate\n", + "the mean value of each set of samples $\\boldsymbol{x}$ etc using the Numpy\n", + "function **np.mean(x)**. We can also extract the eigenvalues of the\n", + "covariance matrix through the **np.linalg.eig()** function." + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "4c493f06", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Importing various packages\n", + "import numpy as np\n", + "\n", + "n = 100\n", + "x = np.random.normal(size=n)\n", + "print(np.mean(x))\n", + "y = 4+3*x+np.random.normal(size=n)\n", + "print(np.mean(y))\n", + "z = x**3+np.random.normal(size=n)\n", + "print(np.mean(z))\n", + "W = np.vstack((x, y, z))\n", + "Sigma = np.cov(W)\n", + "print(Sigma)\n", + "Eigvals, Eigvecs = np.linalg.eig(Sigma)\n", + "print(Eigvals)" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "20bda99f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "%matplotlib inline\n", + "\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from scipy import sparse\n", + "eye = np.eye(4)\n", + "print(eye)\n", + "sparse_mtx = sparse.csr_matrix(eye)\n", + "print(sparse_mtx)\n", + "x = np.linspace(-10,10,100)\n", + "y = np.sin(x)\n", + "plt.plot(x,y,marker='x')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "a9bff117", + "metadata": { + "editable": true + }, + "source": [ + "## Meet the Pandas\n", + "\n", + "\n", + "\n", + "\n", + "

Figure 1:

\n", + "\n", + "\n", + "Another useful Python package is\n", + "[pandas](https://pandas.pydata.org/), which is an open source library\n", + "providing high-performance, easy-to-use data structures and data\n", + "analysis tools for Python. **pandas** stands for panel data, a term borrowed from econometrics and is an efficient library for data analysis with an emphasis on tabular data.\n", + "**pandas** has two major classes, the **DataFrame** class with two-dimensional data objects and tabular data organized in columns and the class **Series** with a focus on one-dimensional data objects. Both classes allow you to index data easily as we will see in the examples below. \n", + "**pandas** allows you also to perform mathematical operations on the data, spanning from simple reshapings of vectors and matrices to statistical operations. \n", + "\n", + "The following simple example shows how we can, in an easy way make tables of our data. Here we define a data set which includes names, place of birth and date of birth, and displays the data in an easy to read way. We will see repeated use of **pandas**, in particular in connection with classification of data." + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "9ce8cf97", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import pandas as pd\n", + "from IPython.display import display\n", + "data = {'First Name': [\"Frodo\", \"Bilbo\", \"Aragorn II\", \"Samwise\"],\n", + " 'Last Name': [\"Baggins\", \"Baggins\",\"Elessar\",\"Gamgee\"],\n", + " 'Place of birth': [\"Shire\", \"Shire\", \"Eriador\", \"Shire\"],\n", + " 'Date of Birth T.A.': [2968, 2890, 2931, 2980]\n", + " }\n", + "data_pandas = pd.DataFrame(data)\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "98b2cebb", + "metadata": { + "editable": true + }, + "source": [ + "In the above we have imported **pandas** with the shorthand **pd**, the latter has become the standard way we import **pandas**. We make then a list of various variables\n", + "and reorganize the aboves lists into a **DataFrame** and then print out a neat table with specific column labels as *Name*, *place of birth* and *date of birth*.\n", + "Displaying these results, we see that the indices are given by the default numbers from zero to three.\n", + "**pandas** is extremely flexible and we can easily change the above indices by defining a new type of indexing as" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "88a877c3", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "data_pandas = pd.DataFrame(data,index=['Frodo','Bilbo','Aragorn','Sam'])\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "0928de6c", + "metadata": { + "editable": true + }, + "source": [ + "Thereafter we display the content of the row which begins with the index **Aragorn**" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "d93baec5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "display(data_pandas.loc['Aragorn'])" + ] + }, + { + "cell_type": "markdown", + "id": "7a5b03a4", + "metadata": { + "editable": true + }, + "source": [ + "We can easily append data to this, for example" + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "2b5968bf", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "new_hobbit = {'First Name': [\"Peregrin\"],\n", + " 'Last Name': [\"Took\"],\n", + " 'Place of birth': [\"Shire\"],\n", + " 'Date of Birth T.A.': [2990]\n", + " }\n", + "data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin']))\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "281789dd", + "metadata": { + "editable": true + }, + "source": [ + "Here are other examples where we use the **DataFrame** functionality to handle arrays, now with more interesting features for us, namely numbers. We set up a matrix \n", + "of dimensionality $10\\times 5$ and compute the mean value and standard deviation of each column. Similarly, we can perform mathematial operations like squaring the matrix elements and many other operations." + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "226cafd5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "import pandas as pd\n", + "from IPython.display import display\n", + "np.random.seed(100)\n", + "# setting up a 10 x 5 matrix\n", + "rows = 10\n", + "cols = 5\n", + "a = np.random.randn(rows,cols)\n", + "df = pd.DataFrame(a)\n", + "display(df)\n", + "print(df.mean())\n", + "print(df.std())\n", + "display(df**2)" + ] + }, + { + "cell_type": "markdown", + "id": "34d1241d", + "metadata": { + "editable": true + }, + "source": [ + "Thereafter we can select specific columns only and plot final results" + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "id": "66de0f40", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "df.columns = ['First', 'Second', 'Third', 'Fourth', 'Fifth']\n", + "df.index = np.arange(10)\n", + "\n", + "display(df)\n", + "print(df['Second'].mean() )\n", + "print(df.info())\n", + "print(df.describe())\n", + "\n", + "from pylab import plt, mpl\n", + "plt.style.use('seaborn')\n", + "mpl.rcParams['font.family'] = 'serif'\n", + "\n", + "df.cumsum().plot(lw=2.0, figsize=(10,6))\n", + "plt.show()\n", + "\n", + "\n", + "df.plot.bar(figsize=(10,6), rot=15)\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "fce1b00d", + "metadata": { + "editable": true + }, + "source": [ + "We can produce a $4\\times 4$ matrix" + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "id": "b31ec8b2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "b = np.arange(16).reshape((4,4))\n", + "print(b)\n", + "df1 = pd.DataFrame(b)\n", + "print(df1)" + ] + }, + { + "cell_type": "markdown", + "id": "399b4f0b", + "metadata": { + "editable": true + }, + "source": [ + "and many other operations. \n", + "\n", + "The **Series** class is another important class included in\n", + "**pandas**. You can view it as a specialization of **DataFrame** but where\n", + "we have just a single column of data. It shares many of the same features as **DataFrame**. As with **DataFrame**,\n", + "most operations are vectorized, achieving thereby a high performance when dealing with computations of arrays, in particular labeled arrays.\n", + "As we will see below it leads also to a very concice code close to the mathematical operations we may be interested in.\n", + "For multidimensional arrays, we recommend strongly [xarray](http://xarray.pydata.org/en/stable/). **xarray** has much of the same flexibility as **pandas**, but allows for the extension to higher dimensions than two. We will see examples later of the usage of both **pandas** and **xarray**." + ] + }, + { + "cell_type": "markdown", + "id": "b3c66ed4", + "metadata": { + "editable": true + }, + "source": [ + "### Simple linear regression model using **scikit-learn**\n", + "\n", + "We start with perhaps our simplest possible example, using **Scikit-Learn** to perform linear regression analysis on a data set produced by us. \n", + "\n", + "What follows is a simple Python code where we have defined a function\n", + "$y$ in terms of the variable $x$. Both are defined as vectors with $100$ entries. \n", + "The numbers in the vector $\\boldsymbol{x}$ are given\n", + "by random numbers generated with a uniform distribution with entries\n", + "$x_i \\in [0,1]$ (more about probability distribution functions\n", + "later). These values are then used to define a function $y(x)$\n", + "(tabulated again as a vector) with a linear dependence on $x$ plus a\n", + "random noise added via the normal distribution.\n", + "\n", + "The Numpy functions are imported used the **import numpy as np**\n", + "statement and the random number generator for the uniform distribution\n", + "is called using the function **np.random.rand()**, where we specificy\n", + "that we want $100$ random variables. Using Numpy we define\n", + "automatically an array with the specified number of elements, $100$ in\n", + "our case. With the Numpy function **randn()** we can compute random\n", + "numbers with the normal distribution (mean value $\\mu$ equal to zero and\n", + "variance $\\sigma^2$ set to one) and produce the values of $y$ assuming a linear\n", + "dependence as function of $x$" + ] + }, + { + "cell_type": "markdown", + "id": "b7ec1f01", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y = 2x+N(0,1),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "5c4c6423", + "metadata": { + "editable": true + }, + "source": [ + "where $N(0,1)$ represents random numbers generated by the normal\n", + "distribution. From **Scikit-Learn** we import then the\n", + "**LinearRegression** functionality and make a prediction $\\tilde{y} =\n", + "\\alpha + \\beta x$ using the function **fit(x,y)**. We call the set of\n", + "data $(\\boldsymbol{x},\\boldsymbol{y})$ for our training data. The Python package\n", + "**scikit-learn** has also a functionality which extracts the above\n", + "fitting parameters $\\alpha$ and $\\beta$ (see below). Later we will\n", + "distinguish between training data and test data.\n", + "\n", + "For plotting we use the Python package\n", + "[matplotlib](https://matplotlib.org/) which produces publication\n", + "quality figures. Feel free to explore the extensive\n", + "[gallery](https://matplotlib.org/gallery/index.html) of examples. In\n", + "this example we plot our original values of $x$ and $y$ as well as the\n", + "prediction **ypredict** ($\\tilde{y}$), which attempts at fitting our\n", + "data with a straight line.\n", + "\n", + "The Python code follows here." + ] + }, + { + "cell_type": "code", + "execution_count": 24, + "id": "b026b228", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Importing various packages\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 2*x+np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "xnew = np.array([[0],[1]])\n", + "ypredict = linreg.predict(xnew)\n", + "\n", + "plt.plot(xnew, ypredict, \"r-\")\n", + "plt.plot(x, y ,'ro')\n", + "plt.axis([0,1.0,0, 5.0])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$y$')\n", + "plt.title(r'Simple Linear Regression')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "121762a4", + "metadata": { + "editable": true + }, + "source": [ + "This example serves several aims. It allows us to demonstrate several\n", + "aspects of data analysis and later machine learning algorithms. The\n", + "immediate visualization shows that our linear fit is not\n", + "impressive. It goes through the data points, but there are many\n", + "outliers which are not reproduced by our linear regression. We could\n", + "now play around with this small program and change for example the\n", + "factor in front of $x$ and the normal distribution. Try to change the\n", + "function $y$ to" + ] + }, + { + "cell_type": "markdown", + "id": "60728127", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y = 10x+0.01 \\times N(0,1),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "acf5b63f", + "metadata": { + "editable": true + }, + "source": [ + "where $x$ is defined as before. Does the fit look better? Indeed, by\n", + "reducing the role of the noise given by the normal distribution we see immediately that\n", + "our linear prediction seemingly reproduces better the training\n", + "set. However, this testing 'by the eye' is obviouly not satisfactory in the\n", + "long run. Here we have only defined the training data and our model, and \n", + "have not discussed a more rigorous approach to the **cost** function.\n", + "\n", + "We need more rigorous criteria in defining whether we have succeeded or\n", + "not in modeling our training data. You will be surprised to see that\n", + "many scientists seldomly venture beyond this 'by the eye' approach. A\n", + "standard approach for the *cost* function is the so-called $\\chi^2$\n", + "function (a variant of the mean-squared error (MSE))" + ] + }, + { + "cell_type": "markdown", + "id": "40040656", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\chi^2 = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}\\frac{(y_i-\\tilde{y}_i)^2}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "86948166", + "metadata": { + "editable": true + }, + "source": [ + "where $\\sigma_i^2$ is the variance (to be defined later) of the entry\n", + "$y_i$. We may not know the explicit value of $\\sigma_i^2$, it serves\n", + "however the aim of scaling the equations and make the cost function\n", + "dimensionless. \n", + "\n", + "Minimizing the cost function is a central aspect of\n", + "our discussions to come. Finding its minima as function of the model\n", + "parameters ($\\alpha$ and $\\beta$ in our case) will be a recurring\n", + "theme in these series of lectures. Essentially all machine learning\n", + "algorithms we will discuss center around the minimization of the\n", + "chosen cost function. This depends in turn on our specific\n", + "model for describing the data, a typical situation in supervised\n", + "learning. Automatizing the search for the minima of the cost function is a\n", + "central ingredient in all algorithms. Typical methods which are\n", + "employed are various variants of **gradient** methods. These will be\n", + "discussed in more detail later. Again, you'll be surprised to hear that\n", + "many practitioners minimize the above function ''by the eye', popularly dubbed as \n", + "'chi by the eye'. That is, change a parameter and see (visually and numerically) that \n", + "the $\\chi^2$ function becomes smaller. \n", + "\n", + "There are many ways to define the cost function. A simpler approach is to look at the relative difference between the training data and the predicted data, that is we define \n", + "the relative error (why would we prefer the MSE instead of the relative error?) as" + ] + }, + { + "cell_type": "markdown", + "id": "2c7af90a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\epsilon_{\\mathrm{relative}}= \\frac{\\vert \\boldsymbol{y} -\\boldsymbol{\\tilde{y}}\\vert}{\\vert \\boldsymbol{y}\\vert}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "aebc9af3", + "metadata": { + "editable": true + }, + "source": [ + "The squared cost function results in an arithmetic mean-unbiased\n", + "estimator, and the absolute-value cost function results in a\n", + "median-unbiased estimator (in the one-dimensional case, and a\n", + "geometric median-unbiased estimator for the multi-dimensional\n", + "case). The squared cost function has the disadvantage that it has the tendency\n", + "to be dominated by outliers.\n", + "\n", + "We can modify easily the above Python code and plot the relative error instead" + ] + }, + { + "cell_type": "code", + "execution_count": 25, + "id": "eaa61d49", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 5*x+0.01*np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "ypredict = linreg.predict(x)\n", + "\n", + "plt.plot(x, np.abs(ypredict-y)/abs(y), \"ro\")\n", + "plt.axis([0,1.0,0.0, 0.5])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$\\epsilon_{\\mathrm{relative}}$')\n", + "plt.title(r'Relative error')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "f964fde3", + "metadata": { + "editable": true + }, + "source": [ + "Depending on the parameter in front of the normal distribution, we may\n", + "have a small or larger relative error. Try to play around with\n", + "different training data sets and study (graphically) the value of the\n", + "relative error.\n", + "\n", + "As mentioned above, **Scikit-Learn** has an impressive functionality.\n", + "We can for example extract the values of $\\alpha$ and $\\beta$ and\n", + "their error estimates, or the variance and standard deviation and many\n", + "other properties from the statistical data analysis. \n", + "\n", + "Here we show an\n", + "example of the functionality of **Scikit-Learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 26, + "id": "c7fe9388", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np \n", + "import matplotlib.pyplot as plt \n", + "from sklearn.linear_model import LinearRegression \n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_squared_log_error, mean_absolute_error\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 2.0+ 5*x+0.5*np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "ypredict = linreg.predict(x)\n", + "print('The intercept alpha: \\n', linreg.intercept_)\n", + "print('Coefficient beta : \\n', linreg.coef_)\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(y, ypredict))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(y, ypredict))\n", + "# Mean squared log error \n", + "print('Mean squared log error: %.2f' % mean_squared_log_error(y, ypredict) )\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(y, ypredict))\n", + "plt.plot(x, ypredict, \"r-\")\n", + "plt.plot(x, y ,'ro')\n", + "plt.axis([0.0,1.0,1.5, 7.0])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$y$')\n", + "plt.title(r'Linear Regression fit ')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "14ad007a", + "metadata": { + "editable": true + }, + "source": [ + "The function **coef** gives us the parameter $\\beta$ of our fit while **intercept** yields \n", + "$\\alpha$. Depending on the constant in front of the normal distribution, we get values near or far from $\\alpha =2$ and $\\beta =5$. Try to play around with different parameters in front of the normal distribution. The function **meansquarederror** gives us the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error or loss defined as" + ] + }, + { + "cell_type": "markdown", + "id": "7dc68b06", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "471f97ed", + "metadata": { + "editable": true + }, + "source": [ + "The smaller the value, the better the fit. Ideally we would like to\n", + "have an MSE equal zero. The attentive reader has probably recognized\n", + "this function as being similar to the $\\chi^2$ function defined above.\n", + "\n", + "The **r2score** function computes $R^2$, the coefficient of\n", + "determination. It provides a measure of how well future samples are\n", + "likely to be predicted by the model. Best possible score is 1.0 and it\n", + "can be negative (because the model can be arbitrarily worse). A\n", + "constant model that always predicts the expected value of $\\boldsymbol{y}$,\n", + "disregarding the input features, would get a $R^2$ score of $0.0$.\n", + "\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "65a80e19", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "37e1d624", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "a81b3061", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "792fa6a1", + "metadata": { + "editable": true + }, + "source": [ + "Another quantity taht we will meet again in our discussions of regression analysis is \n", + " the mean absolute error (MAE), a risk metric corresponding to the expected value of the absolute error loss or what we call the $l1$-norm loss. In our discussion above we presented the relative error.\n", + "The MAE is defined as follows" + ] + }, + { + "cell_type": "markdown", + "id": "2f6df05c", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\text{MAE}(\\boldsymbol{y}, \\boldsymbol{\\tilde{y}}) = \\frac{1}{n} \\sum_{i=0}^{n-1} \\left| y_i - \\tilde{y}_i \\right|.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "9eb85874", + "metadata": { + "editable": true + }, + "source": [ + "We present the \n", + "squared logarithmic (quadratic) error" + ] + }, + { + "cell_type": "markdown", + "id": "a98d9555", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\text{MSLE}(\\boldsymbol{y}, \\boldsymbol{\\tilde{y}}) = \\frac{1}{n} \\sum_{i=0}^{n - 1} (\\log_e (1 + y_i) - \\log_e (1 + \\tilde{y}_i) )^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6766ecc9", + "metadata": { + "editable": true + }, + "source": [ + "where $\\log_e (x)$ stands for the natural logarithm of $x$. This error\n", + "estimate is best to use when targets having exponential growth, such\n", + "as population counts, average sales of a commodity over a span of\n", + "years etc. \n", + "\n", + "Finally, another cost function is the Huber cost function used in robust regression.\n", + "\n", + "The rationale behind this possible cost function is its reduced\n", + "sensitivity to outliers in the data set. In our discussions on\n", + "dimensionality reduction and normalization of data we will meet other\n", + "ways of dealing with outliers.\n", + "\n", + "The Huber cost function is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "fb3a9871", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "H_{\\delta}(\\boldsymbol{a})=\\left\\{\\begin{array}{cc}\\frac{1}{2} \\boldsymbol{a}^{2}& \\text{for }|\\boldsymbol{a}|\\leq \\delta\\\\ \\delta (|\\boldsymbol{a}|-\\frac{1}{2}\\delta ),&\\text{otherwise}.\\end{array}\\right.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "57226a58", + "metadata": { + "editable": true + }, + "source": [ + "Here $\\boldsymbol{a}=\\boldsymbol{y} - \\boldsymbol{\\tilde{y}}$.\n", + "\n", + "We will discuss in more detail these and other functions in the\n", + "various lectures and lab sessions." + ] + }, + { + "cell_type": "markdown", + "id": "b333e60b", + "metadata": { + "editable": true + }, + "source": [ + "### To our real data: nuclear binding energies. Brief reminder on masses and binding energies\n", + "\n", + "Let us now dive into nuclear physics and remind ourselves briefly about some basic features about binding\n", + "energies. A basic quantity which can be measured for the ground\n", + "states of nuclei is the atomic mass $M(N, Z)$ of the neutral atom with\n", + "atomic mass number $A$ and charge $Z$. The number of neutrons is $N$. There are indeed several sophisticated experiments worldwide which allow us to measure this quantity to high precision (parts per million even). \n", + "\n", + "Atomic masses are usually tabulated in terms of the mass excess defined by" + ] + }, + { + "cell_type": "markdown", + "id": "1adb10e6", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\Delta M(N, Z) = M(N, Z) - uA,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "77592d4e", + "metadata": { + "editable": true + }, + "source": [ + "where $u$ is the Atomic Mass Unit" + ] + }, + { + "cell_type": "markdown", + "id": "ba90405f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "u = M(^{12}\\mathrm{C})/12 = 931.4940954(57) \\hspace{0.1cm} \\mathrm{MeV}/c^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "db7cef69", + "metadata": { + "editable": true + }, + "source": [ + "The nucleon masses are" + ] + }, + { + "cell_type": "markdown", + "id": "e12b557f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "m_p = 1.00727646693(9)u,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6f5462e0", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "0247179e", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "m_n = 939.56536(8)\\hspace{0.1cm} \\mathrm{MeV}/c^2 = 1.0086649156(6)u.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "cb6e7ad5", + "metadata": { + "editable": true + }, + "source": [ + "In the [2016 mass evaluation of by W.J.Huang, G.Audi, M.Wang, F.G.Kondev, S.Naimi and X.Xu](http://nuclearmasses.org/resources_folder/Wang_2017_Chinese_Phys_C_41_030003.pdf)\n", + "there are data on masses and decays of 3437 nuclei.\n", + "\n", + "The nuclear binding energy is defined as the energy required to break\n", + "up a given nucleus into its constituent parts of $N$ neutrons and $Z$\n", + "protons. In terms of the atomic masses $M(N, Z)$ the binding energy is\n", + "defined by" + ] + }, + { + "cell_type": "markdown", + "id": "b60eb119", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N, Z) = ZM_H c^2 + Nm_n c^2 - M(N, Z)c^2 ,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b38eb451", + "metadata": { + "editable": true + }, + "source": [ + "where $M_H$ is the mass of the hydrogen atom and $m_n$ is the mass of the neutron.\n", + "In terms of the mass excess the binding energy is given by" + ] + }, + { + "cell_type": "markdown", + "id": "d67cf525", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N, Z) = Z\\Delta_H c^2 + N\\Delta_n c^2 -\\Delta(N, Z)c^2 ,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2f8449b2", + "metadata": { + "editable": true + }, + "source": [ + "where $\\Delta_H c^2 = 7.2890$ MeV and $\\Delta_n c^2 = 8.0713$ MeV.\n", + "\n", + "A popular and physically intuitive model which can be used to parametrize \n", + "the experimental binding energies as function of $A$, is the so-called \n", + "**liquid drop model**. The ansatz is based on the following expression" + ] + }, + { + "cell_type": "markdown", + "id": "55ca6490", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N,Z) = a_1A-a_2A^{2/3}-a_3\\frac{Z^2}{A^{1/3}}-a_4\\frac{(N-Z)^2}{A},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e716655b", + "metadata": { + "editable": true + }, + "source": [ + "where $A$ stands for the number of nucleons and the $a_i$s are parameters which are determined by a fit \n", + "to the experimental data. \n", + "\n", + "To arrive at the above expression we have assumed that we can make the following assumptions:\n", + "\n", + " * There is a volume term $a_1A$ proportional with the number of nucleons (the energy is also an extensive quantity). When an assembly of nucleons of the same size is packed together into the smallest volume, each interior nucleon has a certain number of other nucleons in contact with it. This contribution is proportional to the volume.\n", + "\n", + " * There is a surface energy term $a_2A^{2/3}$. The assumption here is that a nucleon at the surface of a nucleus interacts with fewer other nucleons than one in the interior of the nucleus and hence its binding energy is less. This surface energy term takes that into account and is therefore negative and is proportional to the surface area.\n", + "\n", + " * There is a Coulomb energy term $a_3\\frac{Z^2}{A^{1/3}}$. The electric repulsion between each pair of protons in a nucleus yields less binding. \n", + "\n", + " * There is an asymmetry term $a_4\\frac{(N-Z)^2}{A}$. This term is associated with the Pauli exclusion principle and reflects the fact that the proton-neutron interaction is more attractive on the average than the neutron-neutron and proton-proton interactions.\n", + "\n", + "We could also add a so-called pairing term, which is a correction term that\n", + "arises from the tendency of proton pairs and neutron pairs to\n", + "occur. An even number of particles is more stable than an odd number." + ] + }, + { + "cell_type": "markdown", + "id": "fa63a9de", + "metadata": { + "editable": true + }, + "source": [ + "### Organizing our data\n", + "\n", + "Let us start with reading and organizing our data. \n", + "We start with the compilation of masses and binding energies from 2016.\n", + "After having downloaded this file to our own computer, we are now ready to read the file and start structuring our data.\n", + "\n", + "We start with preparing folders for storing our calculations and the data file over masses and binding energies. We import also various modules that we will find useful in order to present various Machine Learning methods. Here we focus mainly on the functionality of **scikit-learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 27, + "id": "cca448dd", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "import sklearn.linear_model as skl\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"MassEval2016.dat\"),'r')" + ] + }, + { + "cell_type": "markdown", + "id": "aa35aba5", + "metadata": { + "editable": true + }, + "source": [ + "Before we proceed, we define also a function for making our plots. You can obviously avoid this and simply set up various **matplotlib** commands every time you need them. You may however find it convenient to collect all such commands in one function and simply call this function." + ] + }, + { + "cell_type": "code", + "execution_count": 28, + "id": "c854893a", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "from pylab import plt, mpl\n", + "plt.style.use('seaborn')\n", + "mpl.rcParams['font.family'] = 'serif'\n", + "\n", + "def MakePlot(x,y, styles, labels, axlabels):\n", + " plt.figure(figsize=(10,6))\n", + " for i in range(len(x)):\n", + " plt.plot(x[i], y[i], styles[i], label = labels[i])\n", + " plt.xlabel(axlabels[0])\n", + " plt.ylabel(axlabels[1])\n", + " plt.legend(loc=0)" + ] + }, + { + "cell_type": "markdown", + "id": "74b3f235", + "metadata": { + "editable": true + }, + "source": [ + "Our next step is to read the data on experimental binding energies and\n", + "reorganize them as functions of the mass number $A$, the number of\n", + "protons $Z$ and neutrons $N$ using **pandas**. Before we do this it is\n", + "always useful (unless you have a binary file or other types of compressed\n", + "data) to actually open the file and simply take a look at it!\n", + "\n", + "In particular, the program that outputs the final nuclear masses is written in Fortran with a specific format. It means that we need to figure out the format and which columns contain the data we are interested in. Pandas comes with a function that reads formatted output. After having admired the file, we are now ready to start massaging it with **pandas**. The file begins with some basic format information." + ] + }, + { + "cell_type": "code", + "execution_count": 29, + "id": "92e8847b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "\"\"\" \n", + "This is taken from the data file of the mass 2016 evaluation. \n", + "All files are 3436 lines long with 124 character per line. \n", + " Headers are 39 lines long. \n", + " col 1 : Fortran character control: 1 = page feed 0 = line feed \n", + " format : a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5 \n", + " These formats are reflected in the pandas widths variable below, see the statement \n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), \n", + " Pandas has also a variable header, with length 39 in this case. \n", + "\"\"\"" + ] + }, + { + "cell_type": "markdown", + "id": "8cbce288", + "metadata": { + "editable": true + }, + "source": [ + "The data we are interested in are in columns 2, 3, 4 and 11, giving us\n", + "the number of neutrons, protons, mass numbers and binding energies,\n", + "respectively. We add also for the sake of completeness the element name. The data are in fixed-width formatted lines and we will\n", + "covert them into the **pandas** DataFrame structure." + ] + }, + { + "cell_type": "code", + "execution_count": 30, + "id": "b4398581", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Read the experimental data with Pandas\n", + "Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),\n", + " names=('N', 'Z', 'A', 'Element', 'Ebinding'),\n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),\n", + " header=39,\n", + " index_col=False)\n", + "\n", + "# Extrapolated values are indicated by '#' in place of the decimal place, so\n", + "# the Ebinding column won't be numeric. Coerce to float and drop these entries.\n", + "Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')\n", + "Masses = Masses.dropna()\n", + "# Convert from keV to MeV.\n", + "Masses['Ebinding'] /= 1000\n", + "\n", + "# Group the DataFrame by nucleon number, A.\n", + "Masses = Masses.groupby('A')\n", + "# Find the rows of the grouped DataFrame with the maximum binding energy.\n", + "Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])" + ] + }, + { + "cell_type": "markdown", + "id": "0f21a73f", + "metadata": { + "editable": true + }, + "source": [ + "We have now read in the data, grouped them according to the variables we are interested in. \n", + "We see how easy it is to reorganize the data using **pandas**. If we\n", + "were to do these operations in C/C++ or Fortran, we would have had to\n", + "write various functions/subroutines which perform the above\n", + "reorganizations for us. Having reorganized the data, we can now start\n", + "to make some simple fits using both the functionalities in **numpy** and\n", + "**Scikit-Learn** afterwards. \n", + "\n", + "Now we define five variables which contain\n", + "the number of nucleons $A$, the number of protons $Z$ and the number of neutrons $N$, the element name and finally the energies themselves." + ] + }, + { + "cell_type": "code", + "execution_count": 31, + "id": "491e38a4", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "A = Masses['A']\n", + "Z = Masses['Z']\n", + "N = Masses['N']\n", + "Element = Masses['Element']\n", + "Energies = Masses['Ebinding']\n", + "print(Masses)" + ] + }, + { + "cell_type": "markdown", + "id": "07ac2cd3", + "metadata": { + "editable": true + }, + "source": [ + "The next step, and we will define this mathematically later, is to set up the so-called **design matrix**. We will throughout call this matrix $\\boldsymbol{X}$.\n", + "It has dimensionality $p\\times n$, where $n$ is the number of data points and $p$ are the so-called predictors. In our case here they are given by the number of polynomials in $A$ we wish to include in the fit." + ] + }, + { + "cell_type": "code", + "execution_count": 32, + "id": "0462ebdd", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Now we set up the design matrix X\n", + "X = np.zeros((len(A),5))\n", + "X[:,0] = 1\n", + "X[:,1] = A\n", + "X[:,2] = A**(2.0/3.0)\n", + "X[:,3] = A**(-1.0/3.0)\n", + "X[:,4] = A**(-1.0)" + ] + }, + { + "cell_type": "markdown", + "id": "66a31f58", + "metadata": { + "editable": true + }, + "source": [ + "With **scikitlearn** we are now ready to use linear regression and fit our data." + ] + }, + { + "cell_type": "code", + "execution_count": 33, + "id": "7d4bf033", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "clf = skl.LinearRegression().fit(X, Energies)\n", + "fity = clf.predict(X)" + ] + }, + { + "cell_type": "markdown", + "id": "3b9f86de", + "metadata": { + "editable": true + }, + "source": [ + "Pretty simple! \n", + "Now we can print measures of how our fit is doing, the coefficients from the fits and plot the final fit together with our data." + ] + }, + { + "cell_type": "code", + "execution_count": 34, + "id": "b12b4d19", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, fity))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, fity))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, fity))\n", + "print(clf.coef_, clf.intercept_)\n", + "\n", + "Masses['Eapprox'] = fity\n", + "# Generate a plot comparing the experimental with the fitted values values.\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$A = N + Z$')\n", + "ax.set_ylabel(r'$E_\\mathrm{bind}\\,/\\mathrm{MeV}$')\n", + "ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,\n", + " label='Ame2016')\n", + "ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',\n", + " label='Fit')\n", + "ax.legend()\n", + "save_fig(\"Masses2016\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "4cf1be8b", + "metadata": { + "editable": true + }, + "source": [ + "### And what about using neural networks?\n", + "\n", + "The **seaborn** package allows us to visualize data in an efficient way. Note that we use **scikit-learn**'s multi-layer perceptron (or feed forward neural network) \n", + "functionality." + ] + }, + { + "cell_type": "code", + "execution_count": 35, + "id": "e80600eb", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "from sklearn.neural_network import MLPRegressor\n", + "from sklearn.metrics import accuracy_score\n", + "import seaborn as sns\n", + "\n", + "X_train = X\n", + "Y_train = Energies\n", + "n_hidden_neurons = 100\n", + "epochs = 100\n", + "# store models for later use\n", + "eta_vals = np.logspace(-5, 1, 7)\n", + "lmbd_vals = np.logspace(-5, 1, 7)\n", + "# store the models for later use\n", + "DNN_scikit = np.zeros((len(eta_vals), len(lmbd_vals)), dtype=object)\n", + "train_accuracy = np.zeros((len(eta_vals), len(lmbd_vals)))\n", + "sns.set()\n", + "for i, eta in enumerate(eta_vals):\n", + " for j, lmbd in enumerate(lmbd_vals):\n", + " dnn = MLPRegressor(hidden_layer_sizes=(n_hidden_neurons), activation='logistic',\n", + " alpha=lmbd, learning_rate_init=eta, max_iter=epochs)\n", + " dnn.fit(X_train, Y_train)\n", + " DNN_scikit[i][j] = dnn\n", + " train_accuracy[i][j] = dnn.score(X_train, Y_train)\n", + "\n", + "fig, ax = plt.subplots(figsize = (10, 10))\n", + "sns.heatmap(train_accuracy, annot=True, ax=ax, cmap=\"viridis\")\n", + "ax.set_title(\"Training Accuracy\")\n", + "ax.set_ylabel(\"$\\eta$\")\n", + "ax.set_xlabel(\"$\\lambda$\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "d49cd4ec", + "metadata": { + "editable": true + }, + "source": [ + "## A first summary\n", + "\n", + "The aim behind these introductory words was to present to you various\n", + "Python libraries and their functionalities, in particular libraries like\n", + "**numpy**, **pandas**, **xarray** and **matplotlib** and other that make our life much easier\n", + "in handling various data sets and visualizing data. \n", + "\n", + "Furthermore,\n", + "**Scikit-Learn** allows us with few lines of code to implement popular\n", + "Machine Learning algorithms for supervised learning. Later we will meet **Tensorflow**, a powerful library for deep learning. \n", + "Now it is time to dive more into the details of various methods. We will start with linear regression and try to take a deeper look at what it entails." + ] + }, + { + "cell_type": "markdown", + "id": "9a4706c7", + "metadata": { + "editable": true + }, + "source": [ + "## Why Linear Regression (aka Ordinary Least Squares and family)\n", + "\n", + "Fitting a continuous function with linear parameterization in terms of the parameters $\\boldsymbol{\\beta}$.\n", + "* Method of choice for fitting a continuous function!\n", + "\n", + "* Gives an excellent introduction to central Machine Learning features with **understandable pedagogical** links to other methods like **Neural Networks**, **Support Vector Machines** etc\n", + "\n", + "* Analytical expression for the fitting parameters $\\boldsymbol{\\beta}$\n", + "\n", + "* Analytical expressions for statistical propertiers like mean values, variances, confidence intervals and more\n", + "\n", + "* Analytical relation with probabilistic interpretations \n", + "\n", + "* Easy to introduce basic concepts like bias-variance tradeoff, cross-validation, resampling and regularization techniques and many other ML topics\n", + "\n", + "* Easy to code! And links well with classification problems and logistic regression and neural networks\n", + "\n", + "* Allows for **easy** hands-on understanding of gradient descent methods\n", + "\n", + "* and many more features\n", + "\n", + "For more discussions of Ridge and Lasso regression, [Wessel van Wieringen's](https://arxiv.org/abs/1509.09169) article is highly recommended.\n", + "Similarly, [Mehta et al's article](https://arxiv.org/abs/1803.08823) is also recommended." + ] + }, + { + "cell_type": "markdown", + "id": "b720efdc", + "metadata": { + "editable": true + }, + "source": [ + "## Regression analysis, overarching aims\n", + "\n", + "Regression modeling deals with the description of the sampling distribution of a given random variable $y$ and how it varies as function of another variable or a set of such variables $\\boldsymbol{x} =[x_0, x_1,\\dots, x_{n-1}]^T$. \n", + "The first variable is called the **dependent**, the **outcome** or the **response** variable while the set of variables $\\boldsymbol{x}$ is called the independent variable, or the predictor variable or the explanatory variable, or simply just the **inputs**. \n", + "\n", + "A regression model aims at finding a likelihood function $p(\\boldsymbol{y}\\vert \\boldsymbol{x})$ or in the more traditional sense a function $\\boldsymbol{y}(\\boldsymbol{x})$, that is the conditional distribution for $\\boldsymbol{y}$ with a given $\\boldsymbol{x}$. The estimation of $p(\\boldsymbol{y}\\vert \\boldsymbol{x})$ is made using a data set with \n", + "* $n$ cases $i = 0, 1, 2, \\dots, n-1$ \n", + "\n", + "* Response (target, dependent or outcome) variable $y_i$ with $i = 0, 1, 2, \\dots, n-1$ \n", + "\n", + "* $p$ so-called explanatory (independent or predictor or feature) variables $\\boldsymbol{x}_i=[x_{i0}, x_{i1}, \\dots, x_{ip-1}]$ with $i = 0, 1, 2, \\dots, n-1$ and explanatory variables running from $0$ to $p-1$. See below for more explicit examples. \n", + "\n", + " The goal of the regression analysis is to extract/exploit relationship between $\\boldsymbol{y}$ and $\\boldsymbol{x}$ in order to infer specific dependencies, approximations to the likelihood functions, functional relationships and to make predictions, making fits and many other things." + ] + }, + { + "cell_type": "markdown", + "id": "0cb90bb5", + "metadata": { + "editable": true + }, + "source": [ + "## Regression analysis, overarching aims II\n", + "\n", + "Consider an experiment in which $p$ characteristics/features of $n$ samples are\n", + "measured. The data from this experiment, for various explanatory variables $p$ are normally represented by a matrix \n", + "$\\mathbf{X}$.\n", + "\n", + "The matrix $\\mathbf{X}$ is called the *design\n", + "matrix*. Additional information of the samples is available in the\n", + "form of $\\boldsymbol{y}$ (also as above). The variable $\\boldsymbol{y}$ is\n", + "generally referred to as the *response variable*. The aim of\n", + "regression analysis is to explain $\\boldsymbol{y}$ in terms of\n", + "$\\boldsymbol{X}$ through a functional relationship like $y_i =\n", + "f(\\mathbf{X}_{i,\\ast})$. When no prior knowledge on the form of\n", + "$f(\\cdot)$ is available, it is common to assume a linear relationship\n", + "between $\\boldsymbol{X}$ and $\\boldsymbol{y}$. This assumption gives rise to\n", + "the *linear regression model* where $\\boldsymbol{\\beta} = [\\beta_0, \\ldots,\n", + "\\beta_{p-1}]^{T}$ are the *regression parameters*. \n", + "\n", + "Linear regression gives us a set of analytical equations for the parameters $\\beta_j$." + ] + }, + { + "cell_type": "markdown", + "id": "e45e89d3", + "metadata": { + "editable": true + }, + "source": [ + "## Examples\n", + "In order to understand the relation among the predictors (or features or properties) $p$, the set of data $n$ and the target (outcome, output etc) $\\boldsymbol{y}$,\n", + "consider the model we discussed for describing nuclear binding energies. \n", + "\n", + "There we assumed that we could parametrize the data using a polynomial approximation based on the liquid drop model.\n", + "Assuming" + ] + }, + { + "cell_type": "markdown", + "id": "f2471451", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(A) = a_0+a_1A+a_2A^{2/3}+a_3A^{-1/3}+a_4A^{-1},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ee20b9ee", + "metadata": { + "editable": true + }, + "source": [ + "we have five predictors, that is the intercept, the $A$ dependent term, the $A^{2/3}$ term and the $A^{-1/3}$ and $A^{-1}$ terms.\n", + "This gives $p=0,1,2,3,4$. Furthermore we have $n$ entries for each predictor. It means that our design matrix is a \n", + "$p\\times n$ matrix $\\boldsymbol{X}$.\n", + "\n", + "Here the predictors are based on a model we have made. A popular data set which is widely encountered in ML applications is the\n", + "so-called [credit card default data from Taiwan](https://www.sciencedirect.com/science/article/pii/S0957417407006719?via%3Dihub). The data set contains data on $n=30000$ credit card holders with predictors like gender, marital status, age, profession, education, etc. In total there are $24$ such predictors or attributes leading to a design matrix of dimensionality $24 \\times 30000$. This is however a classification problem and we will come back to it when we discuss Logistic Regression." + ] + }, + { + "cell_type": "markdown", + "id": "ebc12338", + "metadata": { + "editable": true + }, + "source": [ + "## General linear models and linear algebra\n", + "Before we proceed let us study a case where we aim at fitting a set of data $\\boldsymbol{y}=[y_0,y_1,\\dots,y_{n-1}]$. We could think of these data as a result of an experiment or a complicated numerical experiment. These data are functions of a series of variables $\\boldsymbol{x}=[x_0,x_1,\\dots,x_{n-1}]$, that is $y_i = y(x_i)$ with $i=0,1,2,\\dots,n-1$. The variables $x_i$ could represent physical quantities like time, temperature, position etc. We assume that $y(x)$ is a smooth function. \n", + "\n", + "Since obtaining these data points may not be trivial, we want to use these data to fit a function which can allow us to make predictions for values of $y$ which are not in the present set. The perhaps simplest approach is to assume we can parametrize our function in terms of a polynomial of degree $n-1$ with $n$ points, that is" + ] + }, + { + "cell_type": "markdown", + "id": "0f5cb3d7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y=y(x) \\rightarrow y(x_i)=\\tilde{y}_i+\\epsilon_i=\\sum_{j=0}^{n-1} \\beta_j x_i^j+\\epsilon_i,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8be293e3", + "metadata": { + "editable": true + }, + "source": [ + "where $\\epsilon_i$ is the error in our approximation." + ] + }, + { + "cell_type": "markdown", + "id": "c0a2104e", + "metadata": { + "editable": true + }, + "source": [ + "## Rewriting the fitting procedure as a linear algebra problem\n", + "For every set of values $y_i,x_i$ we have thus the corresponding set of equations" + ] + }, + { + "cell_type": "markdown", + "id": "93b447fd", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0+\\beta_1x_0^1+\\beta_2x_0^2+\\dots+\\beta_{n-1}x_0^{n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0+\\beta_1x_1^1+\\beta_2x_1^2+\\dots+\\beta_{n-1}x_1^{n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0+\\beta_1x_2^1+\\beta_2x_2^2+\\dots+\\beta_{n-1}x_2^{n-1}+\\epsilon_2\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0+\\beta_1x_{n-1}^1+\\beta_2x_{n-1}^2+\\dots+\\beta_{n-1}x_{n-1}^{n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "94d59403", + "metadata": { + "editable": true + }, + "source": [ + "## Rewriting the fitting procedure as a linear algebra problem, more details\n", + "Defining the vectors" + ] + }, + { + "cell_type": "markdown", + "id": "7b044c26", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = [y_0,y_1, y_2,\\dots, y_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a4ec8b3e", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "7e221f81", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} = [\\beta_0,\\beta_1, \\beta_2,\\dots, \\beta_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "729c5dd3", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "5eb54f13", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\epsilon} = [\\epsilon_0,\\epsilon_1, \\epsilon_2,\\dots, \\epsilon_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "51cad783", + "metadata": { + "editable": true + }, + "source": [ + "and the design matrix" + ] + }, + { + "cell_type": "markdown", + "id": "e38a17c1", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}=\n", + "\\begin{bmatrix} \n", + "1& x_{0}^1 &x_{0}^2& \\dots & \\dots &x_{0}^{n-1}\\\\\n", + "1& x_{1}^1 &x_{1}^2& \\dots & \\dots &x_{1}^{n-1}\\\\\n", + "1& x_{2}^1 &x_{2}^2& \\dots & \\dots &x_{2}^{n-1}\\\\ \n", + "\\dots& \\dots &\\dots& \\dots & \\dots &\\dots\\\\\n", + "1& x_{n-1}^1 &x_{n-1}^2& \\dots & \\dots &x_{n-1}^{n-1}\\\\\n", + "\\end{bmatrix}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c669704a", + "metadata": { + "editable": true + }, + "source": [ + "we can rewrite our equations as" + ] + }, + { + "cell_type": "markdown", + "id": "6c2f1143", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = \\boldsymbol{X}\\boldsymbol{\\beta}+\\boldsymbol{\\epsilon}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "144ef4c2", + "metadata": { + "editable": true + }, + "source": [ + "The above design matrix is called a [Vandermonde matrix](https://en.wikipedia.org/wiki/Vandermonde_matrix)." + ] + }, + { + "cell_type": "markdown", + "id": "d9da387d", + "metadata": { + "editable": true + }, + "source": [ + "## Generalizing the fitting procedure as a linear algebra problem\n", + "\n", + "We are obviously not limited to the above polynomial expansions. We\n", + "could replace the various powers of $x$ with elements of Fourier\n", + "series or instead of $x_i^j$ we could have $\\cos{(j x_i)}$ or $\\sin{(j\n", + "x_i)}$, or time series or other orthogonal functions. For every set\n", + "of values $y_i,x_i$ we can then generalize the equations to" + ] + }, + { + "cell_type": "markdown", + "id": "16d9099a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0x_{00}+\\beta_1x_{01}+\\beta_2x_{02}+\\dots+\\beta_{n-1}x_{0n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0x_{10}+\\beta_1x_{11}+\\beta_2x_{12}+\\dots+\\beta_{n-1}x_{1n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0x_{20}+\\beta_1x_{21}+\\beta_2x_{22}+\\dots+\\beta_{n-1}x_{2n-1}+\\epsilon_2\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{i}&=\\beta_0x_{i0}+\\beta_1x_{i1}+\\beta_2x_{i2}+\\dots+\\beta_{n-1}x_{in-1}+\\epsilon_i\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0x_{n-1,0}+\\beta_1x_{n-1,2}+\\beta_2x_{n-1,2}+\\dots+\\beta_{n-1}x_{n-1,n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "dac4b0d4", + "metadata": { + "editable": true + }, + "source": [ + "**Note that we have $p=n$ here. The matrix is symmetric. This is generally not the case!**" + ] + }, + { + "cell_type": "markdown", + "id": "821d16ee", + "metadata": { + "editable": true + }, + "source": [ + "## Generalizing the fitting procedure as a linear algebra problem\n", + "We redefine in turn the matrix $\\boldsymbol{X}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "64804e58", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}=\n", + "\\begin{bmatrix} \n", + "x_{00}& x_{01} &x_{02}& \\dots & \\dots &x_{0,n-1}\\\\\n", + "x_{10}& x_{11} &x_{12}& \\dots & \\dots &x_{1,n-1}\\\\\n", + "x_{20}& x_{21} &x_{22}& \\dots & \\dots &x_{2,n-1}\\\\ \n", + "\\dots& \\dots &\\dots& \\dots & \\dots &\\dots\\\\\n", + "x_{n-1,0}& x_{n-1,1} &x_{n-1,2}& \\dots & \\dots &x_{n-1,n-1}\\\\\n", + "\\end{bmatrix}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "28be47df", + "metadata": { + "editable": true + }, + "source": [ + "and without loss of generality we rewrite again our equations as" + ] + }, + { + "cell_type": "markdown", + "id": "f15f39a0", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = \\boldsymbol{X}\\boldsymbol{\\beta}+\\boldsymbol{\\epsilon}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "03117774", + "metadata": { + "editable": true + }, + "source": [ + "The left-hand side of this equation is kwown. Our error vector $\\boldsymbol{\\epsilon}$ and the parameter vector $\\boldsymbol{\\beta}$ are our unknow quantities. How can we obtain the optimal set of $\\beta_i$ values?" + ] + }, + { + "cell_type": "markdown", + "id": "2dadc99e", + "metadata": { + "editable": true + }, + "source": [ + "## Optimizing our parameters\n", + "We have defined the matrix $\\boldsymbol{X}$ via the equations" + ] + }, + { + "cell_type": "markdown", + "id": "98d081e3", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0x_{00}+\\beta_1x_{01}+\\beta_2x_{02}+\\dots+\\beta_{n-1}x_{0n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0x_{10}+\\beta_1x_{11}+\\beta_2x_{12}+\\dots+\\beta_{n-1}x_{1n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0x_{20}+\\beta_1x_{21}+\\beta_2x_{22}+\\dots+\\beta_{n-1}x_{2n-1}+\\epsilon_1\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{i}&=\\beta_0x_{i0}+\\beta_1x_{i1}+\\beta_2x_{i2}+\\dots+\\beta_{n-1}x_{in-1}+\\epsilon_1\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0x_{n-1,0}+\\beta_1x_{n-1,2}+\\beta_2x_{n-1,2}+\\dots+\\beta_{n-1}x_{n-1,n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "02c207e3", + "metadata": { + "editable": true + }, + "source": [ + "As we noted above, we stayed with a system with the design matrix \n", + " $\\boldsymbol{X}\\in {\\mathbb{R}}^{n\\times n}$, that is we have $p=n$. For reasons to come later (algorithmic arguments) we will hereafter define \n", + "our matrix as $\\boldsymbol{X}\\in {\\mathbb{R}}^{n\\times p}$, with the predictors refering to the column numbers and the entries $n$ being the row elements." + ] + }, + { + "cell_type": "markdown", + "id": "5b3ba5aa", + "metadata": { + "editable": true + }, + "source": [ + "## Our model for the nuclear binding energies\n", + "\n", + "In our [introductory notes](https://compphysics.github.io/MachineLearning/doc/pub/How2ReadData/html/How2ReadData.html) we looked at the so-called [liquid drop model](https://en.wikipedia.org/wiki/Semi-empirical_mass_formula). Let us remind ourselves about what we did by looking at the code.\n", + "\n", + "We restate the parts of the code we are most interested in." + ] + }, + { + "cell_type": "code", + "execution_count": 36, + "id": "97a71b30", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "from IPython.display import display\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"MassEval2016.dat\"),'r')\n", + "\n", + "\n", + "# Read the experimental data with Pandas\n", + "Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),\n", + " names=('N', 'Z', 'A', 'Element', 'Ebinding'),\n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),\n", + " header=39,\n", + " index_col=False)\n", + "\n", + "# Extrapolated values are indicated by '#' in place of the decimal place, so\n", + "# the Ebinding column won't be numeric. Coerce to float and drop these entries.\n", + "Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')\n", + "Masses = Masses.dropna()\n", + "# Convert from keV to MeV.\n", + "Masses['Ebinding'] /= 1000\n", + "\n", + "# Group the DataFrame by nucleon number, A.\n", + "Masses = Masses.groupby('A')\n", + "# Find the rows of the grouped DataFrame with the maximum binding energy.\n", + "Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])\n", + "A = Masses['A']\n", + "Z = Masses['Z']\n", + "N = Masses['N']\n", + "Element = Masses['Element']\n", + "Energies = Masses['Ebinding']\n", + "\n", + "# Now we set up the design matrix X\n", + "X = np.zeros((len(A),5))\n", + "X[:,0] = 1\n", + "X[:,1] = A\n", + "X[:,2] = A**(2.0/3.0)\n", + "X[:,3] = A**(-1.0/3.0)\n", + "X[:,4] = A**(-1.0)\n", + "# Then nice printout using pandas\n", + "DesignMatrix = pd.DataFrame(X)\n", + "DesignMatrix.index = A\n", + "DesignMatrix.columns = ['1', 'A', 'A^(2/3)', 'A^(-1/3)', '1/A']\n", + "display(DesignMatrix)" + ] + }, + { + "cell_type": "markdown", + "id": "ed7eb3b8", + "metadata": { + "editable": true + }, + "source": [ + "With $\\boldsymbol{\\beta}\\in {\\mathbb{R}}^{p\\times 1}$, it means that we will hereafter write our equations for the approximation as" + ] + }, + { + "cell_type": "markdown", + "id": "1ac7d0e5", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\tilde{y}}= \\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8069a90f", + "metadata": { + "editable": true + }, + "source": [ + "throughout these lectures." + ] + }, + { + "cell_type": "markdown", + "id": "815493e0", + "metadata": { + "editable": true + }, + "source": [ + "## Optimizing our parameters, more details\n", + "With the above we use the design matrix to define the approximation $\\boldsymbol{\\tilde{y}}$ via the unknown quantity $\\boldsymbol{\\beta}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "7880ef32", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\tilde{y}}= \\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ecb0a16f", + "metadata": { + "editable": true + }, + "source": [ + "and in order to find the optimal parameters $\\beta_i$ instead of solving the above linear algebra problem, we define a function which gives a measure of the spread between the values $y_i$ (which represent hopefully the exact values) and the parameterized values $\\tilde{y}_i$, namely" + ] + }, + { + "cell_type": "markdown", + "id": "d8782aad", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\sum_{i=0}^{n-1}\\left(y_i-\\tilde{y}_i\\right)^2=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "d2491d40", + "metadata": { + "editable": true + }, + "source": [ + "or using the matrix $\\boldsymbol{X}$ and in a more compact matrix-vector notation as" + ] + }, + { + "cell_type": "markdown", + "id": "1cf22842", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "71180218", + "metadata": { + "editable": true + }, + "source": [ + "This function is one possible way to define the so-called cost function.\n", + "\n", + "It is also common to define\n", + "the function $C$ as" + ] + }, + { + "cell_type": "markdown", + "id": "1c734def", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{2n}\\sum_{i=0}^{n-1}\\left(y_i-\\tilde{y}_i\\right)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "85f6ff9d", + "metadata": { + "editable": true + }, + "source": [ + "since when taking the first derivative with respect to the unknown parameters $\\beta$, the factor of $2$ cancels out." + ] + }, + { + "cell_type": "markdown", + "id": "1ad66e97", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "\n", + "The function" + ] + }, + { + "cell_type": "markdown", + "id": "6ff8d32a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "fde1cca5", + "metadata": { + "editable": true + }, + "source": [ + "can be linked to the variance of the quantity $y_i$ if we interpret the latter as the mean value. \n", + "When linking (see the discussion below) with the maximum likelihood approach below, we will indeed interpret $y_i$ as a mean value" + ] + }, + { + "cell_type": "markdown", + "id": "5b34fa19", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y_{i}=\\langle y_i \\rangle = \\beta_0x_{i,0}+\\beta_1x_{i,1}+\\beta_2x_{i,2}+\\dots+\\beta_{n-1}x_{i,n-1}+\\epsilon_i,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e93804b1", + "metadata": { + "editable": true + }, + "source": [ + "where $\\langle y_i \\rangle$ is the mean value. Keep in mind also that\n", + "till now we have treated $y_i$ as the exact value. Normally, the\n", + "response (dependent or outcome) variable $y_i$ the outcome of a\n", + "numerical experiment or another type of experiment and is thus only an\n", + "approximation to the true value. It is then always accompanied by an\n", + "error estimate, often limited to a statistical error estimate given by\n", + "the standard deviation discussed earlier. In the discussion here we\n", + "will treat $y_i$ as our exact value for the response variable.\n", + "\n", + "In order to find the parameters $\\beta_i$ we will then minimize the spread of $C(\\boldsymbol{\\beta})$, that is we are going to solve the problem" + ] + }, + { + "cell_type": "markdown", + "id": "e3b2a85a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "{\\displaystyle \\min_{\\boldsymbol{\\beta}\\in\n", + "{\\mathbb{R}}^{p}}}\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "88b484a0", + "metadata": { + "editable": true + }, + "source": [ + "In practical terms it means we will require" + ] + }, + { + "cell_type": "markdown", + "id": "0cae40f7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\beta_j} = \\frac{\\partial }{\\partial \\beta_j}\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}\\right)^2\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3e3c19f7", + "metadata": { + "editable": true + }, + "source": [ + "which results in" + ] + }, + { + "cell_type": "markdown", + "id": "c5d9738a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\beta_j} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}x_{ij}\\left(y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "49bf03c8", + "metadata": { + "editable": true + }, + "source": [ + "or in a matrix-vector form as" + ] + }, + { + "cell_type": "markdown", + "id": "acc52e20", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ecfc4d77", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "We can rewrite" + ] + }, + { + "cell_type": "markdown", + "id": "7bc2250b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3c038483", + "metadata": { + "editable": true + }, + "source": [ + "as" + ] + }, + { + "cell_type": "markdown", + "id": "d530cc8f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\boldsymbol{y} = \\boldsymbol{X}^T\\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "667c132f", + "metadata": { + "editable": true + }, + "source": [ + "and if the matrix $\\boldsymbol{X}^T\\boldsymbol{X}$ is invertible we have the solution" + ] + }, + { + "cell_type": "markdown", + "id": "1a592608", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} =\\left(\\boldsymbol{X}^T\\boldsymbol{X}\\right)^{-1}\\boldsymbol{X}^T\\boldsymbol{y}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "fe968934", + "metadata": { + "editable": true + }, + "source": [ + "We note also that since our design matrix is defined as $\\boldsymbol{X}\\in\n", + "{\\mathbb{R}}^{n\\times p}$, the product $\\boldsymbol{X}^T\\boldsymbol{X} \\in\n", + "{\\mathbb{R}}^{p\\times p}$. In the above case we have that $p \\ll n$,\n", + "in our case $p=5$ meaning that we end up with inverting a small\n", + "$5\\times 5$ matrix. This is a rather common situation, in many cases we end up with low-dimensional\n", + "matrices to invert. The methods discussed here and for many other\n", + "supervised learning algorithms like classification with logistic\n", + "regression or support vector machines, exhibit dimensionalities which\n", + "allow for the usage of direct linear algebra methods such as **LU** decomposition or **Singular Value Decomposition** (SVD) for finding the inverse of the matrix\n", + "$\\boldsymbol{X}^T\\boldsymbol{X}$.\n", + "\n", + "**Small question**: Do you think the example we have at hand here (the nuclear binding energies) can lead to problems in inverting the matrix $\\boldsymbol{X}^T\\boldsymbol{X}$? What kind of problems can we expect?" + ] + }, + { + "cell_type": "markdown", + "id": "f182ccd2", + "metadata": { + "editable": true + }, + "source": [ + "## Some useful matrix and vector expressions\n", + "\n", + "See the handwritten notes at \n", + "\n", + "These notes will be discussed during one of the lectures." + ] + }, + { + "cell_type": "markdown", + "id": "3e23c6ac", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "The residuals $\\boldsymbol{\\epsilon}$ are in turn given by" + ] + }, + { + "cell_type": "markdown", + "id": "2a10b1fa", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\epsilon} = \\boldsymbol{y}-\\boldsymbol{\\tilde{y}} = \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "753ac3da", + "metadata": { + "editable": true + }, + "source": [ + "and with" + ] + }, + { + "cell_type": "markdown", + "id": "abdc427a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)= 0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a959437a", + "metadata": { + "editable": true + }, + "source": [ + "we have" + ] + }, + { + "cell_type": "markdown", + "id": "764e589a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\boldsymbol{\\epsilon}=\\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)= 0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "00e50636", + "metadata": { + "editable": true + }, + "source": [ + "meaning that the solution for $\\boldsymbol{\\beta}$ is the one which minimizes the residuals. Later we will link this with the maximum likelihood approach.\n", + "\n", + "Let us now return to our nuclear binding energies and simply code the above equations." + ] + }, + { + "cell_type": "markdown", + "id": "7f1fb52c", + "metadata": { + "editable": true + }, + "source": [ + "## Own code for Ordinary Least Squares\n", + "\n", + "It is rather straightforward to implement the matrix inversion and obtain the parameters $\\boldsymbol{\\beta}$. After having defined the matrix $\\boldsymbol{X}$ we simply need to \n", + "write" + ] + }, + { + "cell_type": "code", + "execution_count": 37, + "id": "6e7ef8e3", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# matrix inversion to find beta\n", + "beta = np.linalg.inv(X.T.dot(X)).dot(X.T).dot(Energies)\n", + "# and then make the prediction\n", + "ytilde = X @ beta" + ] + }, + { + "cell_type": "markdown", + "id": "e8de20b3", + "metadata": { + "editable": true + }, + "source": [ + "Alternatively, you can use the least squares functionality in **Numpy** as" + ] + }, + { + "cell_type": "code", + "execution_count": 38, + "id": "890c0e17", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "fit = np.linalg.lstsq(X, Energies, rcond =None)[0]\n", + "ytildenp = np.dot(fit,X.T)" + ] + }, + { + "cell_type": "markdown", + "id": "fa46eacd", + "metadata": { + "editable": true + }, + "source": [ + "And finally we plot our fit with and compare with data" + ] + }, + { + "cell_type": "code", + "execution_count": 39, + "id": "dd7bc08b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "Masses['Eapprox'] = ytilde\n", + "# Generate a plot comparing the experimental with the fitted values values.\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$A = N + Z$')\n", + "ax.set_ylabel(r'$E_\\mathrm{bind}\\,/\\mathrm{MeV}$')\n", + "ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,\n", + " label='Ame2016')\n", + "ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',\n", + " label='Fit')\n", + "ax.legend()\n", + "save_fig(\"Masses2016OLS\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "f1dac4cb", + "metadata": { + "editable": true + }, + "source": [ + "## Adding error analysis and training set up\n", + "\n", + "We can easily test our fit by computing the $R2$ score that we discussed in connection with the functionality of **Scikit-Learn** in the introductory slides.\n", + "Since we are not using **Scikit-Learn** here we can define our own $R2$ function as" + ] + }, + { + "cell_type": "code", + "execution_count": 40, + "id": "8253a870", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def R2(y_data, y_model):\n", + " return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)" + ] + }, + { + "cell_type": "markdown", + "id": "25175ebc", + "metadata": { + "editable": true + }, + "source": [ + "and we would be using it as" + ] + }, + { + "cell_type": "code", + "execution_count": 41, + "id": "4106532b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "print(R2(Energies,ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "3ae3cbe1", + "metadata": { + "editable": true + }, + "source": [ + "We can easily add our **MSE** score as" + ] + }, + { + "cell_type": "code", + "execution_count": 42, + "id": "9cbc337d", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def MSE(y_data,y_model):\n", + " n = np.size(y_model)\n", + " return np.sum((y_data-y_model)**2)/n\n", + "\n", + "print(MSE(Energies,ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "1f984da0", + "metadata": { + "editable": true + }, + "source": [ + "and finally the relative error as" + ] + }, + { + "cell_type": "code", + "execution_count": 43, + "id": "91777701", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def RelativeError(y_data,y_model):\n", + " return abs((y_data-y_model)/y_data)\n", + "print(RelativeError(Energies, ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "f504b559", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "Normally, the response (dependent or outcome) variable $y_i$ is the\n", + "outcome of a numerical experiment or another type of experiment and is\n", + "thus only an approximation to the true value. It is then always\n", + "accompanied by an error estimate, often limited to a statistical error\n", + "estimate given by the standard deviation discussed earlier. In the\n", + "discussion here we will treat $y_i$ as our exact value for the\n", + "response variable.\n", + "\n", + "Introducing the standard deviation $\\sigma_i$ for each measurement\n", + "$y_i$, we define now the $\\chi^2$ function (omitting the $1/n$ term)\n", + "as" + ] + }, + { + "cell_type": "markdown", + "id": "67440280", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\chi^2(\\boldsymbol{\\beta})=\\frac{1}{n}\\sum_{i=0}^{n-1}\\frac{\\left(y_i-\\tilde{y}_i\\right)^2}{\\sigma_i^2}=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)^T\\frac{1}{\\boldsymbol{\\Sigma^2}}\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c5190684", + "metadata": { + "editable": true + }, + "source": [ + "where the matrix $\\boldsymbol{\\Sigma}$ is a diagonal matrix with $\\sigma_i$ as matrix elements." + ] + }, + { + "cell_type": "markdown", + "id": "1b878f0e", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "In order to find the parameters $\\beta_i$ we will then minimize the spread of $\\chi^2(\\boldsymbol{\\beta})$ by requiring" + ] + }, + { + "cell_type": "markdown", + "id": "8ab0a4a3", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_j} = \\frac{\\partial }{\\partial \\beta_j}\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(\\frac{y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}}{\\sigma_i}\\right)^2\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e9106c8d", + "metadata": { + "editable": true + }, + "source": [ + "which results in" + ] + }, + { + "cell_type": "markdown", + "id": "03c4e9d7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_j} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}\\frac{x_{ij}}{\\sigma_i}\\left(\\frac{y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}}{\\sigma_i}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "0e4abfb1", + "metadata": { + "editable": true + }, + "source": [ + "or in a matrix-vector form as" + ] + }, + { + "cell_type": "markdown", + "id": "45af97a4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{A}^T\\left( \\boldsymbol{b}-\\boldsymbol{A}\\boldsymbol{\\beta}\\right).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "9ff2fda9", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the matrix $\\boldsymbol{A} =\\boldsymbol{X}/\\boldsymbol{\\Sigma}$ with matrix elements $a_{ij} = x_{ij}/\\sigma_i$ and the vector $\\boldsymbol{b}$ with elements $b_i = y_i/\\sigma_i$." + ] + }, + { + "cell_type": "markdown", + "id": "c1593fa5", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "We can rewrite" + ] + }, + { + "cell_type": "markdown", + "id": "d34ea7db", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{A}^T\\left( \\boldsymbol{b}-\\boldsymbol{A}\\boldsymbol{\\beta}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "699612be", + "metadata": { + "editable": true + }, + "source": [ + "as" + ] + }, + { + "cell_type": "markdown", + "id": "88ae1420", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{A}^T\\boldsymbol{b} = \\boldsymbol{A}^T\\boldsymbol{A}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b827d1c6", + "metadata": { + "editable": true + }, + "source": [ + "and if the matrix $\\boldsymbol{A}^T\\boldsymbol{A}$ is invertible we have the solution" + ] + }, + { + "cell_type": "markdown", + "id": "61900168", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} =\\left(\\boldsymbol{A}^T\\boldsymbol{A}\\right)^{-1}\\boldsymbol{A}^T\\boldsymbol{b}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "eaab94fc", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "If we then introduce the matrix" + ] + }, + { + "cell_type": "markdown", + "id": "c8f542a4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{H} = \\left(\\boldsymbol{A}^T\\boldsymbol{A}\\right)^{-1},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c6e93dd5", + "metadata": { + "editable": true + }, + "source": [ + "we have then the following expression for the parameters $\\beta_j$ (the matrix elements of $\\boldsymbol{H}$ are $h_{ij}$)" + ] + }, + { + "cell_type": "markdown", + "id": "977ca552", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_j = \\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}\\frac{y_i}{\\sigma_i}\\frac{x_{ik}}{\\sigma_i} = \\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}b_ia_{ik}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ff3c6b13", + "metadata": { + "editable": true + }, + "source": [ + "We state without proof the expression for the uncertainty in the parameters $\\beta_j$ as (we leave this as an exercise)" + ] + }, + { + "cell_type": "markdown", + "id": "ebbb06cf", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma^2(\\beta_j) = \\sum_{i=0}^{n-1}\\sigma_i^2\\left( \\frac{\\partial \\beta_j}{\\partial y_i}\\right)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ed6f641e", + "metadata": { + "editable": true + }, + "source": [ + "resulting in" + ] + }, + { + "cell_type": "markdown", + "id": "94c50447", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma^2(\\beta_j) = \\left(\\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}a_{ik}\\right)\\left(\\sum_{l=0}^{p-1}h_{jl}\\sum_{m=0}^{n-1}a_{ml}\\right) = h_{jj}!\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "caa52732", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "The first step here is to approximate the function $y$ with a first-order polynomial, that is we write" + ] + }, + { + "cell_type": "markdown", + "id": "721f3130", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y=y(x) \\rightarrow y(x_i) \\approx \\beta_0+\\beta_1 x_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "1ecec855", + "metadata": { + "editable": true + }, + "source": [ + "By computing the derivatives of $\\chi^2$ with respect to $\\beta_0$ and $\\beta_1$ show that these are given by" + ] + }, + { + "cell_type": "markdown", + "id": "9f72bcd2", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_0} = -2\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(\\frac{y_i-\\beta_0-\\beta_1x_{i}}{\\sigma_i^2}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "952c0807", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "db7a0783", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_1} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}x_i\\left(\\frac{y_i-\\beta_0-\\beta_1x_{i}}{\\sigma_i^2}\\right)\\right]=0.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "03a57144", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "For a linear fit (a first-order polynomial) we don't need to invert a matrix!! \n", + "Defining" + ] + }, + { + "cell_type": "markdown", + "id": "22e8525a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma = \\sum_{i=0}^{n-1}\\frac{1}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e7dc4b83", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_x = \\sum_{i=0}^{n-1}\\frac{x_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "5ceab08b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_y = \\sum_{i=0}^{n-1}\\left(\\frac{y_i}{\\sigma_i^2}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "61608838", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_{xx} = \\sum_{i=0}^{n-1}\\frac{x_ix_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "59fd7508", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_{xy} = \\sum_{i=0}^{n-1}\\frac{y_ix_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "caca22c4", + "metadata": { + "editable": true + }, + "source": [ + "we obtain" + ] + }, + { + "cell_type": "markdown", + "id": "b459c22c", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_0 = \\frac{\\gamma_{xx}\\gamma_y-\\gamma_x\\gamma_y}{\\gamma\\gamma_{xx}-\\gamma_x^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3715e857", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_1 = \\frac{\\gamma_{xy}\\gamma-\\gamma_x\\gamma_y}{\\gamma\\gamma_{xx}-\\gamma_x^2}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3ad3eeb2", + "metadata": { + "editable": true + }, + "source": [ + "This approach (different linear and non-linear regression) suffers\n", + "often from both being underdetermined and overdetermined in the\n", + "unknown coefficients $\\beta_i$. A better approach is to use the\n", + "Singular Value Decomposition (SVD) method discussed next week." + ] + }, + { + "cell_type": "markdown", + "id": "d8b193ef", + "metadata": { + "editable": true + }, + "source": [ + "## Fitting an Equation of State for Dense Nuclear Matter\n", + "\n", + "Before we continue, let us introduce yet another example. We are going to fit the\n", + "nuclear equation of state using results from many-body calculations.\n", + "The equation of state we have made available here, as function of\n", + "density, has been derived using modern nucleon-nucleon potentials with\n", + "[the addition of three-body\n", + "forces](https://www.sciencedirect.com/science/article/pii/S0370157399001106). This\n", + "time the file is presented as a standard **csv** file.\n", + "\n", + "The beginning of the Python code here is similar to what you have seen\n", + "before, with the same initializations and declarations. We use also\n", + "**pandas** again, rather extensively in order to organize our data.\n", + "\n", + "The difference now is that we use **Scikit-Learn's** regression tools\n", + "instead of our own matrix inversion implementation. Furthermore, we\n", + "sneak in **Ridge** regression (to be discussed below) which includes a\n", + "hyperparameter $\\lambda$, also to be explained below." + ] + }, + { + "cell_type": "markdown", + "id": "11007882", + "metadata": { + "editable": true + }, + "source": [ + "## The code" + ] + }, + { + "cell_type": "code", + "execution_count": 44, + "id": "ee8cdde5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import os\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "import matplotlib.pyplot as plt\n", + "import sklearn.linear_model as skl\n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"EoS.csv\"),'r')\n", + "\n", + "# Read the EoS data as csv file and organize the data into two arrays with density and energies\n", + "EoS = pd.read_csv(infile, names=('Density', 'Energy'))\n", + "EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')\n", + "EoS = EoS.dropna()\n", + "Energies = EoS['Energy']\n", + "Density = EoS['Density']\n", + "# The design matrix now as function of various polytrops\n", + "X = np.zeros((len(Density),4))\n", + "X[:,3] = Density**(4.0/3.0)\n", + "X[:,2] = Density\n", + "X[:,1] = Density**(2.0/3.0)\n", + "X[:,0] = 1\n", + "\n", + "# We use now Scikit-Learn's linear regressor and ridge regressor\n", + "# OLS part\n", + "clf = skl.LinearRegression().fit(X, Energies)\n", + "ytilde = clf.predict(X)\n", + "EoS['Eols'] = ytilde\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, ytilde))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, ytilde))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, ytilde))\n", + "print(clf.coef_, clf.intercept_)\n", + "\n", + "# The Ridge regression with a hyperparameter lambda = 0.1\n", + "_lambda = 0.1\n", + "clf_ridge = skl.Ridge(alpha=_lambda).fit(X, Energies)\n", + "yridge = clf_ridge.predict(X)\n", + "EoS['Eridge'] = yridge\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, yridge))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, yridge))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, yridge))\n", + "print(clf_ridge.coef_, clf_ridge.intercept_)\n", + "\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$\\rho[\\mathrm{fm}^{-3}]$')\n", + "ax.set_ylabel(r'Energy per particle')\n", + "ax.plot(EoS['Density'], EoS['Energy'], alpha=0.7, lw=2,\n", + " label='Theoretical data')\n", + "ax.plot(EoS['Density'], EoS['Eols'], alpha=0.7, lw=2, c='m',\n", + " label='OLS')\n", + "ax.plot(EoS['Density'], EoS['Eridge'], alpha=0.7, lw=2, c='g',\n", + " label='Ridge $\\lambda = 0.1$')\n", + "ax.legend()\n", + "save_fig(\"EoSfitting\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "1ea53cff", + "metadata": { + "editable": true + }, + "source": [ + "The above simple polynomial in density $\\rho$ gives an excellent fit\n", + "to the data. \n", + "\n", + "We note also that there is a small deviation between the\n", + "standard OLS and the Ridge regression at higher densities. We discuss this in more detail\n", + "below." + ] + }, + { + "cell_type": "markdown", + "id": "4d725567", + "metadata": { + "editable": true + }, + "source": [ + "## Splitting our Data in Training and Test data\n", + "\n", + "It is normal in essentially all Machine Learning studies to split the\n", + "data in a training set and a test set (sometimes also an additional\n", + "validation set). **Scikit-Learn** has an own function for this. There\n", + "is no explicit recipe for how much data should be included as training\n", + "data and say test data. An accepted rule of thumb is to use\n", + "approximately $2/3$ to $4/5$ of the data as training data. We will\n", + "postpone a discussion of this splitting to the end of these notes and\n", + "our discussion of the so-called **bias-variance** tradeoff. Here we\n", + "limit ourselves to repeat the above equation of state fitting example\n", + "but now splitting the data into a training set and a test set." + ] + }, + { + "cell_type": "code", + "execution_count": 45, + "id": "aa6446c1", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import os\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.model_selection import train_test_split\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "def R2(y_data, y_model):\n", + " return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)\n", + "def MSE(y_data,y_model):\n", + " n = np.size(y_model)\n", + " return np.sum((y_data-y_model)**2)/n\n", + "\n", + "infile = open(data_path(\"EoS.csv\"),'r')\n", + "\n", + "# Read the EoS data as csv file and organized into two arrays with density and energies\n", + "EoS = pd.read_csv(infile, names=('Density', 'Energy'))\n", + "EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')\n", + "EoS = EoS.dropna()\n", + "Energies = EoS['Energy']\n", + "Density = EoS['Density']\n", + "# The design matrix now as function of various polytrops\n", + "X = np.zeros((len(Density),5))\n", + "X[:,0] = 1\n", + "X[:,1] = Density**(2.0/3.0)\n", + "X[:,2] = Density\n", + "X[:,3] = Density**(4.0/3.0)\n", + "X[:,4] = Density**(5.0/3.0)\n", + "# We split the data in test and training data\n", + "X_train, X_test, y_train, y_test = train_test_split(X, Energies, test_size=0.2)\n", + "# matrix inversion to find beta\n", + "beta = np.linalg.inv(X_train.T.dot(X_train)).dot(X_train.T).dot(y_train)\n", + "# and then make the prediction\n", + "ytilde = X_train @ beta\n", + "print(\"Training R2\")\n", + "print(R2(y_train,ytilde))\n", + "print(\"Training MSE\")\n", + "print(MSE(y_train,ytilde))\n", + "ypredict = X_test @ beta\n", + "print(\"Test R2\")\n", + "print(R2(y_test,ypredict))\n", + "print(\"Test MSE\")\n", + "print(MSE(y_test,ypredict))" + ] + }, + { + "cell_type": "markdown", + "id": "7b772723", + "metadata": { + "editable": true + }, + "source": [ + "## Exercises\n", + "\n", + "Here are three possible exercises for week 34" + ] + }, + { + "cell_type": "markdown", + "id": "3cf1d40f", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 1: Setting up various Python environments\n", + "\n", + "The first exercise here is of a mere technical art. We want you to have \n", + "* git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). \n", + "\n", + "* Install various Python packages\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "IPython/Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow \n", + "\n", + "For **Tensorflow**, we recommend following the instructions in the text of \n", + "[Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do)\n", + "\n", + "We will come back to **tensorflow** later. \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for Python2.7)\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment." + ] + }, + { + "cell_type": "markdown", + "id": "59288b2d", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 2: making your own data and exploring scikit-learn\n", + "\n", + "We will generate our own dataset for a function $y(x)$ where $x \\in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\\cal {N}(0,1)$.\n", + "The following simple Python instructions define our $x$ and $y$ values (with 100 data points)." + ] + }, + { + "cell_type": "code", + "execution_count": 46, + "id": "f06dd2f8", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "x = np.random.rand(100,1)\n", + "y = 2.0+5*x*x+0.1*np.random.randn(100,1)" + ] + }, + { + "cell_type": "markdown", + "id": "21171361", + "metadata": { + "editable": true + }, + "source": [ + "1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. \n", + "\n", + "2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. \n", + "\n", + "3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as" + ] + }, + { + "cell_type": "markdown", + "id": "0389374f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e85ae9f5", + "metadata": { + "editable": true + }, + "source": [ + "and the $R^2$ score function.\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "6eac8058", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c9d88838", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "4a767648", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3b1a7735", + "metadata": { + "editable": true + }, + "source": [ + "You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. \n", + "Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits." + ] + }, + { + "cell_type": "markdown", + "id": "8c8ad214", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 3: Split data in test and training data\n", + "\n", + "In this exercise we want you to to compute the MSE for the training\n", + "data and the test data as function of the complexity of a polynomial,\n", + "that is the degree of a given polynomial.\n", + "\n", + "The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf).\n", + "\n", + "Our data is defined by $x\\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis." + ] + }, + { + "cell_type": "code", + "execution_count": 47, + "id": "43caea5f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "np.random.seed()\n", + "n = 100\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)" + ] + }, + { + "cell_type": "markdown", + "id": "9c604beb", + "metadata": { + "editable": true + }, + "source": [ + "where $y$ is the function we want to fit with a given polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "519328d9", + "metadata": { + "editable": true + }, + "source": [ + "**a)**\n", + "Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data." + ] + }, + { + "cell_type": "markdown", + "id": "3fdc48c8", + "metadata": { + "editable": true + }, + "source": [ + "**b)**\n", + "Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**)\n", + "and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "e6774cc5", + "metadata": { + "editable": true + }, + "source": [ + "**c)**\n", + "Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?" + ] + } + ], + "metadata": {}, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/doc/LectureNotes/_build/html/exercisesweek34.html b/doc/LectureNotes/_build/html/exercisesweek34.html new file mode 100644 index 000000000..7f55ce9e2 --- /dev/null +++ b/doc/LectureNotes/_build/html/exercisesweek34.html @@ -0,0 +1,603 @@ + + + + + + + + 19. Exercises week 34 — Applied Data Analysis and Machine Learning + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + \ No newline at end of file diff --git a/doc/LectureNotes/_build/html/genindex.html b/doc/LectureNotes/_build/html/genindex.html index 321feeb59..223b8f408 100644 --- a/doc/LectureNotes/_build/html/genindex.html +++ b/doc/LectureNotes/_build/html/genindex.html @@ -238,6 +238,23 @@ const thebe_selector_output = ".output, .cell_output" +

+ + Weekly material, notes and exercises + +

+ diff --git a/doc/LectureNotes/_build/html/intro.html b/doc/LectureNotes/_build/html/intro.html index cca8a66a4..d0110bbf8 100644 --- a/doc/LectureNotes/_build/html/intro.html +++ b/doc/LectureNotes/_build/html/intro.html @@ -239,6 +239,23 @@ const thebe_selector_output = ".output, .cell_output" +

+ + Weekly material, notes and exercises + +

+ @@ -590,6 +607,8 @@ It provides composable transformations of Python+NumPy programs: differentiate,
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
+
+ + + + + + + + +
+ +
+
+
+
+ + + + +
+ + +
+ + + + + + +
+
+ + + + + + + + + +
+ + +
+ +
+ Contents +
+ +
+
+
+
+
+ +
+

Week 34: Introduction to the course, Logistics and Practicalities

+ +
+
+ +
+

Contents

+
+ +
+
+
+ +
+ + +
+

20. Week 34: Introduction to the course, Logistics and Practicalities

+

Morten Hjorth-Jensen, Department of Physics and Center for Computing in Science Education, University of Oslo, Norway and Department of Physics and Astronomy and Facility for Rare Isotope Beams, Michigan State University, USA

+

Date: Week 34, August 21-25, 2023

+
+

20.1. Overview of first week

+
    +

  1. The sessions on Tuesdays and Wednesdays last four hours for each group (four groups in total) and will include lectures in a flipped mode (promoting active learning) and work on exercices and projects.

    2. +

  2. The sessions will begin with lectures, discussions, questions and answers about the material to be covered every week. Videos and teaching material will be announced in due time.

    3. +

  3. There are four groups:

    4. +
+
    +

  • Tuesdays 815am-12pm and 1215pm-4pm

    • +

  • Wednesdays 815am-12pm and 1215pm-4pm.

    • +
+
    +

  1. On Thursdays we have a regular lecture. These lectures start at 1215pm and end at 2pm and serve the aims of giving an overview over various topics. These lectures will also be recorded.

    2. +
+

The labs are also available till 6pm Tuesdays and Wednesdays. Videos and learning material with reading suggestions will be made available before each week starts.

+
+
+

20.2. Schedule first week

+
    +

  • August 22: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.

    • +

  • August 23: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.

    • +

  • August 24: Lecture: Linear regression, examples and theory

    • +
+
+
+

20.3. Lectures and ComputerLab

+
    +

  • The sessions on Tuesdays and Wednesdays last four hours and will include partly lectures in a flipped mode (promoting active learning) and work on exercices and projects.

    • +

  • Thursdays: regular lectures (12.15pm-2pm)

    • +

  • Weekly reading assignments and videos needed to solve projects and exercises.

    • +

  • Weekly exercises. You can hand in exercises if you want and get an extra score, see below.

    • +

  • Detailed lecture notes, exercises, all programs presented, projects etc can be found at the homepage of the course.

    • +

  • Weekly plans and all other information are on the official website. This info will also be conveyed via weekly emails.

    • +

  • No final exam, three projects that are graded and have to be approved.

    • +
+
+
+

20.4. Communication channels

+
    +

  • Chat and communications via <canvas.uio.no>

    • +

  • Discord channel will be added asap

    • +
+
+
+

20.5. Course Format

+
    +

  • Three compulsory projects. Electronic reports only using Canvas to hand in projects and git as version control software and GitHub for repository (or GitLab) of all your material.

    • +

  • Evaluation and grading: The three projects are graded and each counts 1/3 of the final mark. No final written or oral exam.

    • +
+

a. For the last project each group/participant submits a proposal or works with suggested (by us) proposals for the project.

+

b. If possible, we would like to organize the last project as a workshop where each group presents this to all other participants of the course

+

c. Based on feedback etc, each group finalizes the report and submits for grading.

+
    +

  • Python is the default programming language, but feel free to use C/C++, Julia and/or Fortran or other programming languages. All source codes discussed during the lectures can be found at the webpage and github address of the course.

    • +
+
+
+

20.6. Teachers

+ +
+
+

20.7. Deadlines for projects (tentative)

+
    +

  1. Project 1: October 9 (available September 4) graded with feedback)

    2. +

  2. Project 2: November 6 (available October 6, graded with feedback)

    3. +

  3. Project 3: December 11 (available November 10, graded with feedback)

    4. +
+

Extra Credit (not mandatory), weekly exercise assignments, 10 in total (due Friday same week), 10% additional score. The extra credit assignments are due each Friday and can be uploaed to Canvas in your preferred format (although we prefer jupyter-notebooks). First assignment is for week 35. Each weekly exercise set counts 1%.

+
+
+

20.8. Grading

+

Grades are awarded on a scale from A to F, where A is the best grade and F is a fail. There are three projects which are graded and each project counts 1/3 of the final grade. The total score is thus the average from all three projects.

+

The final number of points is based on the average of all projects and the grade follows the following table:

+
    +

  • 92-100 points: A

    • +

  • 77-91 points: B

    • +

  • 58-76 points: C

    • +

  • 46-57 points: D

    • +

  • 40-45 points: E

    • +

  • 0-39 points: F-failed

    • +
+

In addition you can get an extra 10% score for weekly assignments (10 in total and due each Friday). Each weekly assignment counts 1%.

+
+
+

20.9. Reading material

+

The lecture notes are collected as a jupyter-book at https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/intro.html.

+

In addition to the lecture notes, we recommend the books of Bishop, Hastie et al, Murphy and Goodfellow et al. We will follow these texts closely and the weekly reading assignments refer to these texts. The text by Hastie et al is also widely used in the Machine Learning community. Finally, we also recommend the hands-on text by Geron, see next slide for links.

+
+
+

20.10. Textbooks

+ +
+
+

20.11. Reading suggestions week 34

+

This week: Refresh linear algebra, GBC chapters 1 and 2. HTF chapters 2 and 3. Install scikit-learn. See lecture notes for week 34 at https://compphysics.github.io/MachineLearning/doc/web/course.html (these notes).

+
+
+

20.12. Prerequisites

+

Basic knowledge in programming and mathematics, with an emphasis on +linear algebra. Knowledge of Python or/and C++ as programming +languages is strongly recommended and experience with Jupiter notebook +is recommended. Required courses are the equivalents to the University +of Oslo mathematics courses MAT1100, MAT1110, MAT1120 and at least one +of the corresponding computing and programming courses INF1000/INF1110 +or MAT-INF1100/MAT-INF1100L/BIOS1100/KJM-INF1100. Most universities +offer nowadays a basic programming course (often compulsory) where +Python is the recurring programming language.

+
+
+

20.13. Topics covered in this course: Statistical analysis and optimization of data

+

The course has two central parts

+
    +

  1. Statistical analysis and optimization of data

    2. +

  2. Machine learning

    3. +
+

These topics will be scattered thorughout the course and may not necessarily be taught separately. Rather, we will often take an approach (during the lectures and project/exercise sessions) where say elements from statistical data analysis are mixed with specific Machine Learning algorithms.

+
+
+

20.14. Statistical analysis and optimization of data

+

We plan to cover the following topics:

+
    +

  • Basic concepts, expectation values, variance, covariance, correlation functions and errors;

    • +

  • Simpler models, binomial distribution, the Poisson distribution, simple and multivariate normal distributions;

    • +

  • Central elements of Bayesian statistics and modeling;

    • +

  • Gradient methods for data optimization;

    • +

  • Monte Carlo methods, Markov chains, Gibbs sampling and Metropolis-Hastings sampling (tentative);

    • +

  • Estimation of errors and resampling techniques such as the cross-validation, blocking, bootstrapping and jackknife methods;

    • +

  • Principal Component Analysis (PCA) and its mathematical foundation;

    • +
+
+
+

20.15. Machine Learning

+
    +

  • Pre deep-learning revolution (2008 approx)

    +
      +

    • Linear Regression and Logistic Regression, classification and regression problems;

      • +

    • Bayesian linear and logistic regression, kernel regression;

      • +

    • Decisions trees, Random Forests, Bagging and Boosting methods;

      • +

    • Support vector machines (only survey);

      • +

    • Unsupervised learning and dimensionality reduction, from PCA to clustering;

      • +
    +
    • +

  • Deep learning

    +
      +

    • Neural networks and deep learning;

      • +

    • Convolutional neural networks;

      • +

    • Recurrent neural networks;

      • +

    • Autoencoders

      • +

    • Generative methods with an emphasis on Boltzmann Machines, Variational Autoencoders and Generalized Adversarial Networks;

      • +
    +
    • +
+

Hands-on demonstrations, exercises and projects aim at deepening your understanding of these topics.

+
+ +
+

20.17. Other courses on Data science and Machine Learning at UiO

+ +
+
+

20.18. Other courses on Data science and Machine Learning at UiO, contn

+ +
+
+

20.19. Learning outcomes

+

This course aims at giving you insights and knowledge about many of +the central algorithms used in Data Analysis and Machine Learning. +The course is project based and through various numerical projects, +normally three, you will be exposed to fundamental research problems +in these fields, with the aim to reproduce state of the art scientific +results. Both supervised and unsupervised methods will be covered. The +emphasis is on a frequentist approach, although we will try to link it +with a Bayesian approach as well. You will learn to develop and +structure large codes for studying different cases where Machine +Learning is applied to, get acquainted with computing facilities and +learn to handle large scientific projects. A good scientific and +ethical conduct is emphasized throughout the course. More +specifically, after this course you will

+
    +

  • Learn about basic data analysis, statistical analysis, Bayesian statistics, Monte Carlo sampling, data optimization and machine learning;

    • +

  • Be capable of extending the acquired knowledge to other systems and cases;

    • +

  • Have an understanding of central algorithms used in data analysis and machine learning;

    • +

  • Understand linear methods for regression and classification, from ordinary least squares, via Lasso and Ridge to Logistic regression;

    • +

  • Learn about neural networks and deep learning methods for supervised and unsupervised learning. Emphasis on feed forward neural networks, convolutional and recurrent neural networks;

    • +

  • Learn about about decision trees, random forests, bagging and boosting methods;

    • +

  • Learn about support vector machines and kernel transformations;

    • +

  • Reduction of data sets, from PCA to clustering;

    • +

  • Generative models

    • +

  • Work on numerical projects to illustrate the theory. The projects play a central role and you are expected to know modern programming languages like Python or C++ and/or Fortran (Fortran2003 or later) or Julia or other.

    • +
+
+
+

20.20. Introduction

+

Our emphasis throughout this series of lectures
+is on understanding the mathematical aspects of +different algorithms used in the fields of data analysis and machine learning.

+

However, where possible we will emphasize the +importance of using available software. We start thus with a hands-on +and top-down approach to machine learning. The aim is thus to start with +relevant data or data we have produced +and use these to introduce statistical data analysis +concepts and machine learning algorithms before we delve into the +algorithms themselves. The examples we will use in the beginning, start with simple +polynomials with random noise added. We will use the Python +software package Scikit-Learn and +introduce various machine learning algorithms to make fits of +the data and predictions. We move thereafter to more interesting +cases such as data from say experiments (below we will look at experimental nuclear binding energies as an example). +These are examples where we can easily set up the data and +then use machine learning algorithms included in for example +Scikit-Learn.

+

These examples will serve us the purpose of getting +started. Furthermore, they allow us to catch more than two birds with +a stone. They will allow us to bring in some programming specific +topics and tools as well as showing the power of various Python +libraries for machine learning and statistical data analysis.

+

Although we have projects where you write your own codes, we will also focus on two +specific Python packages for Machine Learning, Scikit-Learn and +Tensorflow with Keras (see below for links etc). Moreover, the examples we +introduce will serve as inputs to many of our discussions later, as +well as allowing you to set up models and produce your own data and +get started with programming.

+
+
+

20.21. AI/ML and some statements you may have heard (and what do they mean?)

+
    +

  1. Fei-Fei Li on ImageNet: map out the entire world of objects (The data that transformed AI research)

    2. +

  2. Russell and Norvig in their popular textbook: relevant to any intellectual task; it is truly a universal field (Artificial Intelligence, A modern approach)

    3. +

  3. Woody Bledsoe puts it more bluntly: in the long run, AI is the only science (quoted in Pamilla McCorduck, Machines who think)

    4. +
+

If you wish to have a critical read on AI/ML from a societal point of view, see Kate Crawford’s recent text Atlas of AI

+

Here: with AI/ML we intend a collection of machine learning methods with an emphasis on statistical learning and data analysis

+
+
+

20.22. What is Machine Learning?

+

Statistics, data science and machine learning form important fields of +research in modern science. They describe how to learn and make +predictions from data, as well as allowing us to extract important +correlations about physical process and the underlying laws of motion +in large data sets. The latter, big data sets, appear frequently in +essentially all disciplines, from the traditional Science, Technology, +Mathematics and Engineering fields to Life Science, Law, education +research, the Humanities and the Social Sciences.

+

It has become more +and more common to see research projects on big data in for example +the Social Sciences where extracting patterns from complicated survey +data is one of many research directions. Having a solid grasp of data +analysis and machine learning is thus becoming central to scientific +computing in many fields, and competences and skills within the fields +of machine learning and scientific computing are nowadays strongly +requested by many potential employers. The latter cannot be +overstated, familiarity with machine learning has almost become a +prerequisite for many of the most exciting employment opportunities, +whether they are in bioinformatics, life science, physics or finance, +in the private or the public sector. This author has had several +students or met students who have been hired recently based on their +skills and competences in scientific computing and data science, often +with marginal knowledge of machine learning.

+

Machine learning is a subfield of computer science, and is closely +related to computational statistics. It evolved from the study of +pattern recognition in artificial intelligence (AI) research, and has +made contributions to AI tasks like computer vision, natural language +processing and speech recognition. Many of the methods we will study are also +strongly rooted in basic mathematics and physics research.

+

Ideally, machine learning represents the science of giving computers +the ability to learn without being explicitly programmed. The idea is +that there exist generic algorithms which can be used to find patterns +in a broad class of data sets without having to write code +specifically for each problem. The algorithm will build its own logic +based on the data. You should however always keep in mind that +machines and algorithms are to a large extent developed by humans. The +insights and knowledge we have about a specific system, play a central +role when we develop a specific machine learning algorithm.

+

Machine learning is an extremely rich field, in spite of its young +age. The increases we have seen during the last three decades in +computational capabilities have been followed by developments of +methods and techniques for analyzing and handling large date sets, +relying heavily on statistics, computer science and mathematics. The +field is rather new and developing rapidly. Popular software packages +written in Python for machine learning like +Scikit-learn, +Tensorflow, +PyTorch and Keras, all +freely available at their respective GitHub sites, encompass +communities of developers in the thousands or more. And the number of +code developers and contributors keeps increasing. Not all the +algorithms and methods can be given a rigorous mathematical +justification, opening up thereby large rooms for experimenting and +trial and error and thereby exciting new developments. However, a +solid command of linear algebra, multivariate theory, probability +theory, statistical data analysis, understanding errors and Monte +Carlo methods are central elements in a proper understanding of many +of algorithms and methods we will discuss.

+
+
+

20.23. Types of Machine Learning

+

The approaches to machine learning are many, but are often split into +two main categories. In supervised learning we know the answer to a +problem, and let the computer deduce the logic behind it. On the other +hand, unsupervised learning is a method for finding patterns and +relationship in data sets without any prior knowledge of the system. +Some authours also operate with a third category, namely +reinforcement learning. This is a paradigm of learning inspired by +behavioral psychology, where learning is achieved by trial-and-error, +solely from rewards and punishment.

+

Another way to categorize machine learning tasks is to consider the +desired output of a system. Some of the most common tasks are:

+
    +

  • Classification: Outputs are divided into two or more classes. The goal is to produce a model that assigns inputs into one of these classes. An example is to identify digits based on pictures of hand-written ones. Classification is typically supervised learning.

    • +

  • Regression: Finding a functional relationship between an input data set and a reference data set. The goal is to construct a function that maps input data to continuous output values.

    • +

  • Clustering: Data are divided into groups with certain common traits, without knowing the different groups beforehand. It is thus a form of unsupervised learning.

    • +
+
+
+

20.24. Essential elements of ML

+

The methods we cover have three main topics in common, irrespective of +whether we deal with supervised or unsupervised learning.

+
    +

  • The first ingredient is normally our data set (which can be subdivided into training, validation and test data). Many find the most difficult part of using Machine Learning to be the set up of your data in a meaningful way.

    • +

  • The second item is a model which is normally a function of some parameters. The model reflects our knowledge of the system (or lack thereof). As an example, if we know that our data show a behavior similar to what would be predicted by a polynomial, fitting our data to a polynomial of some degree would then determin our model.

    • +

  • The last ingredient is a so-called cost/loss function (or error or risk function) which allows us to present an estimate on how good our model is in reproducing the data it is supposed to train.

    • +
+
+
+

20.25. An optimization/minimization problem

+

At the heart of basically all Machine Learning algorithms we will encounter so-called minimization or optimization algorithms. A large family of such methods are so-called gradient methods.

+
+
+

20.26. A Frequentist approach to data analysis

+

When you hear phrases like predictions and estimations and +correlations and causations, what do you think of? May be you think +of the difference between classifying new data points and generating +new data points. +Or perhaps you consider that correlations represent some kind of symmetric statements like +if \(A\) is correlated with \(B\), then \(B\) is correlated with +\(A\). Causation on the other hand is directional, that is if \(A\) causes \(B\), \(B\) does not +necessarily cause \(A\).

+

These concepts are in some sense the difference between machine +learning and statistics. In machine learning and prediction based +tasks, we are often interested in developing algorithms that are +capable of learning patterns from given data in an automated fashion, +and then using these learned patterns to make predictions or +assessments of newly given data. In many cases, our primary concern +is the quality of the predictions or assessments, and we are less +concerned about the underlying patterns that were learned in order +to make these predictions.

+

In machine learning we normally use a so-called frequentist approach, +where the aim is to make predictions and find correlations. We focus +less on for example extracting a probability distribution function (PDF). The PDF can be +used in turn to make estimations and find causations such as given \(A\) +what is the likelihood of finding \(B\).

+
+
+

20.27. What is a good model?

+

In science and engineering we often end up in situations where we want to infer (or learn) a +quantitative model \(M\) for a given set of sample points \(\boldsymbol{X} \in [x_1, x_2,\dots x_N]\).

+

As we will see repeatedely in these lectures, we could try to fit these data points to a model given by a +straight line, or if we wish to be more sophisticated to a more complex +function.

+

The reason for inferring such a model is that it +serves many useful purposes. On the one hand, the model can reveal information +encoded in the data or underlying mechanisms from which the data were generated. For instance, we could discover important +corelations that relate interesting physics interpretations.

+

In addition, it can simplify the representation of the given data set and help +us in making predictions about future data samples.

+

A first important consideration to keep in mind is that inferring the correct model +for a given data set is an elusive, if not impossible, task. The fundamental difficulty +is that if we are not specific about what we mean by a correct model, there +could easily be many different models that fit the given data set equally well.

+
+
+

20.28. What is a good model? Can we define it?

+

The central question is this: what leads us to say that a model is correct or +optimal for a given data set? To make the model inference problem well posed, i.e., +to guarantee that there is a unique optimal model for the given data, we need to +impose additional assumptions or restrictions on the class of models considered. To +this end, we should not be looking for just any model that can describe the data. +Instead, we should look for a model \(M\) that is the best among a restricted class +of models. In addition, to make the model inference problem computationally +tractable, we need to specify how restricted the class of models needs to be. A +common strategy is to start +with the simplest possible class of models that is just necessary to describe the data +or solve the problem at hand. More precisely, the model class should be rich enough +to contain at least one model that can fit the data to a desired accuracy and yet be +restricted enough that it is relatively simple to find the best model for the given data.

+

Thus, the most popular strategy is to start from the +simplest class of models and increase the complexity of the models only when the +simpler models become inadequate. For instance, if we work with a regression problem to fit a set of sample points, one +may first try the simplest class of models, namely linear models, followed obviously by more complex models.

+

How to evaluate which model fits best the data is something we will come back to over and over again in these sets of lectures.

+
+
+

20.29. Software and needed installations

+

We will make extensive use of Python as programming language and its +myriad of available libraries. You will find +Jupyter notebooks invaluable in your work. You can run R +codes in the Jupyter/IPython notebooks, with the immediate benefit of +visualizing your data. You can also use compiled languages like C++, +Rust, Julia, Fortran etc if you prefer. The focus in these lectures will be +on Python.

+

If you have Python installed (we strongly recommend Python3) and you feel +pretty familiar with installing different packages, we recommend that +you install the following Python packages via pip as

+
    +

  1. pip install numpy scipy matplotlib ipython scikit-learn mglearn sympy pandas pillow

    2. +
+

For Python3, replace pip with pip3.

+

For OSX users we recommend, after having installed Xcode, to +install brew. Brew allows for a seamless installation of additional +software via for example

+
    +

  1. brew install python3

    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +you can use pip as well and simply install Python as

+
    +

  1. sudo apt-get install python3 (or python for pyhton2.7)

    2. +
+

etc etc.

+
+
+

20.30. Python installers

+

If you don’t want to perform these operations separately and venture +into the hassle of exploring how to set up dependencies and paths, we +recommend two widely used distrubutions which set up all relevant +dependencies for Python, namely

+ +

which is an open source +distribution of the Python and R programming languages for large-scale +data processing, predictive analytics, and scientific computing, that +aims to simplify package management and deployment. Package versions +are managed by the package management system conda.

+ +

is a Python +distribution for scientific and analytic computing distribution and +analysis environment, available for free and under a commercial +license.

+

Furthermore, Google’s Colab is a free Jupyter notebook environment that requires +no setup and runs entirely in the cloud. Try it out!

+
+
+

20.31. Useful Python libraries

+

Here we list several useful Python libraries we strongly recommend (if you use anaconda many of these are already there)

+
    +

  • NumPy is a highly popular library for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays

    • +

  • The pandas library provides high-performance, easy-to-use data structures and data analysis tools

    • +

  • Xarray is a Python package that makes working with labelled multi-dimensional arrays simple, efficient, and fun!

    • +

  • Scipy (pronounced “Sigh Pie”) is a Python-based ecosystem of open-source software for mathematics, science, and engineering.

    • +

  • Matplotlib is a Python 2D plotting library which produces publication quality figures in a variety of hardcopy formats and interactive environments across platforms.

    • +

  • Autograd can automatically differentiate native Python and Numpy code. It can handle a large subset of Python’s features, including loops, ifs, recursion and closures, and it can even take derivatives of derivatives of derivatives

    • +

  • JAX has now more or less replaced Autograd. JAX is Autograd and XLA, brought together for high-performance numerical computing and machine learning research. It provides composable transformations of Python+NumPy programs: differentiate, vectorize, parallelize, Just-In-Time compile to GPU/TPU, and more.

    • +

  • SymPy is a Python library for symbolic mathematics.

    • +

  • scikit-learn has simple and efficient tools for machine learning, data mining and data analysis

    • +

  • TensorFlow is a Python library for fast numerical computing created and released by Google

    • +

  • Keras is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano

    • +

  • And many more such as pytorch, Theano etc

    • +
+
+
+

20.32. Installing R, C++, cython or Julia

+

You will also find it convenient to utilize R. We will mainly +use Python during our lectures and in various projects and exercises. +Those of you +already familiar with R should feel free to continue using R, keeping +however an eye on the parallel Python set ups. Similarly, if you are a +Python afecionado, feel free to explore R as well. Jupyter(Julia, Python and R) /Ipython +notebook allows you to run R codes and Julia codes interactively in your +browser. The software library R is really tailored for statistical data analysis +and allows for an easy usage of the tools and algorithms we will discuss in these +lectures.

+

To install R with Jupyter notebook +follow the link here

+
+
+

20.33. Installing R, C++, cython, Numba etc

+

For the C++ aficionados, Jupyter/IPython notebook allows you also to +install C++ and run codes written in this language interactively in +the browser. Since we will emphasize writing many of the algorithms +yourself, you can thus opt for either Python or C++ (or Fortran or other compiled languages) as programming +languages.

+

To add more entropy, cython can also be used when running your +notebooks. It means that Python with the jupyter notebook +setup allows you to integrate widely popular softwares and tools for +scientific computing. Similarly, the +Numba Python package delivers increased performance +capabilities with minimal rewrites of your codes. With its +versatility, including symbolic operations, Python offers a unique +computational environment. Your jupyter notebook can easily be +converted into a nicely rendered PDF file or a Latex file for +further processing. For example, convert to latex as

+
    pycod jupyter nbconvert filename.ipynb --to latex 
+
+
+

And to add more versatility, the Python package SymPy is a Python library for symbolic mathematics. It aims to become a full-featured computer algebra system (CAS) and is entirely written in Python.

+

Finally, we recommend strongly using Autograd or JAX for automatic differentiation.

+
+
+

20.34. Numpy examples and Important Matrix and vector handling packages

+

There are several central software libraries for linear algebra and eigenvalue problems. Several of the more +popular ones have been wrapped into ofter software packages like those from the widely used text Numerical Recipes. The original source codes in many of the available packages are often taken from the widely used +software package LAPACK, which follows two other popular packages +developed in the 1970s, namely EISPACK and LINPACK. We describe them shortly here.

+
    +

  • LINPACK: package for linear equations and least square problems.

    • +

  • LAPACK:package for solving symmetric, unsymmetric and generalized eigenvalue problems. From LAPACK’s website http://www.netlib.org it is possible to download for free all source codes from this library. Both C/C++ and Fortran versions are available.

    • +

  • BLAS (I, II and III): (Basic Linear Algebra Subprograms) are routines that provide standard building blocks for performing basic vector and matrix operations. Blas I is vector operations, II vector-matrix operations and III matrix-matrix operations. Highly parallelized and efficient codes, all available for download from http://www.netlib.org.

    • +
+
+
+

20.35. Numpy and arrays

+

Numpy provides an easy way to handle arrays in Python. The standard way to import this library is as

+
+
+
import numpy as np
+
+
+
+
+

Here follows a simple example where we set up an array of ten elements, all determined by random numbers drawn according to the normal distribution,

+
+
+
n = 10
+x = np.random.normal(size=n)
+print(x)
+
+
+
+
+
[ 0.39864527 -1.50219982 -0.24611674 -0.52819229 -0.24533566 -0.85694362
+ -0.81955926  0.14252712 -1.00625843 -0.23732849]
+
+
+
+
+

We defined a vector \(x\) with \(n=10\) elements with its values given by the Normal distribution \(N(0,1)\). +Another alternative is to declare a vector as follows

+
+
+
import numpy as np
+x = np.array([1, 2, 3])
+print(x)
+
+
+
+
+
[1 2 3]
+
+
+
+
+

Here we have defined a vector with three elements, with \(x_0=1\), \(x_1=2\) and \(x_2=3\). Note that both Python and C++ +start numbering array elements from \(0\) and on. This means that a vector with \(n\) elements has a sequence of entities \(x_0, x_1, x_2, \dots, x_{n-1}\). We could also let (recommended) Numpy to compute the logarithms of a specific array as

+
+
+
import numpy as np
+x = np.log(np.array([4, 7, 8]))
+print(x)
+
+
+
+
+
[1.38629436 1.94591015 2.07944154]
+
+
+
+
+

In the last example we used Numpy’s unary function \(np.log\). This function is +highly tuned to compute array elements since the code is vectorized +and does not require looping. We normaly recommend that you use the +Numpy intrinsic functions instead of the corresponding log function +from Python’s math module. The looping is done explicitely by the +np.log function. The alternative, and slower way to compute the +logarithms of a vector would be to write

+
+
+
import numpy as np
+from math import log
+x = np.array([4, 7, 8])
+for i in range(0, len(x)):
+    x[i] = log(x[i])
+print(x)
+
+
+
+
+
[1 1 2]
+
+
+
+
+

We note that our code is much longer already and we need to import the log function from the math module. +The attentive reader will also notice that the output is \([1, 1, 2]\). Python interprets automagically our numbers as integers (like the automatic keyword in C++). To change this we could define our array elements to be double precision numbers as

+
+
+
import numpy as np
+x = np.log(np.array([4, 7, 8], dtype = np.float64))
+print(x)
+
+
+
+
+
[1.38629436 1.94591015 2.07944154]
+
+
+
+
+

or simply write them as double precision numbers (Python uses 64 bits as default for floating point type variables), that is

+
+
+
import numpy as np
+x = np.log(np.array([4.0, 7.0, 8.0]))
+print(x)
+
+
+
+
+
[1.38629436 1.94591015 2.07944154]
+
+
+
+
+

To check the number of bytes (remember that one byte contains eight bits for double precision variables), you can use simple use the itemsize functionality (the array \(x\) is actually an object which inherits the functionalities defined in Numpy) as

+
+
+
import numpy as np
+x = np.log(np.array([4.0, 7.0, 8.0]))
+print(x.itemsize)
+
+
+
+
+
8
+
+
+
+
+
+
+

20.36. Matrices in Python

+

Having defined vectors, we are now ready to try out matrices. We can +define a \(3 \times 3 \) real matrix \(\boldsymbol{A}\) as (recall that we user +lowercase letters for vectors and uppercase letters for matrices)

+
+
+
import numpy as np
+A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))
+print(A)
+
+
+
+
+
[[1.38629436 1.94591015 2.07944154]
+ [1.09861229 2.30258509 2.39789527]
+ [1.38629436 1.60943791 1.94591015]]
+
+
+
+
+

If we use the shape function we would get \((3, 3)\) as output, that is verifying that our matrix is a \(3\times 3\) matrix. We can slice the matrix and print for example the first column (Python organized matrix elements in a row-major order, see below) as

+
+
+
import numpy as np
+A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))
+# print the first column, row-major order and elements start with 0
+print(A[:,0])
+
+
+
+
+
[1.38629436 1.09861229 1.38629436]
+
+
+
+
+

We can continue this was by printing out other columns or rows. The example here prints out the second column

+
+
+
import numpy as np
+A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))
+# print the first column, row-major order and elements start with 0
+print(A[1,:])
+
+
+
+
+
[1.09861229 2.30258509 2.39789527]
+
+
+
+
+

Numpy contains many other functionalities that allow us to slice, subdivide etc etc arrays. We strongly recommend that you look up the Numpy website for more details. Useful functions when defining a matrix are the np.zeros function which declares a matrix of a given dimension and sets all elements to zero

+
+
+
import numpy as np
+n = 10
+# define a matrix of dimension 10 x 10 and set all elements to zero
+A = np.zeros( (n, n) )
+print(A)
+
+
+
+
+
[[0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
+ [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]]
+
+
+
+
+

or initializing all elements to

+
+
+
import numpy as np
+n = 10
+# define a matrix of dimension 10 x 10 and set all elements to one
+A = np.ones( (n, n) )
+print(A)
+
+
+
+
+
[[1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
+ [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]]
+
+
+
+
+

or as unitarily distributed random numbers (see the material on random number generators in the statistics part)

+
+
+
import numpy as np
+n = 10
+# define a matrix of dimension 10 x 10 and set all elements to random numbers with x \in [0, 1]
+A = np.random.rand(n, n)
+print(A)
+
+
+
+
+
[[0.99374388 0.68029036 0.03324117 0.4017308  0.92888885 0.12434807
+  0.9538239  0.15899253 0.48442336 0.72057967]
+ [0.16840081 0.01852793 0.10950766 0.25522012 0.78674758 0.65280103
+  0.220968   0.61634314 0.25115233 0.05622775]
+ [0.87927601 0.35147532 0.62432074 0.16167387 0.49242688 0.12934015
+  0.50812169 0.35123041 0.95054046 0.66638078]
+ [0.08887708 0.95994091 0.21927988 0.05833369 0.49974854 0.20999563
+  0.28102554 0.01479194 0.84942942 0.30702075]
+ [0.29672381 0.99820454 0.56848886 0.32190363 0.13320535 0.31768693
+  0.80280228 0.59654302 0.25116851 0.81681019]
+ [0.50799086 0.13261136 0.33078983 0.53623614 0.14174062 0.06662062
+  0.32183101 0.8338786  0.32766274 0.88595926]
+ [0.12539459 0.60176806 0.12885048 0.00708794 0.66614025 0.70727837
+  0.87849647 0.05491655 0.37157712 0.93325497]
+ [0.56343926 0.23056012 0.97840468 0.82484316 0.9956582  0.53822787
+  0.53219937 0.58312967 0.14617985 0.59468796]
+ [0.67054028 0.4304167  0.38299798 0.16397496 0.81912272 0.54894095
+  0.53254035 0.17826349 0.01551946 0.54969168]
+ [0.92373266 0.86361743 0.27525917 0.59367954 0.85843684 0.40389922
+  0.8074242  0.59323905 0.51075117 0.79552752]]
+
+
+
+
+

As we will see throughout these lectures, there are several extremely useful functionalities in Numpy. +As an example, consider the discussion of the covariance matrix. Suppose we have defined three vectors +\(\boldsymbol{x}, \boldsymbol{y}, \boldsymbol{z}\) with \(n\) elements each. The covariance matrix is defined as

+
+\[\begin{split} +\boldsymbol{\Sigma} = \begin{bmatrix} \sigma_{xx} & \sigma_{xy} & \sigma_{xz} \\ + \sigma_{yx} & \sigma_{yy} & \sigma_{yz} \\ + \sigma_{zx} & \sigma_{zy} & \sigma_{zz} + \end{bmatrix}, +\end{split}\]
+

where for example

+
+\[ +\sigma_{xy} =\frac{1}{n} \sum_{i=0}^{n-1}(x_i- \overline{x})(y_i- \overline{y}). +\]
+

The Numpy function np.cov calculates the covariance elements using the factor \(1/(n-1)\) instead of \(1/n\) since it assumes we do not have the exact mean values. +The following simple function uses the np.vstack function which takes each vector of dimension \(1\times n\) and produces a \(3\times n\) matrix \(\boldsymbol{W}\)

+
+\[\begin{split} +\boldsymbol{W} = \begin{bmatrix} x_0 & x_1 & x_2 & \dots & x_{n-2} & x_{n-1} \\ + y_0 & y_1 & y_2 & \dots & y_{n-2} & y_{n-1} \\ + z_0 & z_1 & z_2 & \dots & z_{n-2} & z_{n-1} \\ + \end{bmatrix}, +\end{split}\]
+

which in turn is converted into into the \(3\times 3\) covariance matrix +\(\boldsymbol{\Sigma}\) via the Numpy function np.cov(). We note that we can also calculate +the mean value of each set of samples \(\boldsymbol{x}\) etc using the Numpy +function np.mean(x). We can also extract the eigenvalues of the +covariance matrix through the np.linalg.eig() function.

+
+
+
# Importing various packages
+import numpy as np
+
+n = 100
+x = np.random.normal(size=n)
+print(np.mean(x))
+y = 4+3*x+np.random.normal(size=n)
+print(np.mean(y))
+z = x**3+np.random.normal(size=n)
+print(np.mean(z))
+W = np.vstack((x, y, z))
+Sigma = np.cov(W)
+print(Sigma)
+Eigvals, Eigvecs = np.linalg.eig(Sigma)
+print(Eigvals)
+
+
+
+
+
-0.007334080205654388
+3.9196398947631166
+0.49680475019226
+[[ 0.83982725  2.39595799  2.20255857]
+ [ 2.39595799  8.04496448  5.72101205]
+ [ 2.20255857  5.72101205 10.77494715]]
+[15.97477238  0.08363087  3.60133562]
+
+
+
+
+
+
+
%matplotlib inline
+
+import numpy as np
+import matplotlib.pyplot as plt
+from scipy import sparse
+eye = np.eye(4)
+print(eye)
+sparse_mtx = sparse.csr_matrix(eye)
+print(sparse_mtx)
+x = np.linspace(-10,10,100)
+y = np.sin(x)
+plt.plot(x,y,marker='x')
+plt.show()
+
+
+
+
+
[[1. 0. 0. 0.]
+ [0. 1. 0. 0.]
+ [0. 0. 1. 0.]
+ [0. 0. 0. 1.]]
+  (0, 0)	1.0
+  (1, 1)	1.0
+  (2, 2)	1.0
+  (3, 3)	1.0
+
+
+_images/week34_74_1.png +
+
+
+
+

20.37. Meet the Pandas

+ + +

Figure 1:

+ +

Another useful Python package is +pandas, which is an open source library +providing high-performance, easy-to-use data structures and data +analysis tools for Python. pandas stands for panel data, a term borrowed from econometrics and is an efficient library for data analysis with an emphasis on tabular data. +pandas has two major classes, the DataFrame class with two-dimensional data objects and tabular data organized in columns and the class Series with a focus on one-dimensional data objects. Both classes allow you to index data easily as we will see in the examples below. +pandas allows you also to perform mathematical operations on the data, spanning from simple reshapings of vectors and matrices to statistical operations.

+

The following simple example shows how we can, in an easy way make tables of our data. Here we define a data set which includes names, place of birth and date of birth, and displays the data in an easy to read way. We will see repeated use of pandas, in particular in connection with classification of data.

+
+
+
import pandas as pd
+from IPython.display import display
+data = {'First Name': ["Frodo", "Bilbo", "Aragorn II", "Samwise"],
+        'Last Name': ["Baggins", "Baggins","Elessar","Gamgee"],
+        'Place of birth': ["Shire", "Shire", "Eriador", "Shire"],
+        'Date of Birth T.A.': [2968, 2890, 2931, 2980]
+        }
+data_pandas = pd.DataFrame(data)
+display(data_pandas)
+
+
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
First NameLast NamePlace of birthDate of Birth T.A.
0FrodoBagginsShire2968
1BilboBagginsShire2890
2Aragorn IIElessarEriador2931
3SamwiseGamgeeShire2980
+
+
+

In the above we have imported pandas with the shorthand pd, the latter has become the standard way we import pandas. We make then a list of various variables +and reorganize the aboves lists into a DataFrame and then print out a neat table with specific column labels as Name, place of birth and date of birth. +Displaying these results, we see that the indices are given by the default numbers from zero to three. +pandas is extremely flexible and we can easily change the above indices by defining a new type of indexing as

+
+
+
data_pandas = pd.DataFrame(data,index=['Frodo','Bilbo','Aragorn','Sam'])
+display(data_pandas)
+
+
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
First NameLast NamePlace of birthDate of Birth T.A.
FrodoFrodoBagginsShire2968
BilboBilboBagginsShire2890
AragornAragorn IIElessarEriador2931
SamSamwiseGamgeeShire2980
+
+
+

Thereafter we display the content of the row which begins with the index Aragorn

+
+
+
display(data_pandas.loc['Aragorn'])
+
+
+
+
+
First Name            Aragorn II
+Last Name                Elessar
+Place of birth           Eriador
+Date of Birth T.A.          2931
+Name: Aragorn, dtype: object
+
+
+
+
+

We can easily append data to this, for example

+
+
+
new_hobbit = {'First Name': ["Peregrin"],
+              'Last Name': ["Took"],
+              'Place of birth': ["Shire"],
+              'Date of Birth T.A.': [2990]
+              }
+data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin']))
+display(data_pandas)
+
+
+
+
+
/var/folders/td/3yk470mj5p931p9dtkk0y6jw0000gn/T/ipykernel_65914/1326197715.py:6: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.
+  data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin']))
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
First NameLast NamePlace of birthDate of Birth T.A.
FrodoFrodoBagginsShire2968
BilboBilboBagginsShire2890
AragornAragorn IIElessarEriador2931
SamSamwiseGamgeeShire2980
PippinPeregrinTookShire2990
+
+
+

Here are other examples where we use the DataFrame functionality to handle arrays, now with more interesting features for us, namely numbers. We set up a matrix +of dimensionality \(10\times 5\) and compute the mean value and standard deviation of each column. Similarly, we can perform mathematial operations like squaring the matrix elements and many other operations.

+
+
+
import numpy as np
+import pandas as pd
+from IPython.display import display
+np.random.seed(100)
+# setting up a 10 x 5 matrix
+rows = 10
+cols = 5
+a = np.random.randn(rows,cols)
+df = pd.DataFrame(a)
+display(df)
+print(df.mean())
+print(df.std())
+display(df**2)
+
+
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
01234
0-1.7497650.3426801.153036-0.2524360.981321
10.5142190.221180-1.070043-0.1894960.255001
2-0.4580270.435163-0.5835950.8168470.672721
3-0.104411-0.5312801.029733-0.438136-1.118318
41.6189821.541605-0.251879-0.8424360.184519
50.9370820.7310001.361556-0.3262380.055676
60.222400-1.443217-0.7563520.8164540.750445
7-0.4559471.189622-1.690617-1.356399-1.232435
8-0.544439-0.6681720.007315-0.6129391.299748
9-1.733096-0.9833100.357508-1.6135791.470714
+
0   -0.175300
+1    0.083527
+2   -0.044334
+3   -0.399836
+4    0.331939
+dtype: float64
+0    1.069584
+1    0.965548
+2    1.018232
+3    0.793167
+4    0.918992
+dtype: float64
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
01234
03.0616790.1174301.3294920.0637240.962990
10.2644210.0489201.1449930.0359090.065026
20.2097890.1893670.3405830.6672390.452553
30.0109020.2822591.0603490.1919631.250636
42.6211022.3765470.0634430.7096980.034047
50.8781230.5343621.8538350.1064310.003100
60.0494622.0828750.5720690.6665970.563167
70.2078881.4152012.8581851.8398181.518895
80.2964140.4464530.0000540.3756941.689345
93.0036200.9668990.1278122.6036362.162999
+
+
+

Thereafter we can select specific columns only and plot final results

+
+
+
df.columns = ['First', 'Second', 'Third', 'Fourth', 'Fifth']
+df.index = np.arange(10)
+
+display(df)
+print(df['Second'].mean() )
+print(df.info())
+print(df.describe())
+
+from pylab import plt, mpl
+plt.style.use('seaborn')
+mpl.rcParams['font.family'] = 'serif'
+
+df.cumsum().plot(lw=2.0, figsize=(10,6))
+plt.show()
+
+
+df.plot.bar(figsize=(10,6), rot=15)
+plt.show()
+
+
+
+
+
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FirstSecondThirdFourthFifth
0-1.7497650.3426801.153036-0.2524360.981321
10.5142190.221180-1.070043-0.1894960.255001
2-0.4580270.435163-0.5835950.8168470.672721
3-0.104411-0.5312801.029733-0.438136-1.118318
41.6189821.541605-0.251879-0.8424360.184519
50.9370820.7310001.361556-0.3262380.055676
60.222400-1.443217-0.7563520.8164540.750445
7-0.4559471.189622-1.690617-1.356399-1.232435
8-0.544439-0.6681720.007315-0.6129391.299748
9-1.733096-0.9833100.357508-1.6135791.470714
+
0.08352721390288316
+<class 'pandas.core.frame.DataFrame'>
+Int64Index: 10 entries, 0 to 9
+Data columns (total 5 columns):
+ #   Column  Non-Null Count  Dtype  
+---  ------  --------------  -----  
+ 0   First   10 non-null     float64
+ 1   Second  10 non-null     float64
+ 2   Third   10 non-null     float64
+ 3   Fourth  10 non-null     float64
+ 4   Fifth   10 non-null     float64
+dtypes: float64(5)
+memory usage: 480.0 bytes
+None
+           First     Second      Third     Fourth      Fifth
+count  10.000000  10.000000  10.000000  10.000000  10.000000
+mean   -0.175300   0.083527  -0.044334  -0.399836   0.331939
+std     1.069584   0.965548   1.018232   0.793167   0.918992
+min    -1.749765  -1.443217  -1.690617  -1.613579  -1.232435
+25%    -0.522836  -0.633949  -0.713163  -0.785061   0.087887
+50%    -0.280179   0.281930  -0.122282  -0.382187   0.463861
+75%     0.441264   0.657041   0.861676  -0.205231   0.923602
+max     1.618982   1.541605   1.361556   0.816847   1.470714
+
+
+_images/week34_86_2.png +_images/week34_86_3.png +
+
+

We can produce a \(4\times 4\) matrix

+
+
+
b = np.arange(16).reshape((4,4))
+print(b)
+df1 = pd.DataFrame(b)
+print(df1)
+
+
+
+
+
[[ 0  1  2  3]
+ [ 4  5  6  7]
+ [ 8  9 10 11]
+ [12 13 14 15]]
+    0   1   2   3
+0   0   1   2   3
+1   4   5   6   7
+2   8   9  10  11
+3  12  13  14  15
+
+
+
+
+

and many other operations.

+

The Series class is another important class included in +pandas. You can view it as a specialization of DataFrame but where +we have just a single column of data. It shares many of the same features as DataFrame. As with DataFrame, +most operations are vectorized, achieving thereby a high performance when dealing with computations of arrays, in particular labeled arrays. +As we will see below it leads also to a very concice code close to the mathematical operations we may be interested in. +For multidimensional arrays, we recommend strongly xarray. xarray has much of the same flexibility as pandas, but allows for the extension to higher dimensions than two. We will see examples later of the usage of both pandas and xarray.

+
+

20.37.1. Simple linear regression model using scikit-learn

+

We start with perhaps our simplest possible example, using Scikit-Learn to perform linear regression analysis on a data set produced by us.

+

What follows is a simple Python code where we have defined a function +\(y\) in terms of the variable \(x\). Both are defined as vectors with \(100\) entries. +The numbers in the vector \(\boldsymbol{x}\) are given +by random numbers generated with a uniform distribution with entries +\(x_i \in [0,1]\) (more about probability distribution functions +later). These values are then used to define a function \(y(x)\) +(tabulated again as a vector) with a linear dependence on \(x\) plus a +random noise added via the normal distribution.

+

The Numpy functions are imported used the import numpy as np +statement and the random number generator for the uniform distribution +is called using the function np.random.rand(), where we specificy +that we want \(100\) random variables. Using Numpy we define +automatically an array with the specified number of elements, \(100\) in +our case. With the Numpy function randn() we can compute random +numbers with the normal distribution (mean value \(\mu\) equal to zero and +variance \(\sigma^2\) set to one) and produce the values of \(y\) assuming a linear +dependence as function of \(x\)

+
+\[ +y = 2x+N(0,1), +\]
+

where \(N(0,1)\) represents random numbers generated by the normal +distribution. From Scikit-Learn we import then the +LinearRegression functionality and make a prediction \(\tilde{y} = +\alpha + \beta x\) using the function fit(x,y). We call the set of +data \((\boldsymbol{x},\boldsymbol{y})\) for our training data. The Python package +scikit-learn has also a functionality which extracts the above +fitting parameters \(\alpha\) and \(\beta\) (see below). Later we will +distinguish between training data and test data.

+

For plotting we use the Python package +matplotlib which produces publication +quality figures. Feel free to explore the extensive +gallery of examples. In +this example we plot our original values of \(x\) and \(y\) as well as the +prediction ypredict (\(\tilde{y}\)), which attempts at fitting our +data with a straight line.

+

The Python code follows here.

+
+
+
# Importing various packages
+import numpy as np
+import matplotlib.pyplot as plt
+from sklearn.linear_model import LinearRegression
+
+x = np.random.rand(100,1)
+y = 2*x+np.random.randn(100,1)
+linreg = LinearRegression()
+linreg.fit(x,y)
+xnew = np.array([[0],[1]])
+ypredict = linreg.predict(xnew)
+
+plt.plot(xnew, ypredict, "r-")
+plt.plot(x, y ,'ro')
+plt.axis([0,1.0,0, 5.0])
+plt.xlabel(r'$x$')
+plt.ylabel(r'$y$')
+plt.title(r'Simple Linear Regression')
+plt.show()
+
+
+
+
+_images/week34_93_0.png +
+
+

This example serves several aims. It allows us to demonstrate several +aspects of data analysis and later machine learning algorithms. The +immediate visualization shows that our linear fit is not +impressive. It goes through the data points, but there are many +outliers which are not reproduced by our linear regression. We could +now play around with this small program and change for example the +factor in front of \(x\) and the normal distribution. Try to change the +function \(y\) to

+
+\[ +y = 10x+0.01 \times N(0,1), +\]
+

where \(x\) is defined as before. Does the fit look better? Indeed, by +reducing the role of the noise given by the normal distribution we see immediately that +our linear prediction seemingly reproduces better the training +set. However, this testing ‘by the eye’ is obviouly not satisfactory in the +long run. Here we have only defined the training data and our model, and +have not discussed a more rigorous approach to the cost function.

+

We need more rigorous criteria in defining whether we have succeeded or +not in modeling our training data. You will be surprised to see that +many scientists seldomly venture beyond this ‘by the eye’ approach. A +standard approach for the cost function is the so-called \(\chi^2\) +function (a variant of the mean-squared error (MSE))

+
+\[ +\chi^2 = \frac{1}{n} +\sum_{i=0}^{n-1}\frac{(y_i-\tilde{y}_i)^2}{\sigma_i^2}, +\]
+

where \(\sigma_i^2\) is the variance (to be defined later) of the entry +\(y_i\). We may not know the explicit value of \(\sigma_i^2\), it serves +however the aim of scaling the equations and make the cost function +dimensionless.

+

Minimizing the cost function is a central aspect of +our discussions to come. Finding its minima as function of the model +parameters (\(\alpha\) and \(\beta\) in our case) will be a recurring +theme in these series of lectures. Essentially all machine learning +algorithms we will discuss center around the minimization of the +chosen cost function. This depends in turn on our specific +model for describing the data, a typical situation in supervised +learning. Automatizing the search for the minima of the cost function is a +central ingredient in all algorithms. Typical methods which are +employed are various variants of gradient methods. These will be +discussed in more detail later. Again, you’ll be surprised to hear that +many practitioners minimize the above function ‘’by the eye’, popularly dubbed as +‘chi by the eye’. That is, change a parameter and see (visually and numerically) that +the \(\chi^2\) function becomes smaller.

+

There are many ways to define the cost function. A simpler approach is to look at the relative difference between the training data and the predicted data, that is we define +the relative error (why would we prefer the MSE instead of the relative error?) as

+
+\[ +\epsilon_{\mathrm{relative}}= \frac{\vert \boldsymbol{y} -\boldsymbol{\tilde{y}}\vert}{\vert \boldsymbol{y}\vert}. +\]
+

The squared cost function results in an arithmetic mean-unbiased +estimator, and the absolute-value cost function results in a +median-unbiased estimator (in the one-dimensional case, and a +geometric median-unbiased estimator for the multi-dimensional +case). The squared cost function has the disadvantage that it has the tendency +to be dominated by outliers.

+

We can modify easily the above Python code and plot the relative error instead

+
+
+
import numpy as np
+import matplotlib.pyplot as plt
+from sklearn.linear_model import LinearRegression
+
+x = np.random.rand(100,1)
+y = 5*x+0.01*np.random.randn(100,1)
+linreg = LinearRegression()
+linreg.fit(x,y)
+ypredict = linreg.predict(x)
+
+plt.plot(x, np.abs(ypredict-y)/abs(y), "ro")
+plt.axis([0,1.0,0.0, 0.5])
+plt.xlabel(r'$x$')
+plt.ylabel(r'$\epsilon_{\mathrm{relative}}$')
+plt.title(r'Relative error')
+plt.show()
+
+
+
+
+_images/week34_101_0.png +
+
+

Depending on the parameter in front of the normal distribution, we may +have a small or larger relative error. Try to play around with +different training data sets and study (graphically) the value of the +relative error.

+

As mentioned above, Scikit-Learn has an impressive functionality. +We can for example extract the values of \(\alpha\) and \(\beta\) and +their error estimates, or the variance and standard deviation and many +other properties from the statistical data analysis.

+

Here we show an +example of the functionality of Scikit-Learn.

+
+
+
import numpy as np 
+import matplotlib.pyplot as plt 
+from sklearn.linear_model import LinearRegression 
+from sklearn.metrics import mean_squared_error, r2_score, mean_squared_log_error, mean_absolute_error
+
+x = np.random.rand(100,1)
+y = 2.0+ 5*x+0.5*np.random.randn(100,1)
+linreg = LinearRegression()
+linreg.fit(x,y)
+ypredict = linreg.predict(x)
+print('The intercept alpha: \n', linreg.intercept_)
+print('Coefficient beta : \n', linreg.coef_)
+# The mean squared error                               
+print("Mean squared error: %.2f" % mean_squared_error(y, ypredict))
+# Explained variance score: 1 is perfect prediction                                 
+print('Variance score: %.2f' % r2_score(y, ypredict))
+# Mean squared log error                                                        
+print('Mean squared log error: %.2f' % mean_squared_log_error(y, ypredict) )
+# Mean absolute error                                                           
+print('Mean absolute error: %.2f' % mean_absolute_error(y, ypredict))
+plt.plot(x, ypredict, "r-")
+plt.plot(x, y ,'ro')
+plt.axis([0.0,1.0,1.5, 7.0])
+plt.xlabel(r'$x$')
+plt.ylabel(r'$y$')
+plt.title(r'Linear Regression fit ')
+plt.show()
+
+
+
+
+
The intercept alpha: 
+ [2.18780801]
+Coefficient beta : 
+ [[4.72228205]]
+Mean squared error: 0.37
+Variance score: 0.83
+Mean squared log error: 0.01
+Mean absolute error: 0.47
+
+
+_images/week34_103_1.png +
+
+

The function coef gives us the parameter \(\beta\) of our fit while intercept yields +\(\alpha\). Depending on the constant in front of the normal distribution, we get values near or far from \(\alpha =2\) and \(\beta =5\). Try to play around with different parameters in front of the normal distribution. The function meansquarederror gives us the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error or loss defined as

+
+\[ +MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +\]
+

The smaller the value, the better the fit. Ideally we would like to +have an MSE equal zero. The attentive reader has probably recognized +this function as being similar to the \(\chi^2\) function defined above.

+

The r2score function computes \(R^2\), the coefficient of +determination. It provides a measure of how well future samples are +likely to be predicted by the model. Best possible score is 1.0 and it +can be negative (because the model can be arbitrarily worse). A +constant model that always predicts the expected value of \(\boldsymbol{y}\), +disregarding the input features, would get a \(R^2\) score of \(0.0\).

+

If \(\tilde{\boldsymbol{y}}_i\) is the predicted value of the \(i-th\) sample and \(y_i\) is the corresponding true value, then the score \(R^2\) is defined as

+
+\[ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +\]
+

where we have defined the mean value of \(\boldsymbol{y}\) as

+
+\[ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +\]
+

Another quantity taht we will meet again in our discussions of regression analysis is +the mean absolute error (MAE), a risk metric corresponding to the expected value of the absolute error loss or what we call the \(l1\)-norm loss. In our discussion above we presented the relative error. +The MAE is defined as follows

+
+\[ +\text{MAE}(\boldsymbol{y}, \boldsymbol{\tilde{y}}) = \frac{1}{n} \sum_{i=0}^{n-1} \left| y_i - \tilde{y}_i \right|. +\]
+

We present the +squared logarithmic (quadratic) error

+
+\[ +\text{MSLE}(\boldsymbol{y}, \boldsymbol{\tilde{y}}) = \frac{1}{n} \sum_{i=0}^{n - 1} (\log_e (1 + y_i) - \log_e (1 + \tilde{y}_i) )^2, +\]
+

where \(\log_e (x)\) stands for the natural logarithm of \(x\). This error +estimate is best to use when targets having exponential growth, such +as population counts, average sales of a commodity over a span of +years etc.

+

Finally, another cost function is the Huber cost function used in robust regression.

+

The rationale behind this possible cost function is its reduced +sensitivity to outliers in the data set. In our discussions on +dimensionality reduction and normalization of data we will meet other +ways of dealing with outliers.

+

The Huber cost function is defined as

+
+\[\begin{split} +H_{\delta}(\boldsymbol{a})=\left\{\begin{array}{cc}\frac{1}{2} \boldsymbol{a}^{2}& \text{for }|\boldsymbol{a}|\leq \delta\\ \delta (|\boldsymbol{a}|-\frac{1}{2}\delta ),&\text{otherwise}.\end{array}\right. +\end{split}\]
+

Here \(\boldsymbol{a}=\boldsymbol{y} - \boldsymbol{\tilde{y}}\).

+

We will discuss in more detail these and other functions in the +various lectures and lab sessions.

+
+
+

20.37.2. To our real data: nuclear binding energies. Brief reminder on masses and binding energies

+

Let us now dive into nuclear physics and remind ourselves briefly about some basic features about binding +energies. A basic quantity which can be measured for the ground +states of nuclei is the atomic mass \(M(N, Z)\) of the neutral atom with +atomic mass number \(A\) and charge \(Z\). The number of neutrons is \(N\). There are indeed several sophisticated experiments worldwide which allow us to measure this quantity to high precision (parts per million even).

+

Atomic masses are usually tabulated in terms of the mass excess defined by

+
+\[ +\Delta M(N, Z) = M(N, Z) - uA, +\]
+

where \(u\) is the Atomic Mass Unit

+
+\[ +u = M(^{12}\mathrm{C})/12 = 931.4940954(57) \hspace{0.1cm} \mathrm{MeV}/c^2. +\]
+

The nucleon masses are

+
+\[ +m_p = 1.00727646693(9)u, +\]
+

and

+
+\[ +m_n = 939.56536(8)\hspace{0.1cm} \mathrm{MeV}/c^2 = 1.0086649156(6)u. +\]
+

In the 2016 mass evaluation of by W.J.Huang, G.Audi, M.Wang, F.G.Kondev, S.Naimi and X.Xu +there are data on masses and decays of 3437 nuclei.

+

The nuclear binding energy is defined as the energy required to break +up a given nucleus into its constituent parts of \(N\) neutrons and \(Z\) +protons. In terms of the atomic masses \(M(N, Z)\) the binding energy is +defined by

+
+\[ +BE(N, Z) = ZM_H c^2 + Nm_n c^2 - M(N, Z)c^2 , +\]
+

where \(M_H\) is the mass of the hydrogen atom and \(m_n\) is the mass of the neutron. +In terms of the mass excess the binding energy is given by

+
+\[ +BE(N, Z) = Z\Delta_H c^2 + N\Delta_n c^2 -\Delta(N, Z)c^2 , +\]
+

where \(\Delta_H c^2 = 7.2890\) MeV and \(\Delta_n c^2 = 8.0713\) MeV.

+

A popular and physically intuitive model which can be used to parametrize +the experimental binding energies as function of \(A\), is the so-called +liquid drop model. The ansatz is based on the following expression

+
+\[ +BE(N,Z) = a_1A-a_2A^{2/3}-a_3\frac{Z^2}{A^{1/3}}-a_4\frac{(N-Z)^2}{A}, +\]
+

where \(A\) stands for the number of nucleons and the \(a_i\)s are parameters which are determined by a fit +to the experimental data.

+

To arrive at the above expression we have assumed that we can make the following assumptions:

+
    +

  • There is a volume term \(a_1A\) proportional with the number of nucleons (the energy is also an extensive quantity). When an assembly of nucleons of the same size is packed together into the smallest volume, each interior nucleon has a certain number of other nucleons in contact with it. This contribution is proportional to the volume.

    • +

  • There is a surface energy term \(a_2A^{2/3}\). The assumption here is that a nucleon at the surface of a nucleus interacts with fewer other nucleons than one in the interior of the nucleus and hence its binding energy is less. This surface energy term takes that into account and is therefore negative and is proportional to the surface area.

    • +

  • There is a Coulomb energy term \(a_3\frac{Z^2}{A^{1/3}}\). The electric repulsion between each pair of protons in a nucleus yields less binding.

    • +

  • There is an asymmetry term \(a_4\frac{(N-Z)^2}{A}\). This term is associated with the Pauli exclusion principle and reflects the fact that the proton-neutron interaction is more attractive on the average than the neutron-neutron and proton-proton interactions.

    • +
+

We could also add a so-called pairing term, which is a correction term that +arises from the tendency of proton pairs and neutron pairs to +occur. An even number of particles is more stable than an odd number.

+
+
+

20.37.3. Organizing our data

+

Let us start with reading and organizing our data. +We start with the compilation of masses and binding energies from 2016. +After having downloaded this file to our own computer, we are now ready to read the file and start structuring our data.

+

We start with preparing folders for storing our calculations and the data file over masses and binding energies. We import also various modules that we will find useful in order to present various Machine Learning methods. Here we focus mainly on the functionality of scikit-learn.

+
+
+
# Common imports
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import sklearn.linear_model as skl
+from sklearn.model_selection import train_test_split
+from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error
+import os
+
+# Where to save the figures and data files
+PROJECT_ROOT_DIR = "Results"
+FIGURE_ID = "Results/FigureFiles"
+DATA_ID = "DataFiles/"
+
+if not os.path.exists(PROJECT_ROOT_DIR):
+    os.mkdir(PROJECT_ROOT_DIR)
+
+if not os.path.exists(FIGURE_ID):
+    os.makedirs(FIGURE_ID)
+
+if not os.path.exists(DATA_ID):
+    os.makedirs(DATA_ID)
+
+def image_path(fig_id):
+    return os.path.join(FIGURE_ID, fig_id)
+
+def data_path(dat_id):
+    return os.path.join(DATA_ID, dat_id)
+
+def save_fig(fig_id):
+    plt.savefig(image_path(fig_id) + ".png", format='png')
+
+infile = open(data_path("MassEval2016.dat"),'r')
+
+
+
+
+

Before we proceed, we define also a function for making our plots. You can obviously avoid this and simply set up various matplotlib commands every time you need them. You may however find it convenient to collect all such commands in one function and simply call this function.

+
+
+
from pylab import plt, mpl
+plt.style.use('seaborn')
+mpl.rcParams['font.family'] = 'serif'
+
+def MakePlot(x,y, styles, labels, axlabels):
+    plt.figure(figsize=(10,6))
+    for i in range(len(x)):
+        plt.plot(x[i], y[i], styles[i], label = labels[i])
+        plt.xlabel(axlabels[0])
+        plt.ylabel(axlabels[1])
+    plt.legend(loc=0)
+
+
+
+
+

Our next step is to read the data on experimental binding energies and +reorganize them as functions of the mass number \(A\), the number of +protons \(Z\) and neutrons \(N\) using pandas. Before we do this it is +always useful (unless you have a binary file or other types of compressed +data) to actually open the file and simply take a look at it!

+

In particular, the program that outputs the final nuclear masses is written in Fortran with a specific format. It means that we need to figure out the format and which columns contain the data we are interested in. Pandas comes with a function that reads formatted output. After having admired the file, we are now ready to start massaging it with pandas. The file begins with some basic format information.

+
+
+
"""                                                                                                                         
+This is taken from the data file of the mass 2016 evaluation.                                                               
+All files are 3436 lines long with 124 character per line.                                                                  
+       Headers are 39 lines long.                                                                                           
+   col 1     :  Fortran character control: 1 = page feed  0 = line feed                                                     
+   format    :  a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5                     
+   These formats are reflected in the pandas widths variable below, see the statement                                       
+   widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),                                                            
+   Pandas has also a variable header, with length 39 in this case.                                                          
+"""
+
+
+
+
+
'                                                                                                                         \nThis is taken from the data file of the mass 2016 evaluation.                                                               \nAll files are 3436 lines long with 124 character per line.                                                                  \n       Headers are 39 lines long.                                                                                           \n   col 1     :  Fortran character control: 1 = page feed  0 = line feed                                                     \n   format    :  a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5                     \n   These formats are reflected in the pandas widths variable below, see the statement                                       \n   widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),                                                            \n   Pandas has also a variable header, with length 39 in this case.                                                          \n'
+
+
+
+
+

The data we are interested in are in columns 2, 3, 4 and 11, giving us +the number of neutrons, protons, mass numbers and binding energies, +respectively. We add also for the sake of completeness the element name. The data are in fixed-width formatted lines and we will +covert them into the pandas DataFrame structure.

+
+
+
# Read the experimental data with Pandas
+Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),
+              names=('N', 'Z', 'A', 'Element', 'Ebinding'),
+              widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),
+              header=39,
+              index_col=False)
+
+# Extrapolated values are indicated by '#' in place of the decimal place, so
+# the Ebinding column won't be numeric. Coerce to float and drop these entries.
+Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')
+Masses = Masses.dropna()
+# Convert from keV to MeV.
+Masses['Ebinding'] /= 1000
+
+# Group the DataFrame by nucleon number, A.
+Masses = Masses.groupby('A')
+# Find the rows of the grouped DataFrame with the maximum binding energy.
+Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])
+
+
+
+
+
---------------------------------------------------------------------------
+ValueError                                Traceback (most recent call last)
+Input In [30], in <cell line: 2>()
+      1 # Read the experimental data with Pandas
+----> 2 Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),
+      3               names=('N', 'Z', 'A', 'Element', 'Ebinding'),
+      4               widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),
+      5               header=39,
+      6               index_col=False)
+      8 # Extrapolated values are indicated by '#' in place of the decimal place, so
+      9 # the Ebinding column won't be numeric. Coerce to float and drop these entries.
+     10 Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')
+
+File ~/miniforge3/envs/myenv/lib/python3.9/site-packages/pandas/util/_decorators.py:311, in deprecate_nonkeyword_arguments.<locals>.decorate.<locals>.wrapper(*args, **kwargs)
+    305 if len(args) > num_allow_args:
+    306     warnings.warn(
+    307         msg.format(arguments=arguments),
+    308         FutureWarning,
+    309         stacklevel=stacklevel,
+    310     )
+--> 311 return func(*args, **kwargs)
+
+File ~/miniforge3/envs/myenv/lib/python3.9/site-packages/pandas/io/parsers/readers.py:871, in read_fwf(filepath_or_buffer, colspecs, widths, infer_nrows, **kwds)
+    869                     len_index = len(index_col)
+    870         if len(names) + len_index != len(colspecs):
+--> 871             raise ValueError("Length of colspecs must match length of names")
+    873 kwds["colspecs"] = colspecs
+    874 kwds["infer_nrows"] = infer_nrows
+
+ValueError: Length of colspecs must match length of names
+
+
+
+
+

We have now read in the data, grouped them according to the variables we are interested in. +We see how easy it is to reorganize the data using pandas. If we +were to do these operations in C/C++ or Fortran, we would have had to +write various functions/subroutines which perform the above +reorganizations for us. Having reorganized the data, we can now start +to make some simple fits using both the functionalities in numpy and +Scikit-Learn afterwards.

+

Now we define five variables which contain +the number of nucleons \(A\), the number of protons \(Z\) and the number of neutrons \(N\), the element name and finally the energies themselves.

+
+
+
A = Masses['A']
+Z = Masses['Z']
+N = Masses['N']
+Element = Masses['Element']
+Energies = Masses['Ebinding']
+print(Masses)
+
+
+
+
+

The next step, and we will define this mathematically later, is to set up the so-called design matrix. We will throughout call this matrix \(\boldsymbol{X}\). +It has dimensionality \(p\times n\), where \(n\) is the number of data points and \(p\) are the so-called predictors. In our case here they are given by the number of polynomials in \(A\) we wish to include in the fit.

+
+
+
# Now we set up the design matrix X
+X = np.zeros((len(A),5))
+X[:,0] = 1
+X[:,1] = A
+X[:,2] = A**(2.0/3.0)
+X[:,3] = A**(-1.0/3.0)
+X[:,4] = A**(-1.0)
+
+
+
+
+

With scikitlearn we are now ready to use linear regression and fit our data.

+
+
+
clf = skl.LinearRegression().fit(X, Energies)
+fity = clf.predict(X)
+
+
+
+
+

Pretty simple!
+Now we can print measures of how our fit is doing, the coefficients from the fits and plot the final fit together with our data.

+
+
+
# The mean squared error                               
+print("Mean squared error: %.2f" % mean_squared_error(Energies, fity))
+# Explained variance score: 1 is perfect prediction                                 
+print('Variance score: %.2f' % r2_score(Energies, fity))
+# Mean absolute error                                                           
+print('Mean absolute error: %.2f' % mean_absolute_error(Energies, fity))
+print(clf.coef_, clf.intercept_)
+
+Masses['Eapprox']  = fity
+# Generate a plot comparing the experimental with the fitted values values.
+fig, ax = plt.subplots()
+ax.set_xlabel(r'$A = N + Z$')
+ax.set_ylabel(r'$E_\mathrm{bind}\,/\mathrm{MeV}$')
+ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,
+            label='Ame2016')
+ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',
+            label='Fit')
+ax.legend()
+save_fig("Masses2016")
+plt.show()
+
+
+
+
+
+
+

20.37.4. And what about using neural networks?

+

The seaborn package allows us to visualize data in an efficient way. Note that we use scikit-learn’s multi-layer perceptron (or feed forward neural network) +functionality.

+
+
+
from sklearn.neural_network import MLPRegressor
+from sklearn.metrics import accuracy_score
+import seaborn as sns
+
+X_train = X
+Y_train = Energies
+n_hidden_neurons = 100
+epochs = 100
+# store models for later use
+eta_vals = np.logspace(-5, 1, 7)
+lmbd_vals = np.logspace(-5, 1, 7)
+# store the models for later use
+DNN_scikit = np.zeros((len(eta_vals), len(lmbd_vals)), dtype=object)
+train_accuracy = np.zeros((len(eta_vals), len(lmbd_vals)))
+sns.set()
+for i, eta in enumerate(eta_vals):
+    for j, lmbd in enumerate(lmbd_vals):
+        dnn = MLPRegressor(hidden_layer_sizes=(n_hidden_neurons), activation='logistic',
+                            alpha=lmbd, learning_rate_init=eta, max_iter=epochs)
+        dnn.fit(X_train, Y_train)
+        DNN_scikit[i][j] = dnn
+        train_accuracy[i][j] = dnn.score(X_train, Y_train)
+
+fig, ax = plt.subplots(figsize = (10, 10))
+sns.heatmap(train_accuracy, annot=True, ax=ax, cmap="viridis")
+ax.set_title("Training Accuracy")
+ax.set_ylabel("$\eta$")
+ax.set_xlabel("$\lambda$")
+plt.show()
+
+
+
+
+
+
+
+

20.38. A first summary

+

The aim behind these introductory words was to present to you various +Python libraries and their functionalities, in particular libraries like +numpy, pandas, xarray and matplotlib and other that make our life much easier +in handling various data sets and visualizing data.

+

Furthermore, +Scikit-Learn allows us with few lines of code to implement popular +Machine Learning algorithms for supervised learning. Later we will meet Tensorflow, a powerful library for deep learning. +Now it is time to dive more into the details of various methods. We will start with linear regression and try to take a deeper look at what it entails.

+
+
+

20.39. Why Linear Regression (aka Ordinary Least Squares and family)

+

Fitting a continuous function with linear parameterization in terms of the parameters \(\boldsymbol{\beta}\).

+
    +

  • Method of choice for fitting a continuous function!

    • +

  • Gives an excellent introduction to central Machine Learning features with understandable pedagogical links to other methods like Neural Networks, Support Vector Machines etc

    • +

  • Analytical expression for the fitting parameters \(\boldsymbol{\beta}\)

    • +

  • Analytical expressions for statistical propertiers like mean values, variances, confidence intervals and more

    • +

  • Analytical relation with probabilistic interpretations

    • +

  • Easy to introduce basic concepts like bias-variance tradeoff, cross-validation, resampling and regularization techniques and many other ML topics

    • +

  • Easy to code! And links well with classification problems and logistic regression and neural networks

    • +

  • Allows for easy hands-on understanding of gradient descent methods

    • +

  • and many more features

    • +
+

For more discussions of Ridge and Lasso regression, Wessel van Wieringen’s article is highly recommended. +Similarly, Mehta et al’s article is also recommended.

+
+
+

20.40. Regression analysis, overarching aims

+

Regression modeling deals with the description of the sampling distribution of a given random variable \(y\) and how it varies as function of another variable or a set of such variables \(\boldsymbol{x} =[x_0, x_1,\dots, x_{n-1}]^T\). +The first variable is called the dependent, the outcome or the response variable while the set of variables \(\boldsymbol{x}\) is called the independent variable, or the predictor variable or the explanatory variable, or simply just the inputs.

+

A regression model aims at finding a likelihood function \(p(\boldsymbol{y}\vert \boldsymbol{x})\) or in the more traditional sense a function \(\boldsymbol{y}(\boldsymbol{x})\), that is the conditional distribution for \(\boldsymbol{y}\) with a given \(\boldsymbol{x}\). The estimation of \(p(\boldsymbol{y}\vert \boldsymbol{x})\) is made using a data set with

+
    +

  • \(n\) cases \(i = 0, 1, 2, \dots, n-1\)

    • +

  • Response (target, dependent or outcome) variable \(y_i\) with \(i = 0, 1, 2, \dots, n-1\)

    • +

  • \(p\) so-called explanatory (independent or predictor or feature) variables \(\boldsymbol{x}_i=[x_{i0}, x_{i1}, \dots, x_{ip-1}]\) with \(i = 0, 1, 2, \dots, n-1\) and explanatory variables running from \(0\) to \(p-1\). See below for more explicit examples.

    • +
+

The goal of the regression analysis is to extract/exploit relationship between \(\boldsymbol{y}\) and \(\boldsymbol{x}\) in order to infer specific dependencies, approximations to the likelihood functions, functional relationships and to make predictions, making fits and many other things.

+
+
+

20.41. Regression analysis, overarching aims II

+

Consider an experiment in which \(p\) characteristics/features of \(n\) samples are +measured. The data from this experiment, for various explanatory variables \(p\) are normally represented by a matrix
+\(\mathbf{X}\).

+

The matrix \(\mathbf{X}\) is called the design +matrix. Additional information of the samples is available in the +form of \(\boldsymbol{y}\) (also as above). The variable \(\boldsymbol{y}\) is +generally referred to as the response variable. The aim of +regression analysis is to explain \(\boldsymbol{y}\) in terms of +\(\boldsymbol{X}\) through a functional relationship like \(y_i = +f(\mathbf{X}_{i,\ast})\). When no prior knowledge on the form of +\(f(\cdot)\) is available, it is common to assume a linear relationship +between \(\boldsymbol{X}\) and \(\boldsymbol{y}\). This assumption gives rise to +the linear regression model where \(\boldsymbol{\beta} = [\beta_0, \ldots, +\beta_{p-1}]^{T}\) are the regression parameters.

+

Linear regression gives us a set of analytical equations for the parameters \(\beta_j\).

+
+
+

20.42. Examples

+

In order to understand the relation among the predictors (or features or properties) \(p\), the set of data \(n\) and the target (outcome, output etc) \(\boldsymbol{y}\), +consider the model we discussed for describing nuclear binding energies.

+

There we assumed that we could parametrize the data using a polynomial approximation based on the liquid drop model. +Assuming

+
+\[ +BE(A) = a_0+a_1A+a_2A^{2/3}+a_3A^{-1/3}+a_4A^{-1}, +\]
+

we have five predictors, that is the intercept, the \(A\) dependent term, the \(A^{2/3}\) term and the \(A^{-1/3}\) and \(A^{-1}\) terms. +This gives \(p=0,1,2,3,4\). Furthermore we have \(n\) entries for each predictor. It means that our design matrix is a +\(p\times n\) matrix \(\boldsymbol{X}\).

+

Here the predictors are based on a model we have made. A popular data set which is widely encountered in ML applications is the +so-called credit card default data from Taiwan. The data set contains data on \(n=30000\) credit card holders with predictors like gender, marital status, age, profession, education, etc. In total there are \(24\) such predictors or attributes leading to a design matrix of dimensionality \(24 \times 30000\). This is however a classification problem and we will come back to it when we discuss Logistic Regression.

+
+
+

20.43. General linear models and linear algebra

+

Before we proceed let us study a case where we aim at fitting a set of data \(\boldsymbol{y}=[y_0,y_1,\dots,y_{n-1}]\). We could think of these data as a result of an experiment or a complicated numerical experiment. These data are functions of a series of variables \(\boldsymbol{x}=[x_0,x_1,\dots,x_{n-1}]\), that is \(y_i = y(x_i)\) with \(i=0,1,2,\dots,n-1\). The variables \(x_i\) could represent physical quantities like time, temperature, position etc. We assume that \(y(x)\) is a smooth function.

+

Since obtaining these data points may not be trivial, we want to use these data to fit a function which can allow us to make predictions for values of \(y\) which are not in the present set. The perhaps simplest approach is to assume we can parametrize our function in terms of a polynomial of degree \(n-1\) with \(n\) points, that is

+
+\[ +y=y(x) \rightarrow y(x_i)=\tilde{y}_i+\epsilon_i=\sum_{j=0}^{n-1} \beta_j x_i^j+\epsilon_i, +\]
+

where \(\epsilon_i\) is the error in our approximation.

+
+
+

20.44. Rewriting the fitting procedure as a linear algebra problem

+

For every set of values \(y_i,x_i\) we have thus the corresponding set of equations

+
+\[\begin{split} +\begin{align*} +y_0&=\beta_0+\beta_1x_0^1+\beta_2x_0^2+\dots+\beta_{n-1}x_0^{n-1}+\epsilon_0\\ +y_1&=\beta_0+\beta_1x_1^1+\beta_2x_1^2+\dots+\beta_{n-1}x_1^{n-1}+\epsilon_1\\ +y_2&=\beta_0+\beta_1x_2^1+\beta_2x_2^2+\dots+\beta_{n-1}x_2^{n-1}+\epsilon_2\\ +\dots & \dots \\ +y_{n-1}&=\beta_0+\beta_1x_{n-1}^1+\beta_2x_{n-1}^2+\dots+\beta_{n-1}x_{n-1}^{n-1}+\epsilon_{n-1}.\\ +\end{align*} +\end{split}\]
+
+
+

20.45. Rewriting the fitting procedure as a linear algebra problem, more details

+

Defining the vectors

+
+\[ +\boldsymbol{y} = [y_0,y_1, y_2,\dots, y_{n-1}]^T, +\]
+

and

+
+\[ +\boldsymbol{\beta} = [\beta_0,\beta_1, \beta_2,\dots, \beta_{n-1}]^T, +\]
+

and

+
+\[ +\boldsymbol{\epsilon} = [\epsilon_0,\epsilon_1, \epsilon_2,\dots, \epsilon_{n-1}]^T, +\]
+

and the design matrix

+
+\[\begin{split} +\boldsymbol{X}= +\begin{bmatrix} +1& x_{0}^1 &x_{0}^2& \dots & \dots &x_{0}^{n-1}\\ +1& x_{1}^1 &x_{1}^2& \dots & \dots &x_{1}^{n-1}\\ +1& x_{2}^1 &x_{2}^2& \dots & \dots &x_{2}^{n-1}\\ +\dots& \dots &\dots& \dots & \dots &\dots\\ +1& x_{n-1}^1 &x_{n-1}^2& \dots & \dots &x_{n-1}^{n-1}\\ +\end{bmatrix} +\end{split}\]
+

we can rewrite our equations as

+
+\[ +\boldsymbol{y} = \boldsymbol{X}\boldsymbol{\beta}+\boldsymbol{\epsilon}. +\]
+

The above design matrix is called a Vandermonde matrix.

+
+
+

20.46. Generalizing the fitting procedure as a linear algebra problem

+

We are obviously not limited to the above polynomial expansions. We +could replace the various powers of \(x\) with elements of Fourier +series or instead of \(x_i^j\) we could have \(\cos{(j x_i)}\) or \(\sin{(j +x_i)}\), or time series or other orthogonal functions. For every set +of values \(y_i,x_i\) we can then generalize the equations to

+
+\[\begin{split} +\begin{align*} +y_0&=\beta_0x_{00}+\beta_1x_{01}+\beta_2x_{02}+\dots+\beta_{n-1}x_{0n-1}+\epsilon_0\\ +y_1&=\beta_0x_{10}+\beta_1x_{11}+\beta_2x_{12}+\dots+\beta_{n-1}x_{1n-1}+\epsilon_1\\ +y_2&=\beta_0x_{20}+\beta_1x_{21}+\beta_2x_{22}+\dots+\beta_{n-1}x_{2n-1}+\epsilon_2\\ +\dots & \dots \\ +y_{i}&=\beta_0x_{i0}+\beta_1x_{i1}+\beta_2x_{i2}+\dots+\beta_{n-1}x_{in-1}+\epsilon_i\\ +\dots & \dots \\ +y_{n-1}&=\beta_0x_{n-1,0}+\beta_1x_{n-1,2}+\beta_2x_{n-1,2}+\dots+\beta_{n-1}x_{n-1,n-1}+\epsilon_{n-1}.\\ +\end{align*} +\end{split}\]
+

Note that we have \(p=n\) here. The matrix is symmetric. This is generally not the case!

+
+
+

20.47. Generalizing the fitting procedure as a linear algebra problem

+

We redefine in turn the matrix \(\boldsymbol{X}\) as

+
+\[\begin{split} +\boldsymbol{X}= +\begin{bmatrix} +x_{00}& x_{01} &x_{02}& \dots & \dots &x_{0,n-1}\\ +x_{10}& x_{11} &x_{12}& \dots & \dots &x_{1,n-1}\\ +x_{20}& x_{21} &x_{22}& \dots & \dots &x_{2,n-1}\\ +\dots& \dots &\dots& \dots & \dots &\dots\\ +x_{n-1,0}& x_{n-1,1} &x_{n-1,2}& \dots & \dots &x_{n-1,n-1}\\ +\end{bmatrix} +\end{split}\]
+

and without loss of generality we rewrite again our equations as

+
+\[ +\boldsymbol{y} = \boldsymbol{X}\boldsymbol{\beta}+\boldsymbol{\epsilon}. +\]
+

The left-hand side of this equation is kwown. Our error vector \(\boldsymbol{\epsilon}\) and the parameter vector \(\boldsymbol{\beta}\) are our unknow quantities. How can we obtain the optimal set of \(\beta_i\) values?

+
+
+

20.48. Optimizing our parameters

+

We have defined the matrix \(\boldsymbol{X}\) via the equations

+
+\[\begin{split} +\begin{align*} +y_0&=\beta_0x_{00}+\beta_1x_{01}+\beta_2x_{02}+\dots+\beta_{n-1}x_{0n-1}+\epsilon_0\\ +y_1&=\beta_0x_{10}+\beta_1x_{11}+\beta_2x_{12}+\dots+\beta_{n-1}x_{1n-1}+\epsilon_1\\ +y_2&=\beta_0x_{20}+\beta_1x_{21}+\beta_2x_{22}+\dots+\beta_{n-1}x_{2n-1}+\epsilon_1\\ +\dots & \dots \\ +y_{i}&=\beta_0x_{i0}+\beta_1x_{i1}+\beta_2x_{i2}+\dots+\beta_{n-1}x_{in-1}+\epsilon_1\\ +\dots & \dots \\ +y_{n-1}&=\beta_0x_{n-1,0}+\beta_1x_{n-1,2}+\beta_2x_{n-1,2}+\dots+\beta_{n-1}x_{n-1,n-1}+\epsilon_{n-1}.\\ +\end{align*} +\end{split}\]
+

As we noted above, we stayed with a system with the design matrix +\(\boldsymbol{X}\in {\mathbb{R}}^{n\times n}\), that is we have \(p=n\). For reasons to come later (algorithmic arguments) we will hereafter define +our matrix as \(\boldsymbol{X}\in {\mathbb{R}}^{n\times p}\), with the predictors refering to the column numbers and the entries \(n\) being the row elements.

+
+
+

20.49. Our model for the nuclear binding energies

+

In our introductory notes we looked at the so-called liquid drop model. Let us remind ourselves about what we did by looking at the code.

+

We restate the parts of the code we are most interested in.

+
+
+
# Common imports
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from IPython.display import display
+import os
+
+# Where to save the figures and data files
+PROJECT_ROOT_DIR = "Results"
+FIGURE_ID = "Results/FigureFiles"
+DATA_ID = "DataFiles/"
+
+if not os.path.exists(PROJECT_ROOT_DIR):
+    os.mkdir(PROJECT_ROOT_DIR)
+
+if not os.path.exists(FIGURE_ID):
+    os.makedirs(FIGURE_ID)
+
+if not os.path.exists(DATA_ID):
+    os.makedirs(DATA_ID)
+
+def image_path(fig_id):
+    return os.path.join(FIGURE_ID, fig_id)
+
+def data_path(dat_id):
+    return os.path.join(DATA_ID, dat_id)
+
+def save_fig(fig_id):
+    plt.savefig(image_path(fig_id) + ".png", format='png')
+
+infile = open(data_path("MassEval2016.dat"),'r')
+
+
+# Read the experimental data with Pandas
+Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),
+              names=('N', 'Z', 'A', 'Element', 'Ebinding'),
+              widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),
+              header=39,
+              index_col=False)
+
+# Extrapolated values are indicated by '#' in place of the decimal place, so
+# the Ebinding column won't be numeric. Coerce to float and drop these entries.
+Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')
+Masses = Masses.dropna()
+# Convert from keV to MeV.
+Masses['Ebinding'] /= 1000
+
+# Group the DataFrame by nucleon number, A.
+Masses = Masses.groupby('A')
+# Find the rows of the grouped DataFrame with the maximum binding energy.
+Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])
+A = Masses['A']
+Z = Masses['Z']
+N = Masses['N']
+Element = Masses['Element']
+Energies = Masses['Ebinding']
+
+# Now we set up the design matrix X
+X = np.zeros((len(A),5))
+X[:,0] = 1
+X[:,1] = A
+X[:,2] = A**(2.0/3.0)
+X[:,3] = A**(-1.0/3.0)
+X[:,4] = A**(-1.0)
+# Then nice printout using pandas
+DesignMatrix = pd.DataFrame(X)
+DesignMatrix.index = A
+DesignMatrix.columns = ['1', 'A', 'A^(2/3)', 'A^(-1/3)', '1/A']
+display(DesignMatrix)
+
+
+
+
+

With \(\boldsymbol{\beta}\in {\mathbb{R}}^{p\times 1}\), it means that we will hereafter write our equations for the approximation as

+
+\[ +\boldsymbol{\tilde{y}}= \boldsymbol{X}\boldsymbol{\beta}, +\]
+

throughout these lectures.

+
+
+

20.50. Optimizing our parameters, more details

+

With the above we use the design matrix to define the approximation \(\boldsymbol{\tilde{y}}\) via the unknown quantity \(\boldsymbol{\beta}\) as

+
+\[ +\boldsymbol{\tilde{y}}= \boldsymbol{X}\boldsymbol{\beta}, +\]
+

and in order to find the optimal parameters \(\beta_i\) instead of solving the above linear algebra problem, we define a function which gives a measure of the spread between the values \(y_i\) (which represent hopefully the exact values) and the parameterized values \(\tilde{y}_i\), namely

+
+\[ +C(\boldsymbol{\beta})=\frac{1}{n}\sum_{i=0}^{n-1}\left(y_i-\tilde{y}_i\right)^2=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)^T\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)\right\}, +\]
+

or using the matrix \(\boldsymbol{X}\) and in a more compact matrix-vector notation as

+
+\[ +C(\boldsymbol{\beta})=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}. +\]
+

This function is one possible way to define the so-called cost function.

+

It is also common to define +the function \(C\) as

+
+\[ +C(\boldsymbol{\beta})=\frac{1}{2n}\sum_{i=0}^{n-1}\left(y_i-\tilde{y}_i\right)^2, +\]
+

since when taking the first derivative with respect to the unknown parameters \(\beta\), the factor of \(2\) cancels out.

+
+
+

20.51. Interpretations and optimizing our parameters

+

The function

+
+\[ +C(\boldsymbol{\beta})=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}, +\]
+

can be linked to the variance of the quantity \(y_i\) if we interpret the latter as the mean value. +When linking (see the discussion below) with the maximum likelihood approach below, we will indeed interpret \(y_i\) as a mean value

+
+\[ +y_{i}=\langle y_i \rangle = \beta_0x_{i,0}+\beta_1x_{i,1}+\beta_2x_{i,2}+\dots+\beta_{n-1}x_{i,n-1}+\epsilon_i, +\]
+

where \(\langle y_i \rangle\) is the mean value. Keep in mind also that +till now we have treated \(y_i\) as the exact value. Normally, the +response (dependent or outcome) variable \(y_i\) the outcome of a +numerical experiment or another type of experiment and is thus only an +approximation to the true value. It is then always accompanied by an +error estimate, often limited to a statistical error estimate given by +the standard deviation discussed earlier. In the discussion here we +will treat \(y_i\) as our exact value for the response variable.

+

In order to find the parameters \(\beta_i\) we will then minimize the spread of \(C(\boldsymbol{\beta})\), that is we are going to solve the problem

+
+\[ +{\displaystyle \min_{\boldsymbol{\beta}\in +{\mathbb{R}}^{p}}}\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}. +\]
+

In practical terms it means we will require

+
+\[ +\frac{\partial C(\boldsymbol{\beta})}{\partial \beta_j} = \frac{\partial }{\partial \beta_j}\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}\right)^2\right]=0, +\]
+

which results in

+
+\[ +\frac{\partial C(\boldsymbol{\beta})}{\partial \beta_j} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}x_{ij}\left(y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}\right)\right]=0, +\]
+

or in a matrix-vector form as

+
+\[ +\frac{\partial C(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right). +\]
+
+
+

20.52. Interpretations and optimizing our parameters

+

We can rewrite

+
+\[ +\frac{\partial C(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right), +\]
+

as

+
+\[ +\boldsymbol{X}^T\boldsymbol{y} = \boldsymbol{X}^T\boldsymbol{X}\boldsymbol{\beta}, +\]
+

and if the matrix \(\boldsymbol{X}^T\boldsymbol{X}\) is invertible we have the solution

+
+\[ +\boldsymbol{\beta} =\left(\boldsymbol{X}^T\boldsymbol{X}\right)^{-1}\boldsymbol{X}^T\boldsymbol{y}. +\]
+

We note also that since our design matrix is defined as \(\boldsymbol{X}\in +{\mathbb{R}}^{n\times p}\), the product \(\boldsymbol{X}^T\boldsymbol{X} \in +{\mathbb{R}}^{p\times p}\). In the above case we have that \(p \ll n\), +in our case \(p=5\) meaning that we end up with inverting a small +\(5\times 5\) matrix. This is a rather common situation, in many cases we end up with low-dimensional +matrices to invert. The methods discussed here and for many other +supervised learning algorithms like classification with logistic +regression or support vector machines, exhibit dimensionalities which +allow for the usage of direct linear algebra methods such as LU decomposition or Singular Value Decomposition (SVD) for finding the inverse of the matrix +\(\boldsymbol{X}^T\boldsymbol{X}\).

+

Small question: Do you think the example we have at hand here (the nuclear binding energies) can lead to problems in inverting the matrix \(\boldsymbol{X}^T\boldsymbol{X}\)? What kind of problems can we expect?

+
+
+

20.53. Some useful matrix and vector expressions

+

See the handwritten notes at https://github.com/CompPhysics/MachineLearning/blob/master/doc/HandWrittenNotes/2022/NotesExercise5Week452022.pdf

+

These notes will be discussed during one of the lectures.

+
+
+

20.54. Interpretations and optimizing our parameters

+

The residuals \(\boldsymbol{\epsilon}\) are in turn given by

+
+\[ +\boldsymbol{\epsilon} = \boldsymbol{y}-\boldsymbol{\tilde{y}} = \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}, +\]
+

and with

+
+\[ +\boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)= 0, +\]
+

we have

+
+\[ +\boldsymbol{X}^T\boldsymbol{\epsilon}=\boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)= 0, +\]
+

meaning that the solution for \(\boldsymbol{\beta}\) is the one which minimizes the residuals. Later we will link this with the maximum likelihood approach.

+

Let us now return to our nuclear binding energies and simply code the above equations.

+
+
+

20.55. Own code for Ordinary Least Squares

+

It is rather straightforward to implement the matrix inversion and obtain the parameters \(\boldsymbol{\beta}\). After having defined the matrix \(\boldsymbol{X}\) we simply need to +write

+
+
+
# matrix inversion to find beta
+beta = np.linalg.inv(X.T.dot(X)).dot(X.T).dot(Energies)
+# and then make the prediction
+ytilde = X @ beta
+
+
+
+
+

Alternatively, you can use the least squares functionality in Numpy as

+
+
+
fit = np.linalg.lstsq(X, Energies, rcond =None)[0]
+ytildenp = np.dot(fit,X.T)
+
+
+
+
+

And finally we plot our fit with and compare with data

+
+
+
Masses['Eapprox']  = ytilde
+# Generate a plot comparing the experimental with the fitted values values.
+fig, ax = plt.subplots()
+ax.set_xlabel(r'$A = N + Z$')
+ax.set_ylabel(r'$E_\mathrm{bind}\,/\mathrm{MeV}$')
+ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,
+            label='Ame2016')
+ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',
+            label='Fit')
+ax.legend()
+save_fig("Masses2016OLS")
+plt.show()
+
+
+
+
+
+
+

20.56. Adding error analysis and training set up

+

We can easily test our fit by computing the \(R2\) score that we discussed in connection with the functionality of Scikit-Learn in the introductory slides. +Since we are not using Scikit-Learn here we can define our own \(R2\) function as

+
+
+
def R2(y_data, y_model):
+    return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)
+
+
+
+
+

and we would be using it as

+
+
+
print(R2(Energies,ytilde))
+
+
+
+
+

We can easily add our MSE score as

+
+
+
def MSE(y_data,y_model):
+    n = np.size(y_model)
+    return np.sum((y_data-y_model)**2)/n
+
+print(MSE(Energies,ytilde))
+
+
+
+
+

and finally the relative error as

+
+
+
def RelativeError(y_data,y_model):
+    return abs((y_data-y_model)/y_data)
+print(RelativeError(Energies, ytilde))
+
+
+
+
+
+
+

20.57. The \(\chi^2\) function

+

Normally, the response (dependent or outcome) variable \(y_i\) is the +outcome of a numerical experiment or another type of experiment and is +thus only an approximation to the true value. It is then always +accompanied by an error estimate, often limited to a statistical error +estimate given by the standard deviation discussed earlier. In the +discussion here we will treat \(y_i\) as our exact value for the +response variable.

+

Introducing the standard deviation \(\sigma_i\) for each measurement +\(y_i\), we define now the \(\chi^2\) function (omitting the \(1/n\) term) +as

+
+\[ +\chi^2(\boldsymbol{\beta})=\frac{1}{n}\sum_{i=0}^{n-1}\frac{\left(y_i-\tilde{y}_i\right)^2}{\sigma_i^2}=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)^T\frac{1}{\boldsymbol{\Sigma^2}}\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)\right\}, +\]
+

where the matrix \(\boldsymbol{\Sigma}\) is a diagonal matrix with \(\sigma_i\) as matrix elements.

+
+
+

20.58. The \(\chi^2\) function

+

In order to find the parameters \(\beta_i\) we will then minimize the spread of \(\chi^2(\boldsymbol{\beta})\) by requiring

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_j} = \frac{\partial }{\partial \beta_j}\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(\frac{y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}}{\sigma_i}\right)^2\right]=0, +\]
+

which results in

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_j} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}\frac{x_{ij}}{\sigma_i}\left(\frac{y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}}{\sigma_i}\right)\right]=0, +\]
+

or in a matrix-vector form as

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{A}^T\left( \boldsymbol{b}-\boldsymbol{A}\boldsymbol{\beta}\right). +\]
+

where we have defined the matrix \(\boldsymbol{A} =\boldsymbol{X}/\boldsymbol{\Sigma}\) with matrix elements \(a_{ij} = x_{ij}/\sigma_i\) and the vector \(\boldsymbol{b}\) with elements \(b_i = y_i/\sigma_i\).

+
+
+

20.59. The \(\chi^2\) function

+

We can rewrite

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{A}^T\left( \boldsymbol{b}-\boldsymbol{A}\boldsymbol{\beta}\right), +\]
+

as

+
+\[ +\boldsymbol{A}^T\boldsymbol{b} = \boldsymbol{A}^T\boldsymbol{A}\boldsymbol{\beta}, +\]
+

and if the matrix \(\boldsymbol{A}^T\boldsymbol{A}\) is invertible we have the solution

+
+\[ +\boldsymbol{\beta} =\left(\boldsymbol{A}^T\boldsymbol{A}\right)^{-1}\boldsymbol{A}^T\boldsymbol{b}. +\]
+
+
+

20.60. The \(\chi^2\) function

+

If we then introduce the matrix

+
+\[ +\boldsymbol{H} = \left(\boldsymbol{A}^T\boldsymbol{A}\right)^{-1}, +\]
+

we have then the following expression for the parameters \(\beta_j\) (the matrix elements of \(\boldsymbol{H}\) are \(h_{ij}\))

+
+\[ +\beta_j = \sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}\frac{y_i}{\sigma_i}\frac{x_{ik}}{\sigma_i} = \sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}b_ia_{ik} +\]
+

We state without proof the expression for the uncertainty in the parameters \(\beta_j\) as (we leave this as an exercise)

+
+\[ +\sigma^2(\beta_j) = \sum_{i=0}^{n-1}\sigma_i^2\left( \frac{\partial \beta_j}{\partial y_i}\right)^2, +\]
+

resulting in

+
+\[ +\sigma^2(\beta_j) = \left(\sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}a_{ik}\right)\left(\sum_{l=0}^{p-1}h_{jl}\sum_{m=0}^{n-1}a_{ml}\right) = h_{jj}! +\]
+
+
+

20.61. The \(\chi^2\) function

+

The first step here is to approximate the function \(y\) with a first-order polynomial, that is we write

+
+\[ +y=y(x) \rightarrow y(x_i) \approx \beta_0+\beta_1 x_i. +\]
+

By computing the derivatives of \(\chi^2\) with respect to \(\beta_0\) and \(\beta_1\) show that these are given by

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_0} = -2\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(\frac{y_i-\beta_0-\beta_1x_{i}}{\sigma_i^2}\right)\right]=0, +\]
+

and

+
+\[ +\frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_1} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}x_i\left(\frac{y_i-\beta_0-\beta_1x_{i}}{\sigma_i^2}\right)\right]=0. +\]
+
+
+

20.62. The \(\chi^2\) function

+

For a linear fit (a first-order polynomial) we don’t need to invert a matrix!!
+Defining

+
+\[ +\gamma = \sum_{i=0}^{n-1}\frac{1}{\sigma_i^2}, +\]
+
+\[ +\gamma_x = \sum_{i=0}^{n-1}\frac{x_{i}}{\sigma_i^2}, +\]
+
+\[ +\gamma_y = \sum_{i=0}^{n-1}\left(\frac{y_i}{\sigma_i^2}\right), +\]
+
+\[ +\gamma_{xx} = \sum_{i=0}^{n-1}\frac{x_ix_{i}}{\sigma_i^2}, +\]
+
+\[ +\gamma_{xy} = \sum_{i=0}^{n-1}\frac{y_ix_{i}}{\sigma_i^2}, +\]
+

we obtain

+
+\[ +\beta_0 = \frac{\gamma_{xx}\gamma_y-\gamma_x\gamma_y}{\gamma\gamma_{xx}-\gamma_x^2}, +\]
+
+\[ +\beta_1 = \frac{\gamma_{xy}\gamma-\gamma_x\gamma_y}{\gamma\gamma_{xx}-\gamma_x^2}. +\]
+

This approach (different linear and non-linear regression) suffers +often from both being underdetermined and overdetermined in the +unknown coefficients \(\beta_i\). A better approach is to use the +Singular Value Decomposition (SVD) method discussed next week.

+
+
+

20.63. Fitting an Equation of State for Dense Nuclear Matter

+

Before we continue, let us introduce yet another example. We are going to fit the +nuclear equation of state using results from many-body calculations. +The equation of state we have made available here, as function of +density, has been derived using modern nucleon-nucleon potentials with +the addition of three-body +forces. This +time the file is presented as a standard csv file.

+

The beginning of the Python code here is similar to what you have seen +before, with the same initializations and declarations. We use also +pandas again, rather extensively in order to organize our data.

+

The difference now is that we use Scikit-Learn’s regression tools +instead of our own matrix inversion implementation. Furthermore, we +sneak in Ridge regression (to be discussed below) which includes a +hyperparameter \(\lambda\), also to be explained below.

+
+
+

20.64. The code

+
+
+
# Common imports
+import os
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import matplotlib.pyplot as plt
+import sklearn.linear_model as skl
+from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error
+
+# Where to save the figures and data files
+PROJECT_ROOT_DIR = "Results"
+FIGURE_ID = "Results/FigureFiles"
+DATA_ID = "DataFiles/"
+
+if not os.path.exists(PROJECT_ROOT_DIR):
+    os.mkdir(PROJECT_ROOT_DIR)
+
+if not os.path.exists(FIGURE_ID):
+    os.makedirs(FIGURE_ID)
+
+if not os.path.exists(DATA_ID):
+    os.makedirs(DATA_ID)
+
+def image_path(fig_id):
+    return os.path.join(FIGURE_ID, fig_id)
+
+def data_path(dat_id):
+    return os.path.join(DATA_ID, dat_id)
+
+def save_fig(fig_id):
+    plt.savefig(image_path(fig_id) + ".png", format='png')
+
+infile = open(data_path("EoS.csv"),'r')
+
+# Read the EoS data as  csv file and organize the data into two arrays with density and energies
+EoS = pd.read_csv(infile, names=('Density', 'Energy'))
+EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')
+EoS = EoS.dropna()
+Energies = EoS['Energy']
+Density = EoS['Density']
+#  The design matrix now as function of various polytrops
+X = np.zeros((len(Density),4))
+X[:,3] = Density**(4.0/3.0)
+X[:,2] = Density
+X[:,1] = Density**(2.0/3.0)
+X[:,0] = 1
+
+# We use now Scikit-Learn's linear regressor and ridge regressor
+# OLS part
+clf = skl.LinearRegression().fit(X, Energies)
+ytilde = clf.predict(X)
+EoS['Eols']  = ytilde
+# The mean squared error                               
+print("Mean squared error: %.2f" % mean_squared_error(Energies, ytilde))
+# Explained variance score: 1 is perfect prediction                                 
+print('Variance score: %.2f' % r2_score(Energies, ytilde))
+# Mean absolute error                                                           
+print('Mean absolute error: %.2f' % mean_absolute_error(Energies, ytilde))
+print(clf.coef_, clf.intercept_)
+
+# The Ridge regression with a hyperparameter lambda = 0.1
+_lambda = 0.1
+clf_ridge = skl.Ridge(alpha=_lambda).fit(X, Energies)
+yridge = clf_ridge.predict(X)
+EoS['Eridge']  = yridge
+# The mean squared error                               
+print("Mean squared error: %.2f" % mean_squared_error(Energies, yridge))
+# Explained variance score: 1 is perfect prediction                                 
+print('Variance score: %.2f' % r2_score(Energies, yridge))
+# Mean absolute error                                                           
+print('Mean absolute error: %.2f' % mean_absolute_error(Energies, yridge))
+print(clf_ridge.coef_, clf_ridge.intercept_)
+
+fig, ax = plt.subplots()
+ax.set_xlabel(r'$\rho[\mathrm{fm}^{-3}]$')
+ax.set_ylabel(r'Energy per particle')
+ax.plot(EoS['Density'], EoS['Energy'], alpha=0.7, lw=2,
+            label='Theoretical data')
+ax.plot(EoS['Density'], EoS['Eols'], alpha=0.7, lw=2, c='m',
+            label='OLS')
+ax.plot(EoS['Density'], EoS['Eridge'], alpha=0.7, lw=2, c='g',
+            label='Ridge $\lambda = 0.1$')
+ax.legend()
+save_fig("EoSfitting")
+plt.show()
+
+
+
+
+

The above simple polynomial in density \(\rho\) gives an excellent fit +to the data.

+

We note also that there is a small deviation between the +standard OLS and the Ridge regression at higher densities. We discuss this in more detail +below.

+
+
+

20.65. Splitting our Data in Training and Test data

+

It is normal in essentially all Machine Learning studies to split the +data in a training set and a test set (sometimes also an additional +validation set). Scikit-Learn has an own function for this. There +is no explicit recipe for how much data should be included as training +data and say test data. An accepted rule of thumb is to use +approximately \(2/3\) to \(4/5\) of the data as training data. We will +postpone a discussion of this splitting to the end of these notes and +our discussion of the so-called bias-variance tradeoff. Here we +limit ourselves to repeat the above equation of state fitting example +but now splitting the data into a training set and a test set.

+
+
+
import os
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from sklearn.model_selection import train_test_split
+# Where to save the figures and data files
+PROJECT_ROOT_DIR = "Results"
+FIGURE_ID = "Results/FigureFiles"
+DATA_ID = "DataFiles/"
+
+if not os.path.exists(PROJECT_ROOT_DIR):
+    os.mkdir(PROJECT_ROOT_DIR)
+
+if not os.path.exists(FIGURE_ID):
+    os.makedirs(FIGURE_ID)
+
+if not os.path.exists(DATA_ID):
+    os.makedirs(DATA_ID)
+
+def image_path(fig_id):
+    return os.path.join(FIGURE_ID, fig_id)
+
+def data_path(dat_id):
+    return os.path.join(DATA_ID, dat_id)
+
+def save_fig(fig_id):
+    plt.savefig(image_path(fig_id) + ".png", format='png')
+
+def R2(y_data, y_model):
+    return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)
+def MSE(y_data,y_model):
+    n = np.size(y_model)
+    return np.sum((y_data-y_model)**2)/n
+
+infile = open(data_path("EoS.csv"),'r')
+
+# Read the EoS data as  csv file and organized into two arrays with density and energies
+EoS = pd.read_csv(infile, names=('Density', 'Energy'))
+EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')
+EoS = EoS.dropna()
+Energies = EoS['Energy']
+Density = EoS['Density']
+#  The design matrix now as function of various polytrops
+X = np.zeros((len(Density),5))
+X[:,0] = 1
+X[:,1] = Density**(2.0/3.0)
+X[:,2] = Density
+X[:,3] = Density**(4.0/3.0)
+X[:,4] = Density**(5.0/3.0)
+# We split the data in test and training data
+X_train, X_test, y_train, y_test = train_test_split(X, Energies, test_size=0.2)
+# matrix inversion to find beta
+beta = np.linalg.inv(X_train.T.dot(X_train)).dot(X_train.T).dot(y_train)
+# and then make the prediction
+ytilde = X_train @ beta
+print("Training R2")
+print(R2(y_train,ytilde))
+print("Training MSE")
+print(MSE(y_train,ytilde))
+ypredict = X_test @ beta
+print("Test R2")
+print(R2(y_test,ypredict))
+print("Test MSE")
+print(MSE(y_test,ypredict))
+
+
+
+
+
+
+

20.66. Exercises

+

Here are three possible exercises for week 34

+
+
+

20.67. Exercise 1: Setting up various Python environments

+

The first exercise here is of a mere technical art. We want you to have

+
    +

  • git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo GitHub facilities.

    • +

  • Install various Python packages

    • +
+

We will make extensive use of Python as programming language and its +myriad of available libraries. You will find +IPython/Jupyter notebooks invaluable in your work. You can run R +codes in the Jupyter/IPython notebooks, with the immediate benefit of +visualizing your data. You can also use compiled languages like C++, +Rust, Fortran etc if you prefer. The focus in these lectures will be +on Python.

+

If you have Python installed (we recommend Python3) and you feel +pretty familiar with installing different packages, we recommend that +you install the following Python packages via pip as

+
    +

  1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow

    2. +
+

For Tensorflow, we recommend following the instructions in the text of +Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O’Reilly

+

We will come back to tensorflow later.

+

For Python3, replace pip with pip3.

+

For OSX users we recommend, after having installed Xcode, to +install brew. Brew allows for a seamless installation of additional +software via for example

+
    +

  1. brew install python3

    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +you can use pip as well and simply install Python as

+
    +

  1. sudo apt-get install python3 (or python for Python2.7)

    2. +
+

If you don’t want to perform these operations separately and venture +into the hassle of exploring how to set up dependencies and paths, we +recommend two widely used distrubutions which set up all relevant +dependencies for Python, namely

+ +

which is an open source +distribution of the Python and R programming languages for large-scale +data processing, predictive analytics, and scientific computing, that +aims to simplify package management and deployment. Package versions +are managed by the package management system conda.

+ +

is a Python +distribution for scientific and analytic computing distribution and +analysis environment, available for free and under a commercial +license.

+

We recommend using Anaconda if you are not too familiar with setting paths in a terminal environment.

+
+
+

20.68. Exercise 2: making your own data and exploring scikit-learn

+

We will generate our own dataset for a function \(y(x)\) where \(x \in [0,1]\) and defined by random numbers computed with the uniform distribution. The function \(y\) is a quadratic polynomial in \(x\) with added stochastic noise according to the normal distribution \(\cal {N}(0,1)\). +The following simple Python instructions define our \(x\) and \(y\) values (with 100 data points).

+
+
+
x = np.random.rand(100,1)
+y = 2.0+5*x*x+0.1*np.random.randn(100,1)
+
+
+
+
+
    +

  1. Write your own code (following the examples under the regression notes) for computing the parametrization of the data set fitting a second-order polynomial.

    2. +

  2. Use thereafter scikit-learn (see again the examples in the regression slides) and compare with your own code.

    3. +

  3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as

    4. +
+
+\[ +MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +\]
+

and the \(R^2\) score function. +If \(\tilde{\boldsymbol{y}}_i\) is the predicted value of the \(i-th\) sample and \(y_i\) is the corresponding true value, then the score \(R^2\) is defined as

+
+\[ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +\]
+

where we have defined the mean value of \(\boldsymbol{y}\) as

+
+\[ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +\]
+

You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. +Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits.

+
+
+

20.69. Exercise 3: Split data in test and training data

+

In this exercise we want you to to compute the MSE for the training +data and the test data as function of the complexity of a polynomial, +that is the degree of a given polynomial.

+

The aim is to reproduce Figure 2.11 of Hastie et al.

+

Our data is defined by \(x\in [-3,3]\) with a total of for example \(n=100\) data points. You should try to vary the number of data points \(n\) in your analysis.

+
+
+
np.random.seed()
+n = 100
+# Make data set.
+x = np.linspace(-3, 3, n).reshape(-1, 1)
+y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)
+
+
+
+
+

where \(y\) is the function we want to fit with a given polynomial.

+

a) +Write a first code which sets up a design matrix \(X\) defined by a fifth-order polynomial and split your data set in training and test data.

+

b) +Write thereafter (using either scikit-learn or your matrix inversion code using for example numpy) +and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial.

+

c) +Add now a model which allows you to make polynomials up to degree \(15\). Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?

+
+
+ + + + +
+ + + +
+ + +
+

previous

+

19. Exercises week 34

+
+ +
+ +
+
+
+

+ + By Morten Hjorth-Jensen
+ + © Copyright 2021.
+

+
+
+ + +
+
+ + + + + \ No newline at end of file diff --git a/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.ipynb b/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.ipynb new file mode 100644 index 000000000..53b974273 --- /dev/null +++ b/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.ipynb @@ -0,0 +1,327 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "241c4a61", + "metadata": { + "editable": true + }, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "a33815b9", + "metadata": { + "editable": true + }, + "source": [ + "# Exercises week 34\n", + "**FYS-STK3155/4155**\n", + "\n", + "Date: **August 21-25, 2023**" + ] + }, + { + "cell_type": "markdown", + "id": "2990585c", + "metadata": { + "editable": true + }, + "source": [ + "## Exercises\n", + "\n", + "Here are three possible exercises for week 34" + ] + }, + { + "cell_type": "markdown", + "id": "be056de0", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 1: Setting up various Python environments\n", + "\n", + "The first exercise here is of a mere technical art. We want you to have \n", + "* git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). \n", + "\n", + "* Install various Python packages\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "IPython/Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow \n", + "\n", + "For **Tensorflow**, we recommend following the instructions in the text of \n", + "[Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do)\n", + "\n", + "We will come back to **tensorflow** later. \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for Python2.7)\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment." + ] + }, + { + "cell_type": "markdown", + "id": "f0f4ffae", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 2: making your own data and exploring scikit-learn\n", + "\n", + "We will generate our own dataset for a function $y(x)$ where $x \\in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\\cal {N}(0,1)$.\n", + "The following simple Python instructions define our $x$ and $y$ values (with 100 data points)." + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "93d6a2b2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "ename": "NameError", + "evalue": "name 'np' is not defined", + "output_type": "error", + "traceback": [ + "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", + "\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)", + "Input \u001b[0;32mIn [1]\u001b[0m, in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0m x \u001b[38;5;241m=\u001b[39m \u001b[43mnp\u001b[49m\u001b[38;5;241m.\u001b[39mrandom\u001b[38;5;241m.\u001b[39mrand(\u001b[38;5;241m100\u001b[39m,\u001b[38;5;241m1\u001b[39m)\n\u001b[1;32m 2\u001b[0m y \u001b[38;5;241m=\u001b[39m \u001b[38;5;241m2.0\u001b[39m\u001b[38;5;241m+\u001b[39m\u001b[38;5;241m5\u001b[39m\u001b[38;5;241m*\u001b[39mx\u001b[38;5;241m*\u001b[39mx\u001b[38;5;241m+\u001b[39m\u001b[38;5;241m0.1\u001b[39m\u001b[38;5;241m*\u001b[39mnp\u001b[38;5;241m.\u001b[39mrandom\u001b[38;5;241m.\u001b[39mrandn(\u001b[38;5;241m100\u001b[39m,\u001b[38;5;241m1\u001b[39m)\n", + "\u001b[0;31mNameError\u001b[0m: name 'np' is not defined" + ] + } + ], + "source": [ + "x = np.random.rand(100,1)\n", + "y = 2.0+5*x*x+0.1*np.random.randn(100,1)" + ] + }, + { + "cell_type": "markdown", + "id": "45392145", + "metadata": { + "editable": true + }, + "source": [ + "1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. \n", + "\n", + "2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. \n", + "\n", + "3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as" + ] + }, + { + "cell_type": "markdown", + "id": "3eed315b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "26038071", + "metadata": { + "editable": true + }, + "source": [ + "and the $R^2$ score function.\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "4c750d55", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "30b4731e", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "49afe51d", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "81f16b80", + "metadata": { + "editable": true + }, + "source": [ + "You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. \n", + "Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits." + ] + }, + { + "cell_type": "markdown", + "id": "cb71cb1c", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 3: Split data in test and training data\n", + "\n", + "In this exercise we want you to to compute the MSE for the training\n", + "data and the test data as function of the complexity of a polynomial,\n", + "that is the degree of a given polynomial.\n", + "\n", + "The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf).\n", + "\n", + "Our data is defined by $x\\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis." + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "b8c919e5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "np.random.seed()\n", + "n = 100\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)" + ] + }, + { + "cell_type": "markdown", + "id": "50629e14", + "metadata": { + "editable": true + }, + "source": [ + "where $y$ is the function we want to fit with a given polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "0578f08c", + "metadata": { + "editable": true + }, + "source": [ + "**a)**\n", + "Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data." + ] + }, + { + "cell_type": "markdown", + "id": "125ec9a1", + "metadata": { + "editable": true + }, + "source": [ + "**b)**\n", + "Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**)\n", + "and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "e5bb2036", + "metadata": { + "editable": true + }, + "source": [ + "**c)**\n", + "Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?" + ] + } + ], + "metadata": { + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.10" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} \ No newline at end of file diff --git a/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.py b/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.py new file mode 100644 index 000000000..cf8d3b297 --- /dev/null +++ b/doc/LectureNotes/_build/jupyter_execute/exercisesweek34.py @@ -0,0 +1,147 @@ +#!/usr/bin/env python +# coding: utf-8 + +# +# + +# # Exercises week 34 +# **FYS-STK3155/4155** +# +# Date: **August 21-25, 2023** + +# ## Exercises +# +# Here are three possible exercises for week 34 + +# ## Exercise 1: Setting up various Python environments +# +# The first exercise here is of a mere technical art. We want you to have +# * git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). +# +# * Install various Python packages +# +# We will make extensive use of Python as programming language and its +# myriad of available libraries. You will find +# IPython/Jupyter notebooks invaluable in your work. You can run **R** +# codes in the Jupyter/IPython notebooks, with the immediate benefit of +# visualizing your data. You can also use compiled languages like C++, +# Rust, Fortran etc if you prefer. The focus in these lectures will be +# on Python. +# +# If you have Python installed (we recommend Python3) and you feel +# pretty familiar with installing different packages, we recommend that +# you install the following Python packages via **pip** as +# +# 1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow +# +# For **Tensorflow**, we recommend following the instructions in the text of +# [Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do) +# +# We will come back to **tensorflow** later. +# +# For Python3, replace **pip** with **pip3**. +# +# For OSX users we recommend, after having installed Xcode, to +# install **brew**. Brew allows for a seamless installation of additional +# software via for example +# +# 1. brew install python3 +# +# For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +# you can use **pip** as well and simply install Python as +# +# 1. sudo apt-get install python3 (or python for Python2.7) +# +# If you don't want to perform these operations separately and venture +# into the hassle of exploring how to set up dependencies and paths, we +# recommend two widely used distrubutions which set up all relevant +# dependencies for Python, namely +# +# * [Anaconda](https://docs.anaconda.com/), +# +# which is an open source +# distribution of the Python and R programming languages for large-scale +# data processing, predictive analytics, and scientific computing, that +# aims to simplify package management and deployment. Package versions +# are managed by the package management system **conda**. +# +# * [Enthought canopy](https://www.enthought.com/product/canopy/) +# +# is a Python +# distribution for scientific and analytic computing distribution and +# analysis environment, available for free and under a commercial +# license. +# +# We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment. + +# ## Exercise 2: making your own data and exploring scikit-learn +# +# We will generate our own dataset for a function $y(x)$ where $x \in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\cal {N}(0,1)$. +# The following simple Python instructions define our $x$ and $y$ values (with 100 data points). + +# In[1]: + + +x = np.random.rand(100,1) +y = 2.0+5*x*x+0.1*np.random.randn(100,1) + + +# 1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. +# +# 2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. +# +# 3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as + +# $$ +# MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +# \sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +# $$ + +# and the $R^2$ score function. +# If $\tilde{\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as + +# $$ +# R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +# $$ + +# where we have defined the mean value of $\boldsymbol{y}$ as + +# $$ +# \bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +# $$ + +# You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. +# Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits. + +# ## Exercise 3: Split data in test and training data +# +# In this exercise we want you to to compute the MSE for the training +# data and the test data as function of the complexity of a polynomial, +# that is the degree of a given polynomial. +# +# The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf). +# +# Our data is defined by $x\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis. + +# In[2]: + + +np.random.seed() +n = 100 +# Make data set. +x = np.linspace(-3, 3, n).reshape(-1, 1) +y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape) + + +# where $y$ is the function we want to fit with a given polynomial. + +# **a)** +# Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data. + +# **b)** +# Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**) +# and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial. + +# **c)** +# Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)? diff --git a/doc/LectureNotes/_build/jupyter_execute/week34.ipynb b/doc/LectureNotes/_build/jupyter_execute/week34.ipynb new file mode 100644 index 000000000..56aab343c --- /dev/null +++ b/doc/LectureNotes/_build/jupyter_execute/week34.ipynb @@ -0,0 +1,5975 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "390f1bd7", + "metadata": { + "editable": true + }, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "2c8cea05", + "metadata": { + "editable": true + }, + "source": [ + "# Week 34: Introduction to the course, Logistics and Practicalities\n", + "**Morten Hjorth-Jensen**, Department of Physics and Center for Computing in Science Education, University of Oslo, Norway and Department of Physics and Astronomy and Facility for Rare Isotope Beams, Michigan State University, USA\n", + "\n", + "Date: **Week 34, August 21-25, 2023**" + ] + }, + { + "cell_type": "markdown", + "id": "725240be", + "metadata": { + "editable": true + }, + "source": [ + "## Overview of first week\n", + "\n", + "1. The sessions on Tuesdays and Wednesdays last four hours for each group (four groups in total) and will include lectures in a flipped mode (promoting active learning) and work on exercices and projects.\n", + "\n", + "2. The sessions will begin with lectures, discussions, questions and answers about the material to be covered every week. Videos and teaching material will be announced in due time.\n", + "\n", + "3. There are four groups:\n", + "\n", + " * Tuesdays 815am-12pm and 1215pm-4pm\n", + "\n", + " * Wednesdays 815am-12pm and 1215pm-4pm.\n", + "\n", + "4. On Thursdays we have a regular lecture. These lectures start at 1215pm and end at 2pm and serve the aims of giving an overview over various topics. These lectures will also be recorded.\n", + "\n", + "The labs are also available till 6pm Tuesdays and Wednesdays. Videos and learning material with reading suggestions will be made available before each week starts." + ] + }, + { + "cell_type": "markdown", + "id": "c18f9a3c", + "metadata": { + "editable": true + }, + "source": [ + "## Schedule first week\n", + "\n", + " * August 22: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.\n", + "\n", + " * August 23: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group.\n", + "\n", + " * August 24: Lecture: Linear regression, examples and theory" + ] + }, + { + "cell_type": "markdown", + "id": "769cbfc6", + "metadata": { + "editable": true + }, + "source": [ + "## Lectures and ComputerLab\n", + "\n", + " * The sessions on Tuesdays and Wednesdays last four hours and will include partly lectures in a flipped mode (promoting active learning) and work on exercices and projects.\n", + "\n", + " * Thursdays: regular lectures (12.15pm-2pm) \n", + "\n", + " * Weekly reading assignments and videos needed to solve projects and exercises.\n", + "\n", + " * Weekly exercises. You can hand in exercises if you want and get an extra score, see below.\n", + "\n", + " * Detailed lecture notes, exercises, all programs presented, projects etc can be found at the homepage of the course.\n", + "\n", + " * Weekly plans and all other information are on the official website. This info will also be conveyed via weekly emails.\n", + "\n", + " * No final exam, three projects that are graded and have to be approved." + ] + }, + { + "cell_type": "markdown", + "id": "5040ec52", + "metadata": { + "editable": true + }, + "source": [ + "## Communication channels\n", + "\n", + "* Chat and communications via \n", + "\n", + "* **Discord** channel will be added asap" + ] + }, + { + "cell_type": "markdown", + "id": "867e5431", + "metadata": { + "editable": true + }, + "source": [ + "## Course Format\n", + "\n", + " * Three compulsory projects. Electronic reports only using [Canvas](https://www.uio.no/english/services/it/education/canvas/) to hand in projects and [git](https://git-scm.com/) as version control software and [GitHub](https://github.com/) for repository (or [GitLab](https://about.gitlab.com/)) of all your material.\n", + "\n", + " * Evaluation and grading: The three projects are graded and each counts 1/3 of the final mark. No final written or oral exam.\n", + "\n", + "a. For the last project each group/participant submits a proposal or works with suggested (by us) proposals for the project.\n", + "\n", + "b. If possible, we would like to organize the last project as a workshop where each group presents this to all other participants of the course\n", + "\n", + "c. Based on feedback etc, each group finalizes the report and submits for grading. \n", + "\n", + " * Python is the default programming language, but feel free to use C/C++, Julia and/or Fortran or other programming languages. All source codes discussed during the lectures can be found at the webpage and [github address](https://github.com/CompPhysics/MachineLearning/tree/master/doc/Programs) of the course." + ] + }, + { + "cell_type": "markdown", + "id": "61ff07a3", + "metadata": { + "editable": true + }, + "source": [ + "## Teachers\n", + "\n", + "* Morten Hjorth-Jensen, morten.hjorth-jensen@fys.uio.no\n", + "\n", + " * **Phone**: +47-48257387\n", + "\n", + " * **Office**: Department of Physics, University of Oslo, Eastern wing, room FØ470 \n", + "\n", + " * **Office hours**: *Anytime*! Individual or group office hours can be arranged either in person or via zoom. Feel free to send an email for planning. \n", + "\n", + "* Ida Torkjellsdatter Storehaug, i.t.storehaug@fys.uio.no\n", + "\n", + "* Fahimeh Najafi, fahimeh.najafi@fys.uio.no\n", + "\n", + "* Mia-Katrin Ose Kvalsund, m.k.o.kvalsund@fys.uio.no\n", + "\n", + "* Karl Henrik Fredly, k.h.fredly@fys.uio.no\n", + "\n", + "* Adam Jakobsen, adam.jakobsen@fys.uio.no\n", + "\n", + "* Daniel Haas Beccatini Lima, d.h.b.lima@fys.uio.no" + ] + }, + { + "cell_type": "markdown", + "id": "afa7d17c", + "metadata": { + "editable": true + }, + "source": [ + "## Deadlines for projects (tentative)\n", + "\n", + "1. Project 1: October 9 (available September 4) graded with feedback)\n", + "\n", + "2. Project 2: November 6 (available October 6, graded with feedback)\n", + "\n", + "3. Project 3: December 11 (available November 10, graded with feedback)\n", + "\n", + "Extra Credit (not mandatory), weekly exercise assignments, 10 in total (due Friday same week), 10% additional score. The extra credit assignments are due each Friday and can be uploaed to **Canvas** in your preferred format (although we prefer jupyter-notebooks). First assignment is for week 35. Each weekly exercise set counts 1%." + ] + }, + { + "cell_type": "markdown", + "id": "37a46b20", + "metadata": { + "editable": true + }, + "source": [ + "## Grading\n", + "\n", + "Grades are awarded on a scale from A to F, where A is the best grade and F is a fail. There are three projects which are graded and each project counts 1/3 of the final grade. The total score is thus the average from all three projects.\n", + "\n", + "The final number of points is based on the average of all projects and the grade follows the following table:\n", + "\n", + " * 92-100 points: A\n", + "\n", + " * 77-91 points: B\n", + "\n", + " * 58-76 points: C\n", + "\n", + " * 46-57 points: D\n", + "\n", + " * 40-45 points: E\n", + "\n", + " * 0-39 points: F-failed\n", + "\n", + "In addition you can get an extra 10% score for weekly assignments (10 in total and due each Friday). Each weekly assignment counts 1%." + ] + }, + { + "cell_type": "markdown", + "id": "40b50d78", + "metadata": { + "editable": true + }, + "source": [ + "## Reading material\n", + "\n", + "The lecture notes are collected as a jupyter-book at .\n", + "\n", + "In addition to the lecture notes, we recommend the books of Bishop, Hastie et al, Murphy and Goodfellow et al. We will follow these texts closely and the weekly reading assignments refer to these texts. The text by Hastie et al is also widely used in the Machine Learning community. Finally, we also recommend the hands-on text by Geron, see next slide for links." + ] + }, + { + "cell_type": "markdown", + "id": "1ac7f5fa", + "metadata": { + "editable": true + }, + "source": [ + "## Textbooks\n", + "\n", + "* [Goodfellow, Bengio, and Courville (GBC), Deep Learning](https://www.deeplearningbook.org/)\n", + "\n", + "* Christopher M. Bishop (CB), Pattern Recognition and Machine Learning\n", + "\n", + "* [Hastie, Tibshirani, and Friedman (HTF), The Elements of Statistical Learning, Springer](https://www.springer.com/gp/book/9780387848570).\n", + "\n", + "* [Aurelien Geron (AG), Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/). This text is very useful since it contains many code examples and hands-on applications of all algorithms discussed in this course.\n", + "\n", + "* [Kevin Murphy (KM), Probabilistic Machine Learning, an Introduction](https://probml.github.io/pml-book/book1.html)" + ] + }, + { + "cell_type": "markdown", + "id": "c74dc137", + "metadata": { + "editable": true + }, + "source": [ + "## Reading suggestions week 34\n", + "\n", + "This week: Refresh linear algebra, GBC chapters 1 and 2. HTF chapters 2 and 3. Install scikit-learn. See lecture notes for week 34 at (these notes)." + ] + }, + { + "cell_type": "markdown", + "id": "6af0d013", + "metadata": { + "editable": true + }, + "source": [ + "## Prerequisites\n", + "\n", + "Basic knowledge in programming and mathematics, with an emphasis on\n", + "linear algebra. Knowledge of Python or/and C++ as programming\n", + "languages is strongly recommended and experience with Jupiter notebook\n", + "is recommended. Required courses are the equivalents to the University\n", + "of Oslo mathematics courses MAT1100, MAT1110, MAT1120 and at least one\n", + "of the corresponding computing and programming courses INF1000/INF1110\n", + "or MAT-INF1100/MAT-INF1100L/BIOS1100/KJM-INF1100. Most universities\n", + "offer nowadays a basic programming course (often compulsory) where\n", + "Python is the recurring programming language." + ] + }, + { + "cell_type": "markdown", + "id": "628fb74c", + "metadata": { + "editable": true + }, + "source": [ + "## Topics covered in this course: Statistical analysis and optimization of data\n", + "\n", + "The course has two central parts\n", + "\n", + "1. Statistical analysis and optimization of data\n", + "\n", + "2. Machine learning\n", + "\n", + "These topics will be scattered thorughout the course and may not necessarily be taught separately. Rather, we will often take an approach (during the lectures and project/exercise sessions) where say elements from statistical data analysis are mixed with specific Machine Learning algorithms." + ] + }, + { + "cell_type": "markdown", + "id": "bcfe0b24", + "metadata": { + "editable": true + }, + "source": [ + "## Statistical analysis and optimization of data\n", + "\n", + "We plan to cover the following topics:\n", + "* Basic concepts, expectation values, variance, covariance, correlation functions and errors;\n", + "\n", + "* Simpler models, binomial distribution, the Poisson distribution, simple and multivariate normal distributions;\n", + "\n", + "* Central elements of Bayesian statistics and modeling;\n", + "\n", + "* Gradient methods for data optimization;\n", + "\n", + "* Monte Carlo methods, Markov chains, Gibbs sampling and Metropolis-Hastings sampling (tentative);\n", + "\n", + "* Estimation of errors and resampling techniques such as the cross-validation, blocking, bootstrapping and jackknife methods;\n", + "\n", + "* Principal Component Analysis (PCA) and its mathematical foundation;" + ] + }, + { + "cell_type": "markdown", + "id": "f153abef", + "metadata": { + "editable": true + }, + "source": [ + "## Machine Learning\n", + "\n", + "* Pre deep-learning revolution (2008 approx)\n", + "\n", + " * Linear Regression and Logistic Regression, classification and regression problems;\n", + "\n", + " * Bayesian linear and logistic regression, kernel regression;\n", + "\n", + " * Decisions trees, Random Forests, Bagging and Boosting methods;\n", + "\n", + " * Support vector machines (only survey);\n", + "\n", + " * Unsupervised learning and dimensionality reduction, from PCA to clustering; \n", + "\n", + "* Deep learning \n", + "\n", + " * Neural networks and deep learning;\n", + "\n", + " * Convolutional neural networks;\n", + "\n", + " * Recurrent neural networks;\n", + "\n", + " * Autoencoders\n", + "\n", + " * Generative methods with an emphasis on Boltzmann Machines, Variational Autoencoders and Generalized Adversarial Networks;\n", + "\n", + "Hands-on demonstrations, exercises and projects aim at deepening your understanding of these topics." + ] + }, + { + "cell_type": "markdown", + "id": "a6b0c4b8", + "metadata": { + "editable": true + }, + "source": [ + "## Extremely useful tools, strongly recommended\n", + "\n", + "**and discussed at the lab sessions.**\n", + "\n", + " * GIT for version control, and GitHub or GitLab as repositories, highly recommended. This will be discussed during the first exercise session\n", + "\n", + " * Anaconda and other Python environments, see intro slides and links to programming resources at " + ] + }, + { + "cell_type": "markdown", + "id": "48eb8817", + "metadata": { + "editable": true + }, + "source": [ + "## Other courses on Data science and Machine Learning at UiO\n", + "\n", + "* [FYS5419 Quantum Computing and Quantum Machine Learning](https://www.uio.no/studier/emner/matnat/fys/FYS5419/index-eng.html)\n", + "\n", + "* [FYS5429 Advanced Machine Learning for the Physical Sciences](https://www.uio.no/studier/emner/matnat/fys/FYS5429/index-eng.html)\n", + "\n", + "* [STK2100 Machine learning and statistical methods for prediction and classification](http://www.uio.no/studier/emner/matnat/math/STK2100/index-eng.html). \n", + "\n", + "* [IN3050/4050 Introduction to Artificial Intelligence and Machine Learning](https://www.uio.no/studier/emner/matnat/ifi/IN3050/index-eng.html). Introductory course in machine learning and AI with an algorithmic approach. \n", + "\n", + "* [STK-INF3000/4000 Selected Topics in Data Science](http://www.uio.no/studier/emner/matnat/math/STK-INF3000/index-eng.html). The course provides insight into selected contemporary relevant topics within Data Science. \n", + "\n", + "* [IN4080 Natural Language Processing](https://www.uio.no/studier/emner/matnat/ifi/IN4080/index.html). Probabilistic and machine learning techniques applied to natural language processing." + ] + }, + { + "cell_type": "markdown", + "id": "e8889321", + "metadata": { + "editable": true + }, + "source": [ + "## Other courses on Data science and Machine Learning at UiO, contn\n", + "\n", + "* [STK-IN4300 Statistical learning methods in Data Science](https://www.uio.no/studier/emner/matnat/math/STK-IN4300/index-eng.html). An advanced introduction to statistical and machine learning. For students with a good mathematics and statistics background.\n", + "\n", + "* [IN3310/4310 Deep Learnig for Image Analysis](https://www.uio.no/studier/emner/matnat/ifi/IN4310/index.html)\n", + "\n", + "* [STK4051 Computational Statistics](https://www.uio.no/studier/emner/matnat/math/STK4051/index-eng.html)\n", + "\n", + "* [STK4021 Applied Bayesian Analysis and Numerical Methods](https://www.uio.no/studier/emner/matnat/math/STK4021/index-eng.html)" + ] + }, + { + "cell_type": "markdown", + "id": "4ad1f253", + "metadata": { + "editable": true + }, + "source": [ + "## Learning outcomes\n", + "\n", + "This course aims at giving you insights and knowledge about many of\n", + "the central algorithms used in Data Analysis and Machine Learning.\n", + "The course is project based and through various numerical projects,\n", + "normally three, you will be exposed to fundamental research problems\n", + "in these fields, with the aim to reproduce state of the art scientific\n", + "results. Both supervised and unsupervised methods will be covered. The\n", + "emphasis is on a frequentist approach, although we will try to link it\n", + "with a Bayesian approach as well. You will learn to develop and\n", + "structure large codes for studying different cases where Machine\n", + "Learning is applied to, get acquainted with computing facilities and\n", + "learn to handle large scientific projects. A good scientific and\n", + "ethical conduct is emphasized throughout the course. More\n", + "specifically, after this course you will\n", + "\n", + "* Learn about basic data analysis, statistical analysis, Bayesian statistics, Monte Carlo sampling, data optimization and machine learning;\n", + "\n", + "* Be capable of extending the acquired knowledge to other systems and cases;\n", + "\n", + "* Have an understanding of central algorithms used in data analysis and machine learning;\n", + "\n", + "* Understand linear methods for regression and classification, from ordinary least squares, via Lasso and Ridge to Logistic regression;\n", + "\n", + "* Learn about neural networks and deep learning methods for supervised and unsupervised learning. Emphasis on feed forward neural networks, convolutional and recurrent neural networks; \n", + "\n", + "* Learn about about decision trees, random forests, bagging and boosting methods;\n", + "\n", + "* Learn about support vector machines and kernel transformations;\n", + "\n", + "* Reduction of data sets, from PCA to clustering;\n", + "\n", + "* Generative models\n", + "\n", + "* Work on numerical projects to illustrate the theory. The projects play a central role and you are expected to know modern programming languages like Python or C++ and/or Fortran (Fortran2003 or later) or Julia or other." + ] + }, + { + "cell_type": "markdown", + "id": "ed3eddab", + "metadata": { + "editable": true + }, + "source": [ + "## Introduction\n", + "\n", + "Our emphasis throughout this series of lectures \n", + "is on understanding the mathematical aspects of\n", + "different algorithms used in the fields of data analysis and machine learning. \n", + "\n", + "However, where possible we will emphasize the\n", + "importance of using available software. We start thus with a hands-on\n", + "and top-down approach to machine learning. The aim is thus to start with\n", + "relevant data or data we have produced \n", + "and use these to introduce statistical data analysis\n", + "concepts and machine learning algorithms before we delve into the\n", + "algorithms themselves. The examples we will use in the beginning, start with simple\n", + "polynomials with random noise added. We will use the Python\n", + "software package [Scikit-Learn](http://scikit-learn.org/stable/) and\n", + "introduce various machine learning algorithms to make fits of\n", + "the data and predictions. We move thereafter to more interesting\n", + "cases such as data from say experiments (below we will look at experimental nuclear binding energies as an example).\n", + "These are examples where we can easily set up the data and\n", + "then use machine learning algorithms included in for example\n", + "**Scikit-Learn**. \n", + "\n", + "These examples will serve us the purpose of getting\n", + "started. Furthermore, they allow us to catch more than two birds with\n", + "a stone. They will allow us to bring in some programming specific\n", + "topics and tools as well as showing the power of various Python \n", + "libraries for machine learning and statistical data analysis. \n", + "\n", + "Although we have projects where you write your own codes, we will also focus on two\n", + "specific Python packages for Machine Learning, Scikit-Learn and\n", + "Tensorflow with Keras (see below for links etc). Moreover, the examples we\n", + "introduce will serve as inputs to many of our discussions later, as\n", + "well as allowing you to set up models and produce your own data and\n", + "get started with programming." + ] + }, + { + "cell_type": "markdown", + "id": "cd6f4f41", + "metadata": { + "editable": true + }, + "source": [ + "## AI/ML and some statements you may have heard (and what do they mean?)\n", + "\n", + "1. Fei-Fei Li on ImageNet: **map out the entire world of objects** ([The data that transformed AI research](https://cacm.acm.org/news/219702-the-data-that-transformed-ai-research-and-possibly-the-world/fulltext))\n", + "\n", + "2. Russell and Norvig in their popular textbook: **relevant to any intellectual task; it is truly a universal field** ([Artificial Intelligence, A modern approach](http://aima.cs.berkeley.edu/))\n", + "\n", + "3. Woody Bledsoe puts it more bluntly: **in the long run, AI is the only science** (quoted in Pamilla McCorduck, [Machines who think](https://www.pamelamccorduck.com/machines-who-think))\n", + "\n", + "If you wish to have a critical read on AI/ML from a societal point of view, see [Kate Crawford's recent text Atlas of AI](https://www.katecrawford.net/)\n", + "\n", + "**Here: with AI/ML we intend a collection of machine learning methods with an emphasis on statistical learning and data analysis**" + ] + }, + { + "cell_type": "markdown", + "id": "9bc6760a", + "metadata": { + "editable": true + }, + "source": [ + "## What is Machine Learning?\n", + "\n", + "Statistics, data science and machine learning form important fields of\n", + "research in modern science. They describe how to learn and make\n", + "predictions from data, as well as allowing us to extract important\n", + "correlations about physical process and the underlying laws of motion\n", + "in large data sets. The latter, big data sets, appear frequently in\n", + "essentially all disciplines, from the traditional Science, Technology,\n", + "Mathematics and Engineering fields to Life Science, Law, education\n", + "research, the Humanities and the Social Sciences. \n", + "\n", + "It has become more\n", + "and more common to see research projects on big data in for example\n", + "the Social Sciences where extracting patterns from complicated survey\n", + "data is one of many research directions. Having a solid grasp of data\n", + "analysis and machine learning is thus becoming central to scientific\n", + "computing in many fields, and competences and skills within the fields\n", + "of machine learning and scientific computing are nowadays strongly\n", + "requested by many potential employers. The latter cannot be\n", + "overstated, familiarity with machine learning has almost become a\n", + "prerequisite for many of the most exciting employment opportunities,\n", + "whether they are in bioinformatics, life science, physics or finance,\n", + "in the private or the public sector. This author has had several\n", + "students or met students who have been hired recently based on their\n", + "skills and competences in scientific computing and data science, often\n", + "with marginal knowledge of machine learning.\n", + "\n", + "Machine learning is a subfield of computer science, and is closely\n", + "related to computational statistics. It evolved from the study of\n", + "pattern recognition in artificial intelligence (AI) research, and has\n", + "made contributions to AI tasks like computer vision, natural language\n", + "processing and speech recognition. Many of the methods we will study are also \n", + "strongly rooted in basic mathematics and physics research. \n", + "\n", + "Ideally, machine learning represents the science of giving computers\n", + "the ability to learn without being explicitly programmed. The idea is\n", + "that there exist generic algorithms which can be used to find patterns\n", + "in a broad class of data sets without having to write code\n", + "specifically for each problem. The algorithm will build its own logic\n", + "based on the data. You should however always keep in mind that\n", + "machines and algorithms are to a large extent developed by humans. The\n", + "insights and knowledge we have about a specific system, play a central\n", + "role when we develop a specific machine learning algorithm. \n", + "\n", + "Machine learning is an extremely rich field, in spite of its young\n", + "age. The increases we have seen during the last three decades in\n", + "computational capabilities have been followed by developments of\n", + "methods and techniques for analyzing and handling large date sets,\n", + "relying heavily on statistics, computer science and mathematics. The\n", + "field is rather new and developing rapidly. Popular software packages\n", + "written in Python for machine learning like\n", + "[Scikit-learn](http://scikit-learn.org/stable/),\n", + "[Tensorflow](https://www.tensorflow.org/),\n", + "[PyTorch](http://pytorch.org/) and [Keras](https://keras.io/), all\n", + "freely available at their respective GitHub sites, encompass\n", + "communities of developers in the thousands or more. And the number of\n", + "code developers and contributors keeps increasing. Not all the\n", + "algorithms and methods can be given a rigorous mathematical\n", + "justification, opening up thereby large rooms for experimenting and\n", + "trial and error and thereby exciting new developments. However, a\n", + "solid command of linear algebra, multivariate theory, probability\n", + "theory, statistical data analysis, understanding errors and Monte\n", + "Carlo methods are central elements in a proper understanding of many\n", + "of algorithms and methods we will discuss." + ] + }, + { + "cell_type": "markdown", + "id": "17c6257a", + "metadata": { + "editable": true + }, + "source": [ + "## Types of Machine Learning\n", + "\n", + "The approaches to machine learning are many, but are often split into\n", + "two main categories. In *supervised learning* we know the answer to a\n", + "problem, and let the computer deduce the logic behind it. On the other\n", + "hand, *unsupervised learning* is a method for finding patterns and\n", + "relationship in data sets without any prior knowledge of the system.\n", + "Some authours also operate with a third category, namely\n", + "*reinforcement learning*. This is a paradigm of learning inspired by\n", + "behavioral psychology, where learning is achieved by trial-and-error,\n", + "solely from rewards and punishment.\n", + "\n", + "Another way to categorize machine learning tasks is to consider the\n", + "desired output of a system. Some of the most common tasks are:\n", + "\n", + " * Classification: Outputs are divided into two or more classes. The goal is to produce a model that assigns inputs into one of these classes. An example is to identify digits based on pictures of hand-written ones. Classification is typically supervised learning.\n", + "\n", + " * Regression: Finding a functional relationship between an input data set and a reference data set. The goal is to construct a function that maps input data to continuous output values.\n", + "\n", + " * Clustering: Data are divided into groups with certain common traits, without knowing the different groups beforehand. It is thus a form of unsupervised learning." + ] + }, + { + "cell_type": "markdown", + "id": "9f4e80cf", + "metadata": { + "editable": true + }, + "source": [ + "## Essential elements of ML\n", + "\n", + "The methods we cover have three main topics in common, irrespective of\n", + "whether we deal with supervised or unsupervised learning.\n", + "* The first ingredient is normally our data set (which can be subdivided into training, validation and test data). Many find the most difficult part of using Machine Learning to be the set up of your data in a meaningful way. \n", + "\n", + "* The second item is a model which is normally a function of some parameters. The model reflects our knowledge of the system (or lack thereof). As an example, if we know that our data show a behavior similar to what would be predicted by a polynomial, fitting our data to a polynomial of some degree would then determin our model. \n", + "\n", + "* The last ingredient is a so-called **cost/loss** function (or error or risk function) which allows us to present an estimate on how good our model is in reproducing the data it is supposed to train." + ] + }, + { + "cell_type": "markdown", + "id": "828f3ad9", + "metadata": { + "editable": true + }, + "source": [ + "## An optimization/minimization problem\n", + "\n", + "At the heart of basically all Machine Learning algorithms we will encounter so-called minimization or optimization algorithms. A large family of such methods are so-called **gradient methods**." + ] + }, + { + "cell_type": "markdown", + "id": "8b12df05", + "metadata": { + "editable": true + }, + "source": [ + "## A Frequentist approach to data analysis\n", + "\n", + "When you hear phrases like **predictions and estimations** and\n", + "**correlations and causations**, what do you think of? May be you think\n", + "of the difference between classifying new data points and generating\n", + "new data points.\n", + "Or perhaps you consider that correlations represent some kind of symmetric statements like\n", + "if $A$ is correlated with $B$, then $B$ is correlated with\n", + "$A$. Causation on the other hand is directional, that is if $A$ causes $B$, $B$ does not\n", + "necessarily cause $A$.\n", + "\n", + "These concepts are in some sense the difference between machine\n", + "learning and statistics. In machine learning and prediction based\n", + "tasks, we are often interested in developing algorithms that are\n", + "capable of learning patterns from given data in an automated fashion,\n", + "and then using these learned patterns to make predictions or\n", + "assessments of newly given data. In many cases, our primary concern\n", + "is the quality of the predictions or assessments, and we are less\n", + "concerned about the underlying patterns that were learned in order\n", + "to make these predictions.\n", + "\n", + "In machine learning we normally use [a so-called frequentist approach](https://en.wikipedia.org/wiki/Frequentist_inference),\n", + "where the aim is to make predictions and find correlations. We focus\n", + "less on for example extracting a probability distribution function (PDF). The PDF can be\n", + "used in turn to make estimations and find causations such as given $A$\n", + "what is the likelihood of finding $B$." + ] + }, + { + "cell_type": "markdown", + "id": "6656791e", + "metadata": { + "editable": true + }, + "source": [ + "## What is a good model?\n", + "\n", + "In science and engineering we often end up in situations where we want to infer (or learn) a\n", + "quantitative model $M$ for a given set of sample points $\\boldsymbol{X} \\in [x_1, x_2,\\dots x_N]$.\n", + "\n", + "As we will see repeatedely in these lectures, we could try to fit these data points to a model given by a\n", + "straight line, or if we wish to be more sophisticated to a more complex\n", + "function.\n", + "\n", + "The reason for inferring such a model is that it\n", + "serves many useful purposes. On the one hand, the model can reveal information\n", + "encoded in the data or underlying mechanisms from which the data were generated. For instance, we could discover important\n", + "corelations that relate interesting physics interpretations.\n", + "\n", + "In addition, it can simplify the representation of the given data set and help\n", + "us in making predictions about future data samples.\n", + "\n", + "A first important consideration to keep in mind is that inferring the *correct* model\n", + "for a given data set is an elusive, if not impossible, task. The fundamental difficulty\n", + "is that if we are not specific about what we mean by a *correct* model, there\n", + "could easily be many different models that fit the given data set *equally well*." + ] + }, + { + "cell_type": "markdown", + "id": "85f146c9", + "metadata": { + "editable": true + }, + "source": [ + "## What is a good model? Can we define it?\n", + "\n", + "The central question is this: what leads us to say that a model is correct or\n", + "optimal for a given data set? To make the model inference problem well posed, i.e.,\n", + "to guarantee that there is a unique optimal model for the given data, we need to\n", + "impose additional assumptions or restrictions on the class of models considered. To\n", + "this end, we should not be looking for just any model that can describe the data.\n", + "Instead, we should look for a **model** $M$ that is the best among a restricted class\n", + "of models. In addition, to make the model inference problem computationally\n", + "tractable, we need to specify how restricted the class of models needs to be. A\n", + "common strategy is to start \n", + "with the simplest possible class of models that is just necessary to describe the data\n", + "or solve the problem at hand. More precisely, the model class should be rich enough\n", + "to contain at least one model that can fit the data to a desired accuracy and yet be\n", + "restricted enough that it is relatively simple to find the best model for the given data.\n", + "\n", + "Thus, the most popular strategy is to start from the\n", + "simplest class of models and increase the complexity of the models only when the\n", + "simpler models become inadequate. For instance, if we work with a regression problem to fit a set of sample points, one\n", + "may first try the simplest class of models, namely linear models, followed obviously by more complex models.\n", + "\n", + "How to evaluate which model fits best the data is something we will come back to over and over again in these sets of lectures." + ] + }, + { + "cell_type": "markdown", + "id": "51f73840", + "metadata": { + "editable": true + }, + "source": [ + "## Software and needed installations\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Julia, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we strongly recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn mglearn sympy pandas pillow \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for pyhton2.7)\n", + "\n", + "etc etc." + ] + }, + { + "cell_type": "markdown", + "id": "404f8680", + "metadata": { + "editable": true + }, + "source": [ + "## Python installers\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "Furthermore, [Google's Colab](https://colab.research.google.com/notebooks/welcome.ipynb) is a free Jupyter notebook environment that requires \n", + "no setup and runs entirely in the cloud. Try it out!" + ] + }, + { + "cell_type": "markdown", + "id": "1b74192a", + "metadata": { + "editable": true + }, + "source": [ + "## Useful Python libraries\n", + "Here we list several useful Python libraries we strongly recommend (if you use anaconda many of these are already there)\n", + "\n", + "* [NumPy](https://www.numpy.org/) is a highly popular library for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays\n", + "\n", + "* [The pandas](https://pandas.pydata.org/) library provides high-performance, easy-to-use data structures and data analysis tools \n", + "\n", + "* [Xarray](http://xarray.pydata.org/en/stable/) is a Python package that makes working with labelled multi-dimensional arrays simple, efficient, and fun!\n", + "\n", + "* [Scipy](https://www.scipy.org/) (pronounced “Sigh Pie”) is a Python-based ecosystem of open-source software for mathematics, science, and engineering. \n", + "\n", + "* [Matplotlib](https://matplotlib.org/) is a Python 2D plotting library which produces publication quality figures in a variety of hardcopy formats and interactive environments across platforms.\n", + "\n", + "* [Autograd](https://github.com/HIPS/autograd) can automatically differentiate native Python and Numpy code. It can handle a large subset of Python's features, including loops, ifs, recursion and closures, and it can even take derivatives of derivatives of derivatives\n", + "\n", + "* [JAX](https://jax.readthedocs.io/en/latest/index.html) has now more or less replaced **Autograd**. JAX is Autograd and XLA, brought together for high-performance numerical computing and machine learning research. It provides composable transformations of Python+NumPy programs: differentiate, vectorize, parallelize, Just-In-Time compile to GPU/TPU, and more.\n", + "\n", + "* [SymPy](https://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. \n", + "\n", + "* [scikit-learn](https://scikit-learn.org/stable/) has simple and efficient tools for machine learning, data mining and data analysis\n", + "\n", + "* [TensorFlow](https://www.tensorflow.org/) is a Python library for fast numerical computing created and released by Google\n", + "\n", + "* [Keras](https://keras.io/) is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano\n", + "\n", + "* And many more such as [pytorch](https://pytorch.org/), [Theano](https://pypi.org/project/Theano/) etc" + ] + }, + { + "cell_type": "markdown", + "id": "32b6b09c", + "metadata": { + "editable": true + }, + "source": [ + "## Installing R, C++, cython or Julia\n", + "\n", + "You will also find it convenient to utilize **R**. We will mainly\n", + "use Python during our lectures and in various projects and exercises.\n", + "Those of you\n", + "already familiar with **R** should feel free to continue using **R**, keeping\n", + "however an eye on the parallel Python set ups. Similarly, if you are a\n", + "Python afecionado, feel free to explore **R** as well. Jupyter(Julia, Python and R) /Ipython\n", + "notebook allows you to run **R** codes and **Julia** codes interactively in your\n", + "browser. The software library **R** is really tailored for statistical data analysis\n", + "and allows for an easy usage of the tools and algorithms we will discuss in these\n", + "lectures.\n", + "\n", + "To install **R** with Jupyter notebook \n", + "[follow the link here](https://mpacer.org/maths/r-kernel-for-ipython-notebook)" + ] + }, + { + "cell_type": "markdown", + "id": "0c8344b1", + "metadata": { + "editable": true + }, + "source": [ + "## Installing R, C++, cython, Numba etc\n", + "\n", + "For the C++ aficionados, Jupyter/IPython notebook allows you also to\n", + "install C++ and run codes written in this language interactively in\n", + "the browser. Since we will emphasize writing many of the algorithms\n", + "yourself, you can thus opt for either Python or C++ (or Fortran or other compiled languages) as programming\n", + "languages.\n", + "\n", + "To add more entropy, **cython** can also be used when running your\n", + "notebooks. It means that Python with the jupyter notebook\n", + "setup allows you to integrate widely popular softwares and tools for\n", + "scientific computing. Similarly, the \n", + "[Numba Python package](https://numba.pydata.org/) delivers increased performance\n", + "capabilities with minimal rewrites of your codes. With its\n", + "versatility, including symbolic operations, Python offers a unique\n", + "computational environment. Your jupyter notebook can easily be\n", + "converted into a nicely rendered **PDF** file or a Latex file for\n", + "further processing. For example, convert to latex as" + ] + }, + { + "cell_type": "markdown", + "id": "40a41b04", + "metadata": { + "editable": true + }, + "source": [ + " pycod jupyter nbconvert filename.ipynb --to latex \n" + ] + }, + { + "cell_type": "markdown", + "id": "367b0b81", + "metadata": { + "editable": true + }, + "source": [ + "And to add more versatility, the Python package [SymPy](http://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. It aims to become a full-featured computer algebra system (CAS) and is entirely written in Python. \n", + "\n", + "Finally, we recommend strongly using Autograd or JAX for automatic differentiation." + ] + }, + { + "cell_type": "markdown", + "id": "e16a079a", + "metadata": { + "editable": true + }, + "source": [ + "## Numpy examples and Important Matrix and vector handling packages\n", + "\n", + "There are several central software libraries for linear algebra and eigenvalue problems. Several of the more\n", + "popular ones have been wrapped into ofter software packages like those from the widely used text **Numerical Recipes**. The original source codes in many of the available packages are often taken from the widely used\n", + "software package LAPACK, which follows two other popular packages\n", + "developed in the 1970s, namely EISPACK and LINPACK. We describe them shortly here.\n", + "\n", + " * LINPACK: package for linear equations and least square problems.\n", + "\n", + " * LAPACK:package for solving symmetric, unsymmetric and generalized eigenvalue problems. From LAPACK's website it is possible to download for free all source codes from this library. Both C/C++ and Fortran versions are available.\n", + "\n", + " * BLAS (I, II and III): (Basic Linear Algebra Subprograms) are routines that provide standard building blocks for performing basic vector and matrix operations. Blas I is vector operations, II vector-matrix operations and III matrix-matrix operations. Highly parallelized and efficient codes, all available for download from ." + ] + }, + { + "cell_type": "markdown", + "id": "244cc56a", + "metadata": { + "editable": true + }, + "source": [ + "## Numpy and arrays\n", + "[Numpy](http://www.numpy.org/) provides an easy way to handle arrays in Python. The standard way to import this library is as" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "1d8845bb", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import numpy as np" + ] + }, + { + "cell_type": "markdown", + "id": "a339aed7", + "metadata": { + "editable": true + }, + "source": [ + "Here follows a simple example where we set up an array of ten elements, all determined by random numbers drawn according to the normal distribution," + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "ce4b3b4e", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[ 0.39864527 -1.50219982 -0.24611674 -0.52819229 -0.24533566 -0.85694362\n", + " -0.81955926 0.14252712 -1.00625843 -0.23732849]\n" + ] + } + ], + "source": [ + "n = 10\n", + "x = np.random.normal(size=n)\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "307170c5", + "metadata": { + "editable": true + }, + "source": [ + "We defined a vector $x$ with $n=10$ elements with its values given by the Normal distribution $N(0,1)$.\n", + "Another alternative is to declare a vector as follows" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "ea4fd74f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1 2 3]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "x = np.array([1, 2, 3])\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "6428f55d", + "metadata": { + "editable": true + }, + "source": [ + "Here we have defined a vector with three elements, with $x_0=1$, $x_1=2$ and $x_2=3$. Note that both Python and C++\n", + "start numbering array elements from $0$ and on. This means that a vector with $n$ elements has a sequence of entities $x_0, x_1, x_2, \\dots, x_{n-1}$. We could also let (recommended) Numpy to compute the logarithms of a specific array as" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "84e4a8fc", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1.38629436 1.94591015 2.07944154]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4, 7, 8]))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "39bd657f", + "metadata": { + "editable": true + }, + "source": [ + "In the last example we used Numpy's unary function $np.log$. This function is\n", + "highly tuned to compute array elements since the code is vectorized\n", + "and does not require looping. We normaly recommend that you use the\n", + "Numpy intrinsic functions instead of the corresponding **log** function\n", + "from Python's **math** module. The looping is done explicitely by the\n", + "**np.log** function. The alternative, and slower way to compute the\n", + "logarithms of a vector would be to write" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "e5a5fe48", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1 1 2]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "from math import log\n", + "x = np.array([4, 7, 8])\n", + "for i in range(0, len(x)):\n", + " x[i] = log(x[i])\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "d74eb874", + "metadata": { + "editable": true + }, + "source": [ + "We note that our code is much longer already and we need to import the **log** function from the **math** module. \n", + "The attentive reader will also notice that the output is $[1, 1, 2]$. Python interprets automagically our numbers as integers (like the **automatic** keyword in C++). To change this we could define our array elements to be double precision numbers as" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "1d657e01", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1.38629436 1.94591015 2.07944154]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4, 7, 8], dtype = np.float64))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "7f82621e", + "metadata": { + "editable": true + }, + "source": [ + "or simply write them as double precision numbers (Python uses 64 bits as default for floating point type variables), that is" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "c4a5e02b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1.38629436 1.94591015 2.07944154]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4.0, 7.0, 8.0]))\n", + "print(x)" + ] + }, + { + "cell_type": "markdown", + "id": "3af71bb0", + "metadata": { + "editable": true + }, + "source": [ + "To check the number of bytes (remember that one byte contains eight bits for double precision variables), you can use simple use the **itemsize** functionality (the array $x$ is actually an object which inherits the functionalities defined in Numpy) as" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "id": "5cf0baf0", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "8\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "x = np.log(np.array([4.0, 7.0, 8.0]))\n", + "print(x.itemsize)" + ] + }, + { + "cell_type": "markdown", + "id": "a70a4db0", + "metadata": { + "editable": true + }, + "source": [ + "## Matrices in Python\n", + "\n", + "Having defined vectors, we are now ready to try out matrices. We can\n", + "define a $3 \\times 3 $ real matrix $\\boldsymbol{A}$ as (recall that we user\n", + "lowercase letters for vectors and uppercase letters for matrices)" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "262e8f46", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[1.38629436 1.94591015 2.07944154]\n", + " [1.09861229 2.30258509 2.39789527]\n", + " [1.38629436 1.60943791 1.94591015]]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "0f43d9cd", + "metadata": { + "editable": true + }, + "source": [ + "If we use the **shape** function we would get $(3, 3)$ as output, that is verifying that our matrix is a $3\\times 3$ matrix. We can slice the matrix and print for example the first column (Python organized matrix elements in a row-major order, see below) as" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "f12d591b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1.38629436 1.09861229 1.38629436]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "# print the first column, row-major order and elements start with 0\n", + "print(A[:,0])" + ] + }, + { + "cell_type": "markdown", + "id": "7c1cc7ed", + "metadata": { + "editable": true + }, + "source": [ + "We can continue this was by printing out other columns or rows. The example here prints out the second column" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "1f365f00", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[1.09861229 2.30258509 2.39789527]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ]))\n", + "# print the first column, row-major order and elements start with 0\n", + "print(A[1,:])" + ] + }, + { + "cell_type": "markdown", + "id": "d5cf7244", + "metadata": { + "editable": true + }, + "source": [ + "Numpy contains many other functionalities that allow us to slice, subdivide etc etc arrays. We strongly recommend that you look up the [Numpy website for more details](http://www.numpy.org/). Useful functions when defining a matrix are the **np.zeros** function which declares a matrix of a given dimension and sets all elements to zero" + ] + }, + { + "cell_type": "code", + "execution_count": 12, + "id": "907260e4", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to zero\n", + "A = np.zeros( (n, n) )\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "375e0579", + "metadata": { + "editable": true + }, + "source": [ + "or initializing all elements to" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "79e19787", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]\n", + " [1. 1. 1. 1. 1. 1. 1. 1. 1. 1.]]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to one\n", + "A = np.ones( (n, n) )\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "be6c16b2", + "metadata": { + "editable": true + }, + "source": [ + "or as unitarily distributed random numbers (see the material on random number generators in the statistics part)" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "901ecfe2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[0.99374388 0.68029036 0.03324117 0.4017308 0.92888885 0.12434807\n", + " 0.9538239 0.15899253 0.48442336 0.72057967]\n", + " [0.16840081 0.01852793 0.10950766 0.25522012 0.78674758 0.65280103\n", + " 0.220968 0.61634314 0.25115233 0.05622775]\n", + " [0.87927601 0.35147532 0.62432074 0.16167387 0.49242688 0.12934015\n", + " 0.50812169 0.35123041 0.95054046 0.66638078]\n", + " [0.08887708 0.95994091 0.21927988 0.05833369 0.49974854 0.20999563\n", + " 0.28102554 0.01479194 0.84942942 0.30702075]\n", + " [0.29672381 0.99820454 0.56848886 0.32190363 0.13320535 0.31768693\n", + " 0.80280228 0.59654302 0.25116851 0.81681019]\n", + " [0.50799086 0.13261136 0.33078983 0.53623614 0.14174062 0.06662062\n", + " 0.32183101 0.8338786 0.32766274 0.88595926]\n", + " [0.12539459 0.60176806 0.12885048 0.00708794 0.66614025 0.70727837\n", + " 0.87849647 0.05491655 0.37157712 0.93325497]\n", + " [0.56343926 0.23056012 0.97840468 0.82484316 0.9956582 0.53822787\n", + " 0.53219937 0.58312967 0.14617985 0.59468796]\n", + " [0.67054028 0.4304167 0.38299798 0.16397496 0.81912272 0.54894095\n", + " 0.53254035 0.17826349 0.01551946 0.54969168]\n", + " [0.92373266 0.86361743 0.27525917 0.59367954 0.85843684 0.40389922\n", + " 0.8074242 0.59323905 0.51075117 0.79552752]]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "n = 10\n", + "# define a matrix of dimension 10 x 10 and set all elements to random numbers with x \\in [0, 1]\n", + "A = np.random.rand(n, n)\n", + "print(A)" + ] + }, + { + "cell_type": "markdown", + "id": "be629f7b", + "metadata": { + "editable": true + }, + "source": [ + "As we will see throughout these lectures, there are several extremely useful functionalities in Numpy.\n", + "As an example, consider the discussion of the covariance matrix. Suppose we have defined three vectors\n", + "$\\boldsymbol{x}, \\boldsymbol{y}, \\boldsymbol{z}$ with $n$ elements each. The covariance matrix is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "980cf441", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\Sigma} = \\begin{bmatrix} \\sigma_{xx} & \\sigma_{xy} & \\sigma_{xz} \\\\\n", + " \\sigma_{yx} & \\sigma_{yy} & \\sigma_{yz} \\\\\n", + " \\sigma_{zx} & \\sigma_{zy} & \\sigma_{zz} \n", + " \\end{bmatrix},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "d2ffe1f6", + "metadata": { + "editable": true + }, + "source": [ + "where for example" + ] + }, + { + "cell_type": "markdown", + "id": "4f4efdd4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma_{xy} =\\frac{1}{n} \\sum_{i=0}^{n-1}(x_i- \\overline{x})(y_i- \\overline{y}).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ee572849", + "metadata": { + "editable": true + }, + "source": [ + "The Numpy function **np.cov** calculates the covariance elements using the factor $1/(n-1)$ instead of $1/n$ since it assumes we do not have the exact mean values. \n", + "The following simple function uses the **np.vstack** function which takes each vector of dimension $1\\times n$ and produces a $3\\times n$ matrix $\\boldsymbol{W}$" + ] + }, + { + "cell_type": "markdown", + "id": "df0a5acc", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{W} = \\begin{bmatrix} x_0 & x_1 & x_2 & \\dots & x_{n-2} & x_{n-1} \\\\\n", + " y_0 & y_1 & y_2 & \\dots & y_{n-2} & y_{n-1} \\\\\n", + "\t\t\t z_0 & z_1 & z_2 & \\dots & z_{n-2} & z_{n-1} \\\\\n", + " \\end{bmatrix},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7bc72a77", + "metadata": { + "editable": true + }, + "source": [ + "which in turn is converted into into the $3\\times 3$ covariance matrix\n", + "$\\boldsymbol{\\Sigma}$ via the Numpy function **np.cov()**. We note that we can also calculate\n", + "the mean value of each set of samples $\\boldsymbol{x}$ etc using the Numpy\n", + "function **np.mean(x)**. We can also extract the eigenvalues of the\n", + "covariance matrix through the **np.linalg.eig()** function." + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "4c493f06", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "-0.007334080205654388\n", + "3.9196398947631166\n", + "0.49680475019226\n", + "[[ 0.83982725 2.39595799 2.20255857]\n", + " [ 2.39595799 8.04496448 5.72101205]\n", + " [ 2.20255857 5.72101205 10.77494715]]\n", + "[15.97477238 0.08363087 3.60133562]\n" + ] + } + ], + "source": [ + "# Importing various packages\n", + "import numpy as np\n", + "\n", + "n = 100\n", + "x = np.random.normal(size=n)\n", + "print(np.mean(x))\n", + "y = 4+3*x+np.random.normal(size=n)\n", + "print(np.mean(y))\n", + "z = x**3+np.random.normal(size=n)\n", + "print(np.mean(z))\n", + "W = np.vstack((x, y, z))\n", + "Sigma = np.cov(W)\n", + "print(Sigma)\n", + "Eigvals, Eigvecs = np.linalg.eig(Sigma)\n", + "print(Eigvals)" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "20bda99f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[1. 0. 0. 0.]\n", + " [0. 1. 0. 0.]\n", + " [0. 0. 1. 0.]\n", + " [0. 0. 0. 1.]]\n", + " (0, 0)\t1.0\n", + " (1, 1)\t1.0\n", + " (2, 2)\t1.0\n", + " (3, 3)\t1.0\n" + ] + }, + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_74_1.png" + } + }, + "output_type": "display_data" + } + ], + "source": [ + "%matplotlib inline\n", + "\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from scipy import sparse\n", + "eye = np.eye(4)\n", + "print(eye)\n", + "sparse_mtx = sparse.csr_matrix(eye)\n", + "print(sparse_mtx)\n", + "x = np.linspace(-10,10,100)\n", + "y = np.sin(x)\n", + "plt.plot(x,y,marker='x')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "a9bff117", + "metadata": { + "editable": true + }, + "source": [ + "## Meet the Pandas\n", + "\n", + "\n", + "\n", + "\n", + "

Figure 1:

\n", + "\n", + "\n", + "Another useful Python package is\n", + "[pandas](https://pandas.pydata.org/), which is an open source library\n", + "providing high-performance, easy-to-use data structures and data\n", + "analysis tools for Python. **pandas** stands for panel data, a term borrowed from econometrics and is an efficient library for data analysis with an emphasis on tabular data.\n", + "**pandas** has two major classes, the **DataFrame** class with two-dimensional data objects and tabular data organized in columns and the class **Series** with a focus on one-dimensional data objects. Both classes allow you to index data easily as we will see in the examples below. \n", + "**pandas** allows you also to perform mathematical operations on the data, spanning from simple reshapings of vectors and matrices to statistical operations. \n", + "\n", + "The following simple example shows how we can, in an easy way make tables of our data. Here we define a data set which includes names, place of birth and date of birth, and displays the data in an easy to read way. We will see repeated use of **pandas**, in particular in connection with classification of data." + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "9ce8cf97", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
First NameLast NamePlace of birthDate of Birth T.A.
0FrodoBagginsShire2968
1BilboBagginsShire2890
2Aragorn IIElessarEriador2931
3SamwiseGamgeeShire2980
\n", + "
" + ], + "text/plain": [ + " First Name Last Name Place of birth Date of Birth T.A.\n", + "0 Frodo Baggins Shire 2968\n", + "1 Bilbo Baggins Shire 2890\n", + "2 Aragorn II Elessar Eriador 2931\n", + "3 Samwise Gamgee Shire 2980" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "import pandas as pd\n", + "from IPython.display import display\n", + "data = {'First Name': [\"Frodo\", \"Bilbo\", \"Aragorn II\", \"Samwise\"],\n", + " 'Last Name': [\"Baggins\", \"Baggins\",\"Elessar\",\"Gamgee\"],\n", + " 'Place of birth': [\"Shire\", \"Shire\", \"Eriador\", \"Shire\"],\n", + " 'Date of Birth T.A.': [2968, 2890, 2931, 2980]\n", + " }\n", + "data_pandas = pd.DataFrame(data)\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "98b2cebb", + "metadata": { + "editable": true + }, + "source": [ + "In the above we have imported **pandas** with the shorthand **pd**, the latter has become the standard way we import **pandas**. We make then a list of various variables\n", + "and reorganize the aboves lists into a **DataFrame** and then print out a neat table with specific column labels as *Name*, *place of birth* and *date of birth*.\n", + "Displaying these results, we see that the indices are given by the default numbers from zero to three.\n", + "**pandas** is extremely flexible and we can easily change the above indices by defining a new type of indexing as" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "88a877c3", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
First NameLast NamePlace of birthDate of Birth T.A.
FrodoFrodoBagginsShire2968
BilboBilboBagginsShire2890
AragornAragorn IIElessarEriador2931
SamSamwiseGamgeeShire2980
\n", + "
" + ], + "text/plain": [ + " First Name Last Name Place of birth Date of Birth T.A.\n", + "Frodo Frodo Baggins Shire 2968\n", + "Bilbo Bilbo Baggins Shire 2890\n", + "Aragorn Aragorn II Elessar Eriador 2931\n", + "Sam Samwise Gamgee Shire 2980" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "data_pandas = pd.DataFrame(data,index=['Frodo','Bilbo','Aragorn','Sam'])\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "0928de6c", + "metadata": { + "editable": true + }, + "source": [ + "Thereafter we display the content of the row which begins with the index **Aragorn**" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "d93baec5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/plain": [ + "First Name Aragorn II\n", + "Last Name Elessar\n", + "Place of birth Eriador\n", + "Date of Birth T.A. 2931\n", + "Name: Aragorn, dtype: object" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "display(data_pandas.loc['Aragorn'])" + ] + }, + { + "cell_type": "markdown", + "id": "7a5b03a4", + "metadata": { + "editable": true + }, + "source": [ + "We can easily append data to this, for example" + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "2b5968bf", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/var/folders/td/3yk470mj5p931p9dtkk0y6jw0000gn/T/ipykernel_65914/1326197715.py:6: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n", + " data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin']))\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
First NameLast NamePlace of birthDate of Birth T.A.
FrodoFrodoBagginsShire2968
BilboBilboBagginsShire2890
AragornAragorn IIElessarEriador2931
SamSamwiseGamgeeShire2980
PippinPeregrinTookShire2990
\n", + "
" + ], + "text/plain": [ + " First Name Last Name Place of birth Date of Birth T.A.\n", + "Frodo Frodo Baggins Shire 2968\n", + "Bilbo Bilbo Baggins Shire 2890\n", + "Aragorn Aragorn II Elessar Eriador 2931\n", + "Sam Samwise Gamgee Shire 2980\n", + "Pippin Peregrin Took Shire 2990" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "new_hobbit = {'First Name': [\"Peregrin\"],\n", + " 'Last Name': [\"Took\"],\n", + " 'Place of birth': [\"Shire\"],\n", + " 'Date of Birth T.A.': [2990]\n", + " }\n", + "data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin']))\n", + "display(data_pandas)" + ] + }, + { + "cell_type": "markdown", + "id": "281789dd", + "metadata": { + "editable": true + }, + "source": [ + "Here are other examples where we use the **DataFrame** functionality to handle arrays, now with more interesting features for us, namely numbers. We set up a matrix \n", + "of dimensionality $10\\times 5$ and compute the mean value and standard deviation of each column. Similarly, we can perform mathematial operations like squaring the matrix elements and many other operations." + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "226cafd5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
01234
0-1.7497650.3426801.153036-0.2524360.981321
10.5142190.221180-1.070043-0.1894960.255001
2-0.4580270.435163-0.5835950.8168470.672721
3-0.104411-0.5312801.029733-0.438136-1.118318
41.6189821.541605-0.251879-0.8424360.184519
50.9370820.7310001.361556-0.3262380.055676
60.222400-1.443217-0.7563520.8164540.750445
7-0.4559471.189622-1.690617-1.356399-1.232435
8-0.544439-0.6681720.007315-0.6129391.299748
9-1.733096-0.9833100.357508-1.6135791.470714
\n", + "
" + ], + "text/plain": [ + " 0 1 2 3 4\n", + "0 -1.749765 0.342680 1.153036 -0.252436 0.981321\n", + "1 0.514219 0.221180 -1.070043 -0.189496 0.255001\n", + "2 -0.458027 0.435163 -0.583595 0.816847 0.672721\n", + "3 -0.104411 -0.531280 1.029733 -0.438136 -1.118318\n", + "4 1.618982 1.541605 -0.251879 -0.842436 0.184519\n", + "5 0.937082 0.731000 1.361556 -0.326238 0.055676\n", + "6 0.222400 -1.443217 -0.756352 0.816454 0.750445\n", + "7 -0.455947 1.189622 -1.690617 -1.356399 -1.232435\n", + "8 -0.544439 -0.668172 0.007315 -0.612939 1.299748\n", + "9 -1.733096 -0.983310 0.357508 -1.613579 1.470714" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "0 -0.175300\n", + "1 0.083527\n", + "2 -0.044334\n", + "3 -0.399836\n", + "4 0.331939\n", + "dtype: float64\n", + "0 1.069584\n", + "1 0.965548\n", + "2 1.018232\n", + "3 0.793167\n", + "4 0.918992\n", + "dtype: float64\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
01234
03.0616790.1174301.3294920.0637240.962990
10.2644210.0489201.1449930.0359090.065026
20.2097890.1893670.3405830.6672390.452553
30.0109020.2822591.0603490.1919631.250636
42.6211022.3765470.0634430.7096980.034047
50.8781230.5343621.8538350.1064310.003100
60.0494622.0828750.5720690.6665970.563167
70.2078881.4152012.8581851.8398181.518895
80.2964140.4464530.0000540.3756941.689345
93.0036200.9668990.1278122.6036362.162999
\n", + "
" + ], + "text/plain": [ + " 0 1 2 3 4\n", + "0 3.061679 0.117430 1.329492 0.063724 0.962990\n", + "1 0.264421 0.048920 1.144993 0.035909 0.065026\n", + "2 0.209789 0.189367 0.340583 0.667239 0.452553\n", + "3 0.010902 0.282259 1.060349 0.191963 1.250636\n", + "4 2.621102 2.376547 0.063443 0.709698 0.034047\n", + "5 0.878123 0.534362 1.853835 0.106431 0.003100\n", + "6 0.049462 2.082875 0.572069 0.666597 0.563167\n", + "7 0.207888 1.415201 2.858185 1.839818 1.518895\n", + "8 0.296414 0.446453 0.000054 0.375694 1.689345\n", + "9 3.003620 0.966899 0.127812 2.603636 2.162999" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "import numpy as np\n", + "import pandas as pd\n", + "from IPython.display import display\n", + "np.random.seed(100)\n", + "# setting up a 10 x 5 matrix\n", + "rows = 10\n", + "cols = 5\n", + "a = np.random.randn(rows,cols)\n", + "df = pd.DataFrame(a)\n", + "display(df)\n", + "print(df.mean())\n", + "print(df.std())\n", + "display(df**2)" + ] + }, + { + "cell_type": "markdown", + "id": "34d1241d", + "metadata": { + "editable": true + }, + "source": [ + "Thereafter we can select specific columns only and plot final results" + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "id": "66de0f40", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
FirstSecondThirdFourthFifth
0-1.7497650.3426801.153036-0.2524360.981321
10.5142190.221180-1.070043-0.1894960.255001
2-0.4580270.435163-0.5835950.8168470.672721
3-0.104411-0.5312801.029733-0.438136-1.118318
41.6189821.541605-0.251879-0.8424360.184519
50.9370820.7310001.361556-0.3262380.055676
60.222400-1.443217-0.7563520.8164540.750445
7-0.4559471.189622-1.690617-1.356399-1.232435
8-0.544439-0.6681720.007315-0.6129391.299748
9-1.733096-0.9833100.357508-1.6135791.470714
\n", + "
" + ], + "text/plain": [ + " First Second Third Fourth Fifth\n", + "0 -1.749765 0.342680 1.153036 -0.252436 0.981321\n", + "1 0.514219 0.221180 -1.070043 -0.189496 0.255001\n", + "2 -0.458027 0.435163 -0.583595 0.816847 0.672721\n", + "3 -0.104411 -0.531280 1.029733 -0.438136 -1.118318\n", + "4 1.618982 1.541605 -0.251879 -0.842436 0.184519\n", + "5 0.937082 0.731000 1.361556 -0.326238 0.055676\n", + "6 0.222400 -1.443217 -0.756352 0.816454 0.750445\n", + "7 -0.455947 1.189622 -1.690617 -1.356399 -1.232435\n", + "8 -0.544439 -0.668172 0.007315 -0.612939 1.299748\n", + "9 -1.733096 -0.983310 0.357508 -1.613579 1.470714" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "0.08352721390288316\n", + "\n", + "Int64Index: 10 entries, 0 to 9\n", + "Data columns (total 5 columns):\n", + " # Column Non-Null Count Dtype \n", + "--- ------ -------------- ----- \n", + " 0 First 10 non-null float64\n", + " 1 Second 10 non-null float64\n", + " 2 Third 10 non-null float64\n", + " 3 Fourth 10 non-null float64\n", + " 4 Fifth 10 non-null float64\n", + "dtypes: float64(5)\n", + "memory usage: 480.0 bytes\n", + "None\n", + " First Second Third Fourth Fifth\n", + "count 10.000000 10.000000 10.000000 10.000000 10.000000\n", + "mean -0.175300 0.083527 -0.044334 -0.399836 0.331939\n", + "std 1.069584 0.965548 1.018232 0.793167 0.918992\n", + "min -1.749765 -1.443217 -1.690617 -1.613579 -1.232435\n", + "25% -0.522836 -0.633949 -0.713163 -0.785061 0.087887\n", + "50% -0.280179 0.281930 -0.122282 -0.382187 0.463861\n", + "75% 0.441264 0.657041 0.861676 -0.205231 0.923602\n", + "max 1.618982 1.541605 1.361556 0.816847 1.470714\n" + ] + }, + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_86_2.png" + } + }, + "output_type": "display_data" + }, + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_86_3.png" + } + }, + "output_type": "display_data" + } + ], + "source": [ + "df.columns = ['First', 'Second', 'Third', 'Fourth', 'Fifth']\n", + "df.index = np.arange(10)\n", + "\n", + "display(df)\n", + "print(df['Second'].mean() )\n", + "print(df.info())\n", + "print(df.describe())\n", + "\n", + "from pylab import plt, mpl\n", + "plt.style.use('seaborn')\n", + "mpl.rcParams['font.family'] = 'serif'\n", + "\n", + "df.cumsum().plot(lw=2.0, figsize=(10,6))\n", + "plt.show()\n", + "\n", + "\n", + "df.plot.bar(figsize=(10,6), rot=15)\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "fce1b00d", + "metadata": { + "editable": true + }, + "source": [ + "We can produce a $4\\times 4$ matrix" + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "id": "b31ec8b2", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[[ 0 1 2 3]\n", + " [ 4 5 6 7]\n", + " [ 8 9 10 11]\n", + " [12 13 14 15]]\n", + " 0 1 2 3\n", + "0 0 1 2 3\n", + "1 4 5 6 7\n", + "2 8 9 10 11\n", + "3 12 13 14 15\n" + ] + } + ], + "source": [ + "b = np.arange(16).reshape((4,4))\n", + "print(b)\n", + "df1 = pd.DataFrame(b)\n", + "print(df1)" + ] + }, + { + "cell_type": "markdown", + "id": "399b4f0b", + "metadata": { + "editable": true + }, + "source": [ + "and many other operations. \n", + "\n", + "The **Series** class is another important class included in\n", + "**pandas**. You can view it as a specialization of **DataFrame** but where\n", + "we have just a single column of data. It shares many of the same features as **DataFrame**. As with **DataFrame**,\n", + "most operations are vectorized, achieving thereby a high performance when dealing with computations of arrays, in particular labeled arrays.\n", + "As we will see below it leads also to a very concice code close to the mathematical operations we may be interested in.\n", + "For multidimensional arrays, we recommend strongly [xarray](http://xarray.pydata.org/en/stable/). **xarray** has much of the same flexibility as **pandas**, but allows for the extension to higher dimensions than two. We will see examples later of the usage of both **pandas** and **xarray**." + ] + }, + { + "cell_type": "markdown", + "id": "b3c66ed4", + "metadata": { + "editable": true + }, + "source": [ + "### Simple linear regression model using **scikit-learn**\n", + "\n", + "We start with perhaps our simplest possible example, using **Scikit-Learn** to perform linear regression analysis on a data set produced by us. \n", + "\n", + "What follows is a simple Python code where we have defined a function\n", + "$y$ in terms of the variable $x$. Both are defined as vectors with $100$ entries. \n", + "The numbers in the vector $\\boldsymbol{x}$ are given\n", + "by random numbers generated with a uniform distribution with entries\n", + "$x_i \\in [0,1]$ (more about probability distribution functions\n", + "later). These values are then used to define a function $y(x)$\n", + "(tabulated again as a vector) with a linear dependence on $x$ plus a\n", + "random noise added via the normal distribution.\n", + "\n", + "The Numpy functions are imported used the **import numpy as np**\n", + "statement and the random number generator for the uniform distribution\n", + "is called using the function **np.random.rand()**, where we specificy\n", + "that we want $100$ random variables. Using Numpy we define\n", + "automatically an array with the specified number of elements, $100$ in\n", + "our case. With the Numpy function **randn()** we can compute random\n", + "numbers with the normal distribution (mean value $\\mu$ equal to zero and\n", + "variance $\\sigma^2$ set to one) and produce the values of $y$ assuming a linear\n", + "dependence as function of $x$" + ] + }, + { + "cell_type": "markdown", + "id": "b7ec1f01", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y = 2x+N(0,1),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "5c4c6423", + "metadata": { + "editable": true + }, + "source": [ + "where $N(0,1)$ represents random numbers generated by the normal\n", + "distribution. From **Scikit-Learn** we import then the\n", + "**LinearRegression** functionality and make a prediction $\\tilde{y} =\n", + "\\alpha + \\beta x$ using the function **fit(x,y)**. We call the set of\n", + "data $(\\boldsymbol{x},\\boldsymbol{y})$ for our training data. The Python package\n", + "**scikit-learn** has also a functionality which extracts the above\n", + "fitting parameters $\\alpha$ and $\\beta$ (see below). Later we will\n", + "distinguish between training data and test data.\n", + "\n", + "For plotting we use the Python package\n", + "[matplotlib](https://matplotlib.org/) which produces publication\n", + "quality figures. Feel free to explore the extensive\n", + "[gallery](https://matplotlib.org/gallery/index.html) of examples. In\n", + "this example we plot our original values of $x$ and $y$ as well as the\n", + "prediction **ypredict** ($\\tilde{y}$), which attempts at fitting our\n", + "data with a straight line.\n", + "\n", + "The Python code follows here." + ] + }, + { + "cell_type": "code", + "execution_count": 24, + "id": "b026b228", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_93_0.png" + } + }, + "output_type": "display_data" + } + ], + "source": [ + "# Importing various packages\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 2*x+np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "xnew = np.array([[0],[1]])\n", + "ypredict = linreg.predict(xnew)\n", + "\n", + "plt.plot(xnew, ypredict, \"r-\")\n", + "plt.plot(x, y ,'ro')\n", + "plt.axis([0,1.0,0, 5.0])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$y$')\n", + "plt.title(r'Simple Linear Regression')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "121762a4", + "metadata": { + "editable": true + }, + "source": [ + "This example serves several aims. It allows us to demonstrate several\n", + "aspects of data analysis and later machine learning algorithms. The\n", + "immediate visualization shows that our linear fit is not\n", + "impressive. It goes through the data points, but there are many\n", + "outliers which are not reproduced by our linear regression. We could\n", + "now play around with this small program and change for example the\n", + "factor in front of $x$ and the normal distribution. Try to change the\n", + "function $y$ to" + ] + }, + { + "cell_type": "markdown", + "id": "60728127", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y = 10x+0.01 \\times N(0,1),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "acf5b63f", + "metadata": { + "editable": true + }, + "source": [ + "where $x$ is defined as before. Does the fit look better? Indeed, by\n", + "reducing the role of the noise given by the normal distribution we see immediately that\n", + "our linear prediction seemingly reproduces better the training\n", + "set. However, this testing 'by the eye' is obviouly not satisfactory in the\n", + "long run. Here we have only defined the training data and our model, and \n", + "have not discussed a more rigorous approach to the **cost** function.\n", + "\n", + "We need more rigorous criteria in defining whether we have succeeded or\n", + "not in modeling our training data. You will be surprised to see that\n", + "many scientists seldomly venture beyond this 'by the eye' approach. A\n", + "standard approach for the *cost* function is the so-called $\\chi^2$\n", + "function (a variant of the mean-squared error (MSE))" + ] + }, + { + "cell_type": "markdown", + "id": "40040656", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\chi^2 = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}\\frac{(y_i-\\tilde{y}_i)^2}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "86948166", + "metadata": { + "editable": true + }, + "source": [ + "where $\\sigma_i^2$ is the variance (to be defined later) of the entry\n", + "$y_i$. We may not know the explicit value of $\\sigma_i^2$, it serves\n", + "however the aim of scaling the equations and make the cost function\n", + "dimensionless. \n", + "\n", + "Minimizing the cost function is a central aspect of\n", + "our discussions to come. Finding its minima as function of the model\n", + "parameters ($\\alpha$ and $\\beta$ in our case) will be a recurring\n", + "theme in these series of lectures. Essentially all machine learning\n", + "algorithms we will discuss center around the minimization of the\n", + "chosen cost function. This depends in turn on our specific\n", + "model for describing the data, a typical situation in supervised\n", + "learning. Automatizing the search for the minima of the cost function is a\n", + "central ingredient in all algorithms. Typical methods which are\n", + "employed are various variants of **gradient** methods. These will be\n", + "discussed in more detail later. Again, you'll be surprised to hear that\n", + "many practitioners minimize the above function ''by the eye', popularly dubbed as \n", + "'chi by the eye'. That is, change a parameter and see (visually and numerically) that \n", + "the $\\chi^2$ function becomes smaller. \n", + "\n", + "There are many ways to define the cost function. A simpler approach is to look at the relative difference between the training data and the predicted data, that is we define \n", + "the relative error (why would we prefer the MSE instead of the relative error?) as" + ] + }, + { + "cell_type": "markdown", + "id": "2c7af90a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\epsilon_{\\mathrm{relative}}= \\frac{\\vert \\boldsymbol{y} -\\boldsymbol{\\tilde{y}}\\vert}{\\vert \\boldsymbol{y}\\vert}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "aebc9af3", + "metadata": { + "editable": true + }, + "source": [ + "The squared cost function results in an arithmetic mean-unbiased\n", + "estimator, and the absolute-value cost function results in a\n", + "median-unbiased estimator (in the one-dimensional case, and a\n", + "geometric median-unbiased estimator for the multi-dimensional\n", + "case). The squared cost function has the disadvantage that it has the tendency\n", + "to be dominated by outliers.\n", + "\n", + "We can modify easily the above Python code and plot the relative error instead" + ] + }, + { + "cell_type": "code", + "execution_count": 25, + "id": "eaa61d49", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_101_0.png" + } + }, + "output_type": "display_data" + } + ], + "source": [ + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 5*x+0.01*np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "ypredict = linreg.predict(x)\n", + "\n", + "plt.plot(x, np.abs(ypredict-y)/abs(y), \"ro\")\n", + "plt.axis([0,1.0,0.0, 0.5])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$\\epsilon_{\\mathrm{relative}}$')\n", + "plt.title(r'Relative error')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "f964fde3", + "metadata": { + "editable": true + }, + "source": [ + "Depending on the parameter in front of the normal distribution, we may\n", + "have a small or larger relative error. Try to play around with\n", + "different training data sets and study (graphically) the value of the\n", + "relative error.\n", + "\n", + "As mentioned above, **Scikit-Learn** has an impressive functionality.\n", + "We can for example extract the values of $\\alpha$ and $\\beta$ and\n", + "their error estimates, or the variance and standard deviation and many\n", + "other properties from the statistical data analysis. \n", + "\n", + "Here we show an\n", + "example of the functionality of **Scikit-Learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 26, + "id": "c7fe9388", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "The intercept alpha: \n", + " [2.18780801]\n", + "Coefficient beta : \n", + " [[4.72228205]]\n", + "Mean squared error: 0.37\n", + "Variance score: 0.83\n", + "Mean squared log error: 0.01\n", + "Mean absolute error: 0.47\n" + ] + }, + { + "data": { + "image/png": "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\n", + "text/plain": [ + "
" + ] + }, + "metadata": { + "filenames": { + "image/png": "/Users/mhjensen/Teaching/MachineLearning/doc/LectureNotes/_build/jupyter_execute/week34_103_1.png" + } + }, + "output_type": "display_data" + } + ], + "source": [ + "import numpy as np \n", + "import matplotlib.pyplot as plt \n", + "from sklearn.linear_model import LinearRegression \n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_squared_log_error, mean_absolute_error\n", + "\n", + "x = np.random.rand(100,1)\n", + "y = 2.0+ 5*x+0.5*np.random.randn(100,1)\n", + "linreg = LinearRegression()\n", + "linreg.fit(x,y)\n", + "ypredict = linreg.predict(x)\n", + "print('The intercept alpha: \\n', linreg.intercept_)\n", + "print('Coefficient beta : \\n', linreg.coef_)\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(y, ypredict))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(y, ypredict))\n", + "# Mean squared log error \n", + "print('Mean squared log error: %.2f' % mean_squared_log_error(y, ypredict) )\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(y, ypredict))\n", + "plt.plot(x, ypredict, \"r-\")\n", + "plt.plot(x, y ,'ro')\n", + "plt.axis([0.0,1.0,1.5, 7.0])\n", + "plt.xlabel(r'$x$')\n", + "plt.ylabel(r'$y$')\n", + "plt.title(r'Linear Regression fit ')\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "14ad007a", + "metadata": { + "editable": true + }, + "source": [ + "The function **coef** gives us the parameter $\\beta$ of our fit while **intercept** yields \n", + "$\\alpha$. Depending on the constant in front of the normal distribution, we get values near or far from $\\alpha =2$ and $\\beta =5$. Try to play around with different parameters in front of the normal distribution. The function **meansquarederror** gives us the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error or loss defined as" + ] + }, + { + "cell_type": "markdown", + "id": "7dc68b06", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "471f97ed", + "metadata": { + "editable": true + }, + "source": [ + "The smaller the value, the better the fit. Ideally we would like to\n", + "have an MSE equal zero. The attentive reader has probably recognized\n", + "this function as being similar to the $\\chi^2$ function defined above.\n", + "\n", + "The **r2score** function computes $R^2$, the coefficient of\n", + "determination. It provides a measure of how well future samples are\n", + "likely to be predicted by the model. Best possible score is 1.0 and it\n", + "can be negative (because the model can be arbitrarily worse). A\n", + "constant model that always predicts the expected value of $\\boldsymbol{y}$,\n", + "disregarding the input features, would get a $R^2$ score of $0.0$.\n", + "\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "65a80e19", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "37e1d624", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "a81b3061", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "792fa6a1", + "metadata": { + "editable": true + }, + "source": [ + "Another quantity taht we will meet again in our discussions of regression analysis is \n", + " the mean absolute error (MAE), a risk metric corresponding to the expected value of the absolute error loss or what we call the $l1$-norm loss. In our discussion above we presented the relative error.\n", + "The MAE is defined as follows" + ] + }, + { + "cell_type": "markdown", + "id": "2f6df05c", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\text{MAE}(\\boldsymbol{y}, \\boldsymbol{\\tilde{y}}) = \\frac{1}{n} \\sum_{i=0}^{n-1} \\left| y_i - \\tilde{y}_i \\right|.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "9eb85874", + "metadata": { + "editable": true + }, + "source": [ + "We present the \n", + "squared logarithmic (quadratic) error" + ] + }, + { + "cell_type": "markdown", + "id": "a98d9555", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\text{MSLE}(\\boldsymbol{y}, \\boldsymbol{\\tilde{y}}) = \\frac{1}{n} \\sum_{i=0}^{n - 1} (\\log_e (1 + y_i) - \\log_e (1 + \\tilde{y}_i) )^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6766ecc9", + "metadata": { + "editable": true + }, + "source": [ + "where $\\log_e (x)$ stands for the natural logarithm of $x$. This error\n", + "estimate is best to use when targets having exponential growth, such\n", + "as population counts, average sales of a commodity over a span of\n", + "years etc. \n", + "\n", + "Finally, another cost function is the Huber cost function used in robust regression.\n", + "\n", + "The rationale behind this possible cost function is its reduced\n", + "sensitivity to outliers in the data set. In our discussions on\n", + "dimensionality reduction and normalization of data we will meet other\n", + "ways of dealing with outliers.\n", + "\n", + "The Huber cost function is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "fb3a9871", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "H_{\\delta}(\\boldsymbol{a})=\\left\\{\\begin{array}{cc}\\frac{1}{2} \\boldsymbol{a}^{2}& \\text{for }|\\boldsymbol{a}|\\leq \\delta\\\\ \\delta (|\\boldsymbol{a}|-\\frac{1}{2}\\delta ),&\\text{otherwise}.\\end{array}\\right.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "57226a58", + "metadata": { + "editable": true + }, + "source": [ + "Here $\\boldsymbol{a}=\\boldsymbol{y} - \\boldsymbol{\\tilde{y}}$.\n", + "\n", + "We will discuss in more detail these and other functions in the\n", + "various lectures and lab sessions." + ] + }, + { + "cell_type": "markdown", + "id": "b333e60b", + "metadata": { + "editable": true + }, + "source": [ + "### To our real data: nuclear binding energies. Brief reminder on masses and binding energies\n", + "\n", + "Let us now dive into nuclear physics and remind ourselves briefly about some basic features about binding\n", + "energies. A basic quantity which can be measured for the ground\n", + "states of nuclei is the atomic mass $M(N, Z)$ of the neutral atom with\n", + "atomic mass number $A$ and charge $Z$. The number of neutrons is $N$. There are indeed several sophisticated experiments worldwide which allow us to measure this quantity to high precision (parts per million even). \n", + "\n", + "Atomic masses are usually tabulated in terms of the mass excess defined by" + ] + }, + { + "cell_type": "markdown", + "id": "1adb10e6", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\Delta M(N, Z) = M(N, Z) - uA,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "77592d4e", + "metadata": { + "editable": true + }, + "source": [ + "where $u$ is the Atomic Mass Unit" + ] + }, + { + "cell_type": "markdown", + "id": "ba90405f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "u = M(^{12}\\mathrm{C})/12 = 931.4940954(57) \\hspace{0.1cm} \\mathrm{MeV}/c^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "db7cef69", + "metadata": { + "editable": true + }, + "source": [ + "The nucleon masses are" + ] + }, + { + "cell_type": "markdown", + "id": "e12b557f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "m_p = 1.00727646693(9)u,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6f5462e0", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "0247179e", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "m_n = 939.56536(8)\\hspace{0.1cm} \\mathrm{MeV}/c^2 = 1.0086649156(6)u.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "cb6e7ad5", + "metadata": { + "editable": true + }, + "source": [ + "In the [2016 mass evaluation of by W.J.Huang, G.Audi, M.Wang, F.G.Kondev, S.Naimi and X.Xu](http://nuclearmasses.org/resources_folder/Wang_2017_Chinese_Phys_C_41_030003.pdf)\n", + "there are data on masses and decays of 3437 nuclei.\n", + "\n", + "The nuclear binding energy is defined as the energy required to break\n", + "up a given nucleus into its constituent parts of $N$ neutrons and $Z$\n", + "protons. In terms of the atomic masses $M(N, Z)$ the binding energy is\n", + "defined by" + ] + }, + { + "cell_type": "markdown", + "id": "b60eb119", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N, Z) = ZM_H c^2 + Nm_n c^2 - M(N, Z)c^2 ,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b38eb451", + "metadata": { + "editable": true + }, + "source": [ + "where $M_H$ is the mass of the hydrogen atom and $m_n$ is the mass of the neutron.\n", + "In terms of the mass excess the binding energy is given by" + ] + }, + { + "cell_type": "markdown", + "id": "d67cf525", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N, Z) = Z\\Delta_H c^2 + N\\Delta_n c^2 -\\Delta(N, Z)c^2 ,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2f8449b2", + "metadata": { + "editable": true + }, + "source": [ + "where $\\Delta_H c^2 = 7.2890$ MeV and $\\Delta_n c^2 = 8.0713$ MeV.\n", + "\n", + "A popular and physically intuitive model which can be used to parametrize \n", + "the experimental binding energies as function of $A$, is the so-called \n", + "**liquid drop model**. The ansatz is based on the following expression" + ] + }, + { + "cell_type": "markdown", + "id": "55ca6490", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(N,Z) = a_1A-a_2A^{2/3}-a_3\\frac{Z^2}{A^{1/3}}-a_4\\frac{(N-Z)^2}{A},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e716655b", + "metadata": { + "editable": true + }, + "source": [ + "where $A$ stands for the number of nucleons and the $a_i$s are parameters which are determined by a fit \n", + "to the experimental data. \n", + "\n", + "To arrive at the above expression we have assumed that we can make the following assumptions:\n", + "\n", + " * There is a volume term $a_1A$ proportional with the number of nucleons (the energy is also an extensive quantity). When an assembly of nucleons of the same size is packed together into the smallest volume, each interior nucleon has a certain number of other nucleons in contact with it. This contribution is proportional to the volume.\n", + "\n", + " * There is a surface energy term $a_2A^{2/3}$. The assumption here is that a nucleon at the surface of a nucleus interacts with fewer other nucleons than one in the interior of the nucleus and hence its binding energy is less. This surface energy term takes that into account and is therefore negative and is proportional to the surface area.\n", + "\n", + " * There is a Coulomb energy term $a_3\\frac{Z^2}{A^{1/3}}$. The electric repulsion between each pair of protons in a nucleus yields less binding. \n", + "\n", + " * There is an asymmetry term $a_4\\frac{(N-Z)^2}{A}$. This term is associated with the Pauli exclusion principle and reflects the fact that the proton-neutron interaction is more attractive on the average than the neutron-neutron and proton-proton interactions.\n", + "\n", + "We could also add a so-called pairing term, which is a correction term that\n", + "arises from the tendency of proton pairs and neutron pairs to\n", + "occur. An even number of particles is more stable than an odd number." + ] + }, + { + "cell_type": "markdown", + "id": "fa63a9de", + "metadata": { + "editable": true + }, + "source": [ + "### Organizing our data\n", + "\n", + "Let us start with reading and organizing our data. \n", + "We start with the compilation of masses and binding energies from 2016.\n", + "After having downloaded this file to our own computer, we are now ready to read the file and start structuring our data.\n", + "\n", + "We start with preparing folders for storing our calculations and the data file over masses and binding energies. We import also various modules that we will find useful in order to present various Machine Learning methods. Here we focus mainly on the functionality of **scikit-learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 27, + "id": "cca448dd", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "import sklearn.linear_model as skl\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"MassEval2016.dat\"),'r')" + ] + }, + { + "cell_type": "markdown", + "id": "aa35aba5", + "metadata": { + "editable": true + }, + "source": [ + "Before we proceed, we define also a function for making our plots. You can obviously avoid this and simply set up various **matplotlib** commands every time you need them. You may however find it convenient to collect all such commands in one function and simply call this function." + ] + }, + { + "cell_type": "code", + "execution_count": 28, + "id": "c854893a", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "from pylab import plt, mpl\n", + "plt.style.use('seaborn')\n", + "mpl.rcParams['font.family'] = 'serif'\n", + "\n", + "def MakePlot(x,y, styles, labels, axlabels):\n", + " plt.figure(figsize=(10,6))\n", + " for i in range(len(x)):\n", + " plt.plot(x[i], y[i], styles[i], label = labels[i])\n", + " plt.xlabel(axlabels[0])\n", + " plt.ylabel(axlabels[1])\n", + " plt.legend(loc=0)" + ] + }, + { + "cell_type": "markdown", + "id": "74b3f235", + "metadata": { + "editable": true + }, + "source": [ + "Our next step is to read the data on experimental binding energies and\n", + "reorganize them as functions of the mass number $A$, the number of\n", + "protons $Z$ and neutrons $N$ using **pandas**. Before we do this it is\n", + "always useful (unless you have a binary file or other types of compressed\n", + "data) to actually open the file and simply take a look at it!\n", + "\n", + "In particular, the program that outputs the final nuclear masses is written in Fortran with a specific format. It means that we need to figure out the format and which columns contain the data we are interested in. Pandas comes with a function that reads formatted output. After having admired the file, we are now ready to start massaging it with **pandas**. The file begins with some basic format information." + ] + }, + { + "cell_type": "code", + "execution_count": 29, + "id": "92e8847b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "data": { + "text/plain": [ + "' \\nThis is taken from the data file of the mass 2016 evaluation. \\nAll files are 3436 lines long with 124 character per line. \\n Headers are 39 lines long. \\n col 1 : Fortran character control: 1 = page feed 0 = line feed \\n format : a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5 \\n These formats are reflected in the pandas widths variable below, see the statement \\n widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), \\n Pandas has also a variable header, with length 39 in this case. \\n'" + ] + }, + "execution_count": 29, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "\"\"\" \n", + "This is taken from the data file of the mass 2016 evaluation. \n", + "All files are 3436 lines long with 124 character per line. \n", + " Headers are 39 lines long. \n", + " col 1 : Fortran character control: 1 = page feed 0 = line feed \n", + " format : a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5 \n", + " These formats are reflected in the pandas widths variable below, see the statement \n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), \n", + " Pandas has also a variable header, with length 39 in this case. \n", + "\"\"\"" + ] + }, + { + "cell_type": "markdown", + "id": "8cbce288", + "metadata": { + "editable": true + }, + "source": [ + "The data we are interested in are in columns 2, 3, 4 and 11, giving us\n", + "the number of neutrons, protons, mass numbers and binding energies,\n", + "respectively. We add also for the sake of completeness the element name. The data are in fixed-width formatted lines and we will\n", + "covert them into the **pandas** DataFrame structure." + ] + }, + { + "cell_type": "code", + "execution_count": 30, + "id": "b4398581", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [ + { + "ename": "ValueError", + "evalue": "Length of colspecs must match length of names", + "output_type": "error", + "traceback": [ + "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", + "\u001b[0;31mValueError\u001b[0m Traceback (most recent call last)", + "Input \u001b[0;32mIn [30]\u001b[0m, in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[38;5;66;03m# Read the experimental data with Pandas\u001b[39;00m\n\u001b[0;32m----> 2\u001b[0m Masses \u001b[38;5;241m=\u001b[39m \u001b[43mpd\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mread_fwf\u001b[49m\u001b[43m(\u001b[49m\u001b[43minfile\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43musecols\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m2\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m3\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m4\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m6\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m11\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 3\u001b[0m \u001b[43m \u001b[49m\u001b[43mnames\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mN\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mZ\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mA\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mElement\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43mEbinding\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 4\u001b[0m \u001b[43m \u001b[49m\u001b[43mwidths\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m3\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m5\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m5\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m5\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m3\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m4\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m13\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m11\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m11\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m9\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m2\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m11\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m9\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m3\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m12\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m11\u001b[39;49m\u001b[43m,\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 5\u001b[0m \u001b[43m \u001b[49m\u001b[43mheader\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m39\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[1;32m 6\u001b[0m \u001b[43m \u001b[49m\u001b[43mindex_col\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mFalse\u001b[39;49;00m\u001b[43m)\u001b[49m\n\u001b[1;32m 8\u001b[0m \u001b[38;5;66;03m# Extrapolated values are indicated by '#' in place of the decimal place, so\u001b[39;00m\n\u001b[1;32m 9\u001b[0m \u001b[38;5;66;03m# the Ebinding column won't be numeric. Coerce to float and drop these entries.\u001b[39;00m\n\u001b[1;32m 10\u001b[0m Masses[\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mEbinding\u001b[39m\u001b[38;5;124m'\u001b[39m] \u001b[38;5;241m=\u001b[39m pd\u001b[38;5;241m.\u001b[39mto_numeric(Masses[\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mEbinding\u001b[39m\u001b[38;5;124m'\u001b[39m], errors\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mcoerce\u001b[39m\u001b[38;5;124m'\u001b[39m)\n", + "File \u001b[0;32m~/miniforge3/envs/myenv/lib/python3.9/site-packages/pandas/util/_decorators.py:311\u001b[0m, in \u001b[0;36mdeprecate_nonkeyword_arguments..decorate..wrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 305\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mlen\u001b[39m(args) \u001b[38;5;241m>\u001b[39m num_allow_args:\n\u001b[1;32m 306\u001b[0m warnings\u001b[38;5;241m.\u001b[39mwarn(\n\u001b[1;32m 307\u001b[0m msg\u001b[38;5;241m.\u001b[39mformat(arguments\u001b[38;5;241m=\u001b[39marguments),\n\u001b[1;32m 308\u001b[0m \u001b[38;5;167;01mFutureWarning\u001b[39;00m,\n\u001b[1;32m 309\u001b[0m stacklevel\u001b[38;5;241m=\u001b[39mstacklevel,\n\u001b[1;32m 310\u001b[0m )\n\u001b[0;32m--> 311\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n", + "File \u001b[0;32m~/miniforge3/envs/myenv/lib/python3.9/site-packages/pandas/io/parsers/readers.py:871\u001b[0m, in \u001b[0;36mread_fwf\u001b[0;34m(filepath_or_buffer, colspecs, widths, infer_nrows, **kwds)\u001b[0m\n\u001b[1;32m 869\u001b[0m len_index \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlen\u001b[39m(index_col)\n\u001b[1;32m 870\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mlen\u001b[39m(names) \u001b[38;5;241m+\u001b[39m len_index \u001b[38;5;241m!=\u001b[39m \u001b[38;5;28mlen\u001b[39m(colspecs):\n\u001b[0;32m--> 871\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mLength of colspecs must match length of names\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 873\u001b[0m kwds[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcolspecs\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m colspecs\n\u001b[1;32m 874\u001b[0m kwds[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124minfer_nrows\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m infer_nrows\n", + "\u001b[0;31mValueError\u001b[0m: Length of colspecs must match length of names" + ] + } + ], + "source": [ + "# Read the experimental data with Pandas\n", + "Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),\n", + " names=('N', 'Z', 'A', 'Element', 'Ebinding'),\n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),\n", + " header=39,\n", + " index_col=False)\n", + "\n", + "# Extrapolated values are indicated by '#' in place of the decimal place, so\n", + "# the Ebinding column won't be numeric. Coerce to float and drop these entries.\n", + "Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')\n", + "Masses = Masses.dropna()\n", + "# Convert from keV to MeV.\n", + "Masses['Ebinding'] /= 1000\n", + "\n", + "# Group the DataFrame by nucleon number, A.\n", + "Masses = Masses.groupby('A')\n", + "# Find the rows of the grouped DataFrame with the maximum binding energy.\n", + "Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])" + ] + }, + { + "cell_type": "markdown", + "id": "0f21a73f", + "metadata": { + "editable": true + }, + "source": [ + "We have now read in the data, grouped them according to the variables we are interested in. \n", + "We see how easy it is to reorganize the data using **pandas**. If we\n", + "were to do these operations in C/C++ or Fortran, we would have had to\n", + "write various functions/subroutines which perform the above\n", + "reorganizations for us. Having reorganized the data, we can now start\n", + "to make some simple fits using both the functionalities in **numpy** and\n", + "**Scikit-Learn** afterwards. \n", + "\n", + "Now we define five variables which contain\n", + "the number of nucleons $A$, the number of protons $Z$ and the number of neutrons $N$, the element name and finally the energies themselves." + ] + }, + { + "cell_type": "code", + "execution_count": 31, + "id": "491e38a4", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "A = Masses['A']\n", + "Z = Masses['Z']\n", + "N = Masses['N']\n", + "Element = Masses['Element']\n", + "Energies = Masses['Ebinding']\n", + "print(Masses)" + ] + }, + { + "cell_type": "markdown", + "id": "07ac2cd3", + "metadata": { + "editable": true + }, + "source": [ + "The next step, and we will define this mathematically later, is to set up the so-called **design matrix**. We will throughout call this matrix $\\boldsymbol{X}$.\n", + "It has dimensionality $p\\times n$, where $n$ is the number of data points and $p$ are the so-called predictors. In our case here they are given by the number of polynomials in $A$ we wish to include in the fit." + ] + }, + { + "cell_type": "code", + "execution_count": 32, + "id": "0462ebdd", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Now we set up the design matrix X\n", + "X = np.zeros((len(A),5))\n", + "X[:,0] = 1\n", + "X[:,1] = A\n", + "X[:,2] = A**(2.0/3.0)\n", + "X[:,3] = A**(-1.0/3.0)\n", + "X[:,4] = A**(-1.0)" + ] + }, + { + "cell_type": "markdown", + "id": "66a31f58", + "metadata": { + "editable": true + }, + "source": [ + "With **scikitlearn** we are now ready to use linear regression and fit our data." + ] + }, + { + "cell_type": "code", + "execution_count": 33, + "id": "7d4bf033", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "clf = skl.LinearRegression().fit(X, Energies)\n", + "fity = clf.predict(X)" + ] + }, + { + "cell_type": "markdown", + "id": "3b9f86de", + "metadata": { + "editable": true + }, + "source": [ + "Pretty simple! \n", + "Now we can print measures of how our fit is doing, the coefficients from the fits and plot the final fit together with our data." + ] + }, + { + "cell_type": "code", + "execution_count": 34, + "id": "b12b4d19", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, fity))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, fity))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, fity))\n", + "print(clf.coef_, clf.intercept_)\n", + "\n", + "Masses['Eapprox'] = fity\n", + "# Generate a plot comparing the experimental with the fitted values values.\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$A = N + Z$')\n", + "ax.set_ylabel(r'$E_\\mathrm{bind}\\,/\\mathrm{MeV}$')\n", + "ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,\n", + " label='Ame2016')\n", + "ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',\n", + " label='Fit')\n", + "ax.legend()\n", + "save_fig(\"Masses2016\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "4cf1be8b", + "metadata": { + "editable": true + }, + "source": [ + "### And what about using neural networks?\n", + "\n", + "The **seaborn** package allows us to visualize data in an efficient way. Note that we use **scikit-learn**'s multi-layer perceptron (or feed forward neural network) \n", + "functionality." + ] + }, + { + "cell_type": "code", + "execution_count": 35, + "id": "e80600eb", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "from sklearn.neural_network import MLPRegressor\n", + "from sklearn.metrics import accuracy_score\n", + "import seaborn as sns\n", + "\n", + "X_train = X\n", + "Y_train = Energies\n", + "n_hidden_neurons = 100\n", + "epochs = 100\n", + "# store models for later use\n", + "eta_vals = np.logspace(-5, 1, 7)\n", + "lmbd_vals = np.logspace(-5, 1, 7)\n", + "# store the models for later use\n", + "DNN_scikit = np.zeros((len(eta_vals), len(lmbd_vals)), dtype=object)\n", + "train_accuracy = np.zeros((len(eta_vals), len(lmbd_vals)))\n", + "sns.set()\n", + "for i, eta in enumerate(eta_vals):\n", + " for j, lmbd in enumerate(lmbd_vals):\n", + " dnn = MLPRegressor(hidden_layer_sizes=(n_hidden_neurons), activation='logistic',\n", + " alpha=lmbd, learning_rate_init=eta, max_iter=epochs)\n", + " dnn.fit(X_train, Y_train)\n", + " DNN_scikit[i][j] = dnn\n", + " train_accuracy[i][j] = dnn.score(X_train, Y_train)\n", + "\n", + "fig, ax = plt.subplots(figsize = (10, 10))\n", + "sns.heatmap(train_accuracy, annot=True, ax=ax, cmap=\"viridis\")\n", + "ax.set_title(\"Training Accuracy\")\n", + "ax.set_ylabel(\"$\\eta$\")\n", + "ax.set_xlabel(\"$\\lambda$\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "d49cd4ec", + "metadata": { + "editable": true + }, + "source": [ + "## A first summary\n", + "\n", + "The aim behind these introductory words was to present to you various\n", + "Python libraries and their functionalities, in particular libraries like\n", + "**numpy**, **pandas**, **xarray** and **matplotlib** and other that make our life much easier\n", + "in handling various data sets and visualizing data. \n", + "\n", + "Furthermore,\n", + "**Scikit-Learn** allows us with few lines of code to implement popular\n", + "Machine Learning algorithms for supervised learning. Later we will meet **Tensorflow**, a powerful library for deep learning. \n", + "Now it is time to dive more into the details of various methods. We will start with linear regression and try to take a deeper look at what it entails." + ] + }, + { + "cell_type": "markdown", + "id": "9a4706c7", + "metadata": { + "editable": true + }, + "source": [ + "## Why Linear Regression (aka Ordinary Least Squares and family)\n", + "\n", + "Fitting a continuous function with linear parameterization in terms of the parameters $\\boldsymbol{\\beta}$.\n", + "* Method of choice for fitting a continuous function!\n", + "\n", + "* Gives an excellent introduction to central Machine Learning features with **understandable pedagogical** links to other methods like **Neural Networks**, **Support Vector Machines** etc\n", + "\n", + "* Analytical expression for the fitting parameters $\\boldsymbol{\\beta}$\n", + "\n", + "* Analytical expressions for statistical propertiers like mean values, variances, confidence intervals and more\n", + "\n", + "* Analytical relation with probabilistic interpretations \n", + "\n", + "* Easy to introduce basic concepts like bias-variance tradeoff, cross-validation, resampling and regularization techniques and many other ML topics\n", + "\n", + "* Easy to code! And links well with classification problems and logistic regression and neural networks\n", + "\n", + "* Allows for **easy** hands-on understanding of gradient descent methods\n", + "\n", + "* and many more features\n", + "\n", + "For more discussions of Ridge and Lasso regression, [Wessel van Wieringen's](https://arxiv.org/abs/1509.09169) article is highly recommended.\n", + "Similarly, [Mehta et al's article](https://arxiv.org/abs/1803.08823) is also recommended." + ] + }, + { + "cell_type": "markdown", + "id": "b720efdc", + "metadata": { + "editable": true + }, + "source": [ + "## Regression analysis, overarching aims\n", + "\n", + "Regression modeling deals with the description of the sampling distribution of a given random variable $y$ and how it varies as function of another variable or a set of such variables $\\boldsymbol{x} =[x_0, x_1,\\dots, x_{n-1}]^T$. \n", + "The first variable is called the **dependent**, the **outcome** or the **response** variable while the set of variables $\\boldsymbol{x}$ is called the independent variable, or the predictor variable or the explanatory variable, or simply just the **inputs**. \n", + "\n", + "A regression model aims at finding a likelihood function $p(\\boldsymbol{y}\\vert \\boldsymbol{x})$ or in the more traditional sense a function $\\boldsymbol{y}(\\boldsymbol{x})$, that is the conditional distribution for $\\boldsymbol{y}$ with a given $\\boldsymbol{x}$. The estimation of $p(\\boldsymbol{y}\\vert \\boldsymbol{x})$ is made using a data set with \n", + "* $n$ cases $i = 0, 1, 2, \\dots, n-1$ \n", + "\n", + "* Response (target, dependent or outcome) variable $y_i$ with $i = 0, 1, 2, \\dots, n-1$ \n", + "\n", + "* $p$ so-called explanatory (independent or predictor or feature) variables $\\boldsymbol{x}_i=[x_{i0}, x_{i1}, \\dots, x_{ip-1}]$ with $i = 0, 1, 2, \\dots, n-1$ and explanatory variables running from $0$ to $p-1$. See below for more explicit examples. \n", + "\n", + " The goal of the regression analysis is to extract/exploit relationship between $\\boldsymbol{y}$ and $\\boldsymbol{x}$ in order to infer specific dependencies, approximations to the likelihood functions, functional relationships and to make predictions, making fits and many other things." + ] + }, + { + "cell_type": "markdown", + "id": "0cb90bb5", + "metadata": { + "editable": true + }, + "source": [ + "## Regression analysis, overarching aims II\n", + "\n", + "Consider an experiment in which $p$ characteristics/features of $n$ samples are\n", + "measured. The data from this experiment, for various explanatory variables $p$ are normally represented by a matrix \n", + "$\\mathbf{X}$.\n", + "\n", + "The matrix $\\mathbf{X}$ is called the *design\n", + "matrix*. Additional information of the samples is available in the\n", + "form of $\\boldsymbol{y}$ (also as above). The variable $\\boldsymbol{y}$ is\n", + "generally referred to as the *response variable*. The aim of\n", + "regression analysis is to explain $\\boldsymbol{y}$ in terms of\n", + "$\\boldsymbol{X}$ through a functional relationship like $y_i =\n", + "f(\\mathbf{X}_{i,\\ast})$. When no prior knowledge on the form of\n", + "$f(\\cdot)$ is available, it is common to assume a linear relationship\n", + "between $\\boldsymbol{X}$ and $\\boldsymbol{y}$. This assumption gives rise to\n", + "the *linear regression model* where $\\boldsymbol{\\beta} = [\\beta_0, \\ldots,\n", + "\\beta_{p-1}]^{T}$ are the *regression parameters*. \n", + "\n", + "Linear regression gives us a set of analytical equations for the parameters $\\beta_j$." + ] + }, + { + "cell_type": "markdown", + "id": "e45e89d3", + "metadata": { + "editable": true + }, + "source": [ + "## Examples\n", + "In order to understand the relation among the predictors (or features or properties) $p$, the set of data $n$ and the target (outcome, output etc) $\\boldsymbol{y}$,\n", + "consider the model we discussed for describing nuclear binding energies. \n", + "\n", + "There we assumed that we could parametrize the data using a polynomial approximation based on the liquid drop model.\n", + "Assuming" + ] + }, + { + "cell_type": "markdown", + "id": "f2471451", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "BE(A) = a_0+a_1A+a_2A^{2/3}+a_3A^{-1/3}+a_4A^{-1},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ee20b9ee", + "metadata": { + "editable": true + }, + "source": [ + "we have five predictors, that is the intercept, the $A$ dependent term, the $A^{2/3}$ term and the $A^{-1/3}$ and $A^{-1}$ terms.\n", + "This gives $p=0,1,2,3,4$. Furthermore we have $n$ entries for each predictor. It means that our design matrix is a \n", + "$p\\times n$ matrix $\\boldsymbol{X}$.\n", + "\n", + "Here the predictors are based on a model we have made. A popular data set which is widely encountered in ML applications is the\n", + "so-called [credit card default data from Taiwan](https://www.sciencedirect.com/science/article/pii/S0957417407006719?via%3Dihub). The data set contains data on $n=30000$ credit card holders with predictors like gender, marital status, age, profession, education, etc. In total there are $24$ such predictors or attributes leading to a design matrix of dimensionality $24 \\times 30000$. This is however a classification problem and we will come back to it when we discuss Logistic Regression." + ] + }, + { + "cell_type": "markdown", + "id": "ebc12338", + "metadata": { + "editable": true + }, + "source": [ + "## General linear models and linear algebra\n", + "Before we proceed let us study a case where we aim at fitting a set of data $\\boldsymbol{y}=[y_0,y_1,\\dots,y_{n-1}]$. We could think of these data as a result of an experiment or a complicated numerical experiment. These data are functions of a series of variables $\\boldsymbol{x}=[x_0,x_1,\\dots,x_{n-1}]$, that is $y_i = y(x_i)$ with $i=0,1,2,\\dots,n-1$. The variables $x_i$ could represent physical quantities like time, temperature, position etc. We assume that $y(x)$ is a smooth function. \n", + "\n", + "Since obtaining these data points may not be trivial, we want to use these data to fit a function which can allow us to make predictions for values of $y$ which are not in the present set. The perhaps simplest approach is to assume we can parametrize our function in terms of a polynomial of degree $n-1$ with $n$ points, that is" + ] + }, + { + "cell_type": "markdown", + "id": "0f5cb3d7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y=y(x) \\rightarrow y(x_i)=\\tilde{y}_i+\\epsilon_i=\\sum_{j=0}^{n-1} \\beta_j x_i^j+\\epsilon_i,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8be293e3", + "metadata": { + "editable": true + }, + "source": [ + "where $\\epsilon_i$ is the error in our approximation." + ] + }, + { + "cell_type": "markdown", + "id": "c0a2104e", + "metadata": { + "editable": true + }, + "source": [ + "## Rewriting the fitting procedure as a linear algebra problem\n", + "For every set of values $y_i,x_i$ we have thus the corresponding set of equations" + ] + }, + { + "cell_type": "markdown", + "id": "93b447fd", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0+\\beta_1x_0^1+\\beta_2x_0^2+\\dots+\\beta_{n-1}x_0^{n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0+\\beta_1x_1^1+\\beta_2x_1^2+\\dots+\\beta_{n-1}x_1^{n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0+\\beta_1x_2^1+\\beta_2x_2^2+\\dots+\\beta_{n-1}x_2^{n-1}+\\epsilon_2\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0+\\beta_1x_{n-1}^1+\\beta_2x_{n-1}^2+\\dots+\\beta_{n-1}x_{n-1}^{n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "94d59403", + "metadata": { + "editable": true + }, + "source": [ + "## Rewriting the fitting procedure as a linear algebra problem, more details\n", + "Defining the vectors" + ] + }, + { + "cell_type": "markdown", + "id": "7b044c26", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = [y_0,y_1, y_2,\\dots, y_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a4ec8b3e", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "7e221f81", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} = [\\beta_0,\\beta_1, \\beta_2,\\dots, \\beta_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "729c5dd3", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "5eb54f13", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\epsilon} = [\\epsilon_0,\\epsilon_1, \\epsilon_2,\\dots, \\epsilon_{n-1}]^T,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "51cad783", + "metadata": { + "editable": true + }, + "source": [ + "and the design matrix" + ] + }, + { + "cell_type": "markdown", + "id": "e38a17c1", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}=\n", + "\\begin{bmatrix} \n", + "1& x_{0}^1 &x_{0}^2& \\dots & \\dots &x_{0}^{n-1}\\\\\n", + "1& x_{1}^1 &x_{1}^2& \\dots & \\dots &x_{1}^{n-1}\\\\\n", + "1& x_{2}^1 &x_{2}^2& \\dots & \\dots &x_{2}^{n-1}\\\\ \n", + "\\dots& \\dots &\\dots& \\dots & \\dots &\\dots\\\\\n", + "1& x_{n-1}^1 &x_{n-1}^2& \\dots & \\dots &x_{n-1}^{n-1}\\\\\n", + "\\end{bmatrix}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c669704a", + "metadata": { + "editable": true + }, + "source": [ + "we can rewrite our equations as" + ] + }, + { + "cell_type": "markdown", + "id": "6c2f1143", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = \\boldsymbol{X}\\boldsymbol{\\beta}+\\boldsymbol{\\epsilon}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "144ef4c2", + "metadata": { + "editable": true + }, + "source": [ + "The above design matrix is called a [Vandermonde matrix](https://en.wikipedia.org/wiki/Vandermonde_matrix)." + ] + }, + { + "cell_type": "markdown", + "id": "d9da387d", + "metadata": { + "editable": true + }, + "source": [ + "## Generalizing the fitting procedure as a linear algebra problem\n", + "\n", + "We are obviously not limited to the above polynomial expansions. We\n", + "could replace the various powers of $x$ with elements of Fourier\n", + "series or instead of $x_i^j$ we could have $\\cos{(j x_i)}$ or $\\sin{(j\n", + "x_i)}$, or time series or other orthogonal functions. For every set\n", + "of values $y_i,x_i$ we can then generalize the equations to" + ] + }, + { + "cell_type": "markdown", + "id": "16d9099a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0x_{00}+\\beta_1x_{01}+\\beta_2x_{02}+\\dots+\\beta_{n-1}x_{0n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0x_{10}+\\beta_1x_{11}+\\beta_2x_{12}+\\dots+\\beta_{n-1}x_{1n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0x_{20}+\\beta_1x_{21}+\\beta_2x_{22}+\\dots+\\beta_{n-1}x_{2n-1}+\\epsilon_2\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{i}&=\\beta_0x_{i0}+\\beta_1x_{i1}+\\beta_2x_{i2}+\\dots+\\beta_{n-1}x_{in-1}+\\epsilon_i\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0x_{n-1,0}+\\beta_1x_{n-1,2}+\\beta_2x_{n-1,2}+\\dots+\\beta_{n-1}x_{n-1,n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "dac4b0d4", + "metadata": { + "editable": true + }, + "source": [ + "**Note that we have $p=n$ here. The matrix is symmetric. This is generally not the case!**" + ] + }, + { + "cell_type": "markdown", + "id": "821d16ee", + "metadata": { + "editable": true + }, + "source": [ + "## Generalizing the fitting procedure as a linear algebra problem\n", + "We redefine in turn the matrix $\\boldsymbol{X}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "64804e58", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}=\n", + "\\begin{bmatrix} \n", + "x_{00}& x_{01} &x_{02}& \\dots & \\dots &x_{0,n-1}\\\\\n", + "x_{10}& x_{11} &x_{12}& \\dots & \\dots &x_{1,n-1}\\\\\n", + "x_{20}& x_{21} &x_{22}& \\dots & \\dots &x_{2,n-1}\\\\ \n", + "\\dots& \\dots &\\dots& \\dots & \\dots &\\dots\\\\\n", + "x_{n-1,0}& x_{n-1,1} &x_{n-1,2}& \\dots & \\dots &x_{n-1,n-1}\\\\\n", + "\\end{bmatrix}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "28be47df", + "metadata": { + "editable": true + }, + "source": [ + "and without loss of generality we rewrite again our equations as" + ] + }, + { + "cell_type": "markdown", + "id": "f15f39a0", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{y} = \\boldsymbol{X}\\boldsymbol{\\beta}+\\boldsymbol{\\epsilon}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "03117774", + "metadata": { + "editable": true + }, + "source": [ + "The left-hand side of this equation is kwown. Our error vector $\\boldsymbol{\\epsilon}$ and the parameter vector $\\boldsymbol{\\beta}$ are our unknow quantities. How can we obtain the optimal set of $\\beta_i$ values?" + ] + }, + { + "cell_type": "markdown", + "id": "2dadc99e", + "metadata": { + "editable": true + }, + "source": [ + "## Optimizing our parameters\n", + "We have defined the matrix $\\boldsymbol{X}$ via the equations" + ] + }, + { + "cell_type": "markdown", + "id": "98d081e3", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\begin{align*}\n", + "y_0&=\\beta_0x_{00}+\\beta_1x_{01}+\\beta_2x_{02}+\\dots+\\beta_{n-1}x_{0n-1}+\\epsilon_0\\\\\n", + "y_1&=\\beta_0x_{10}+\\beta_1x_{11}+\\beta_2x_{12}+\\dots+\\beta_{n-1}x_{1n-1}+\\epsilon_1\\\\\n", + "y_2&=\\beta_0x_{20}+\\beta_1x_{21}+\\beta_2x_{22}+\\dots+\\beta_{n-1}x_{2n-1}+\\epsilon_1\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{i}&=\\beta_0x_{i0}+\\beta_1x_{i1}+\\beta_2x_{i2}+\\dots+\\beta_{n-1}x_{in-1}+\\epsilon_1\\\\\n", + "\\dots & \\dots \\\\\n", + "y_{n-1}&=\\beta_0x_{n-1,0}+\\beta_1x_{n-1,2}+\\beta_2x_{n-1,2}+\\dots+\\beta_{n-1}x_{n-1,n-1}+\\epsilon_{n-1}.\\\\\n", + "\\end{align*}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "02c207e3", + "metadata": { + "editable": true + }, + "source": [ + "As we noted above, we stayed with a system with the design matrix \n", + " $\\boldsymbol{X}\\in {\\mathbb{R}}^{n\\times n}$, that is we have $p=n$. For reasons to come later (algorithmic arguments) we will hereafter define \n", + "our matrix as $\\boldsymbol{X}\\in {\\mathbb{R}}^{n\\times p}$, with the predictors refering to the column numbers and the entries $n$ being the row elements." + ] + }, + { + "cell_type": "markdown", + "id": "5b3ba5aa", + "metadata": { + "editable": true + }, + "source": [ + "## Our model for the nuclear binding energies\n", + "\n", + "In our [introductory notes](https://compphysics.github.io/MachineLearning/doc/pub/How2ReadData/html/How2ReadData.html) we looked at the so-called [liquid drop model](https://en.wikipedia.org/wiki/Semi-empirical_mass_formula). Let us remind ourselves about what we did by looking at the code.\n", + "\n", + "We restate the parts of the code we are most interested in." + ] + }, + { + "cell_type": "code", + "execution_count": 36, + "id": "97a71b30", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "from IPython.display import display\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"MassEval2016.dat\"),'r')\n", + "\n", + "\n", + "# Read the experimental data with Pandas\n", + "Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11),\n", + " names=('N', 'Z', 'A', 'Element', 'Ebinding'),\n", + " widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1),\n", + " header=39,\n", + " index_col=False)\n", + "\n", + "# Extrapolated values are indicated by '#' in place of the decimal place, so\n", + "# the Ebinding column won't be numeric. Coerce to float and drop these entries.\n", + "Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce')\n", + "Masses = Masses.dropna()\n", + "# Convert from keV to MeV.\n", + "Masses['Ebinding'] /= 1000\n", + "\n", + "# Group the DataFrame by nucleon number, A.\n", + "Masses = Masses.groupby('A')\n", + "# Find the rows of the grouped DataFrame with the maximum binding energy.\n", + "Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()])\n", + "A = Masses['A']\n", + "Z = Masses['Z']\n", + "N = Masses['N']\n", + "Element = Masses['Element']\n", + "Energies = Masses['Ebinding']\n", + "\n", + "# Now we set up the design matrix X\n", + "X = np.zeros((len(A),5))\n", + "X[:,0] = 1\n", + "X[:,1] = A\n", + "X[:,2] = A**(2.0/3.0)\n", + "X[:,3] = A**(-1.0/3.0)\n", + "X[:,4] = A**(-1.0)\n", + "# Then nice printout using pandas\n", + "DesignMatrix = pd.DataFrame(X)\n", + "DesignMatrix.index = A\n", + "DesignMatrix.columns = ['1', 'A', 'A^(2/3)', 'A^(-1/3)', '1/A']\n", + "display(DesignMatrix)" + ] + }, + { + "cell_type": "markdown", + "id": "ed7eb3b8", + "metadata": { + "editable": true + }, + "source": [ + "With $\\boldsymbol{\\beta}\\in {\\mathbb{R}}^{p\\times 1}$, it means that we will hereafter write our equations for the approximation as" + ] + }, + { + "cell_type": "markdown", + "id": "1ac7d0e5", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\tilde{y}}= \\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "8069a90f", + "metadata": { + "editable": true + }, + "source": [ + "throughout these lectures." + ] + }, + { + "cell_type": "markdown", + "id": "815493e0", + "metadata": { + "editable": true + }, + "source": [ + "## Optimizing our parameters, more details\n", + "With the above we use the design matrix to define the approximation $\\boldsymbol{\\tilde{y}}$ via the unknown quantity $\\boldsymbol{\\beta}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "7880ef32", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\tilde{y}}= \\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ecb0a16f", + "metadata": { + "editable": true + }, + "source": [ + "and in order to find the optimal parameters $\\beta_i$ instead of solving the above linear algebra problem, we define a function which gives a measure of the spread between the values $y_i$ (which represent hopefully the exact values) and the parameterized values $\\tilde{y}_i$, namely" + ] + }, + { + "cell_type": "markdown", + "id": "d8782aad", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\sum_{i=0}^{n-1}\\left(y_i-\\tilde{y}_i\\right)^2=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "d2491d40", + "metadata": { + "editable": true + }, + "source": [ + "or using the matrix $\\boldsymbol{X}$ and in a more compact matrix-vector notation as" + ] + }, + { + "cell_type": "markdown", + "id": "1cf22842", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "71180218", + "metadata": { + "editable": true + }, + "source": [ + "This function is one possible way to define the so-called cost function.\n", + "\n", + "It is also common to define\n", + "the function $C$ as" + ] + }, + { + "cell_type": "markdown", + "id": "1c734def", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{2n}\\sum_{i=0}^{n-1}\\left(y_i-\\tilde{y}_i\\right)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "85f6ff9d", + "metadata": { + "editable": true + }, + "source": [ + "since when taking the first derivative with respect to the unknown parameters $\\beta$, the factor of $2$ cancels out." + ] + }, + { + "cell_type": "markdown", + "id": "1ad66e97", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "\n", + "The function" + ] + }, + { + "cell_type": "markdown", + "id": "6ff8d32a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "C(\\boldsymbol{\\beta})=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "fde1cca5", + "metadata": { + "editable": true + }, + "source": [ + "can be linked to the variance of the quantity $y_i$ if we interpret the latter as the mean value. \n", + "When linking (see the discussion below) with the maximum likelihood approach below, we will indeed interpret $y_i$ as a mean value" + ] + }, + { + "cell_type": "markdown", + "id": "5b34fa19", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y_{i}=\\langle y_i \\rangle = \\beta_0x_{i,0}+\\beta_1x_{i,1}+\\beta_2x_{i,2}+\\dots+\\beta_{n-1}x_{i,n-1}+\\epsilon_i,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e93804b1", + "metadata": { + "editable": true + }, + "source": [ + "where $\\langle y_i \\rangle$ is the mean value. Keep in mind also that\n", + "till now we have treated $y_i$ as the exact value. Normally, the\n", + "response (dependent or outcome) variable $y_i$ the outcome of a\n", + "numerical experiment or another type of experiment and is thus only an\n", + "approximation to the true value. It is then always accompanied by an\n", + "error estimate, often limited to a statistical error estimate given by\n", + "the standard deviation discussed earlier. In the discussion here we\n", + "will treat $y_i$ as our exact value for the response variable.\n", + "\n", + "In order to find the parameters $\\beta_i$ we will then minimize the spread of $C(\\boldsymbol{\\beta})$, that is we are going to solve the problem" + ] + }, + { + "cell_type": "markdown", + "id": "e3b2a85a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "{\\displaystyle \\min_{\\boldsymbol{\\beta}\\in\n", + "{\\mathbb{R}}^{p}}}\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)^T\\left(\\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)\\right\\}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "88b484a0", + "metadata": { + "editable": true + }, + "source": [ + "In practical terms it means we will require" + ] + }, + { + "cell_type": "markdown", + "id": "0cae40f7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\beta_j} = \\frac{\\partial }{\\partial \\beta_j}\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}\\right)^2\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3e3c19f7", + "metadata": { + "editable": true + }, + "source": [ + "which results in" + ] + }, + { + "cell_type": "markdown", + "id": "c5d9738a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\beta_j} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}x_{ij}\\left(y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "49bf03c8", + "metadata": { + "editable": true + }, + "source": [ + "or in a matrix-vector form as" + ] + }, + { + "cell_type": "markdown", + "id": "acc52e20", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ecfc4d77", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "We can rewrite" + ] + }, + { + "cell_type": "markdown", + "id": "7bc2250b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial C(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3c038483", + "metadata": { + "editable": true + }, + "source": [ + "as" + ] + }, + { + "cell_type": "markdown", + "id": "d530cc8f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\boldsymbol{y} = \\boldsymbol{X}^T\\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "667c132f", + "metadata": { + "editable": true + }, + "source": [ + "and if the matrix $\\boldsymbol{X}^T\\boldsymbol{X}$ is invertible we have the solution" + ] + }, + { + "cell_type": "markdown", + "id": "1a592608", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} =\\left(\\boldsymbol{X}^T\\boldsymbol{X}\\right)^{-1}\\boldsymbol{X}^T\\boldsymbol{y}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "fe968934", + "metadata": { + "editable": true + }, + "source": [ + "We note also that since our design matrix is defined as $\\boldsymbol{X}\\in\n", + "{\\mathbb{R}}^{n\\times p}$, the product $\\boldsymbol{X}^T\\boldsymbol{X} \\in\n", + "{\\mathbb{R}}^{p\\times p}$. In the above case we have that $p \\ll n$,\n", + "in our case $p=5$ meaning that we end up with inverting a small\n", + "$5\\times 5$ matrix. This is a rather common situation, in many cases we end up with low-dimensional\n", + "matrices to invert. The methods discussed here and for many other\n", + "supervised learning algorithms like classification with logistic\n", + "regression or support vector machines, exhibit dimensionalities which\n", + "allow for the usage of direct linear algebra methods such as **LU** decomposition or **Singular Value Decomposition** (SVD) for finding the inverse of the matrix\n", + "$\\boldsymbol{X}^T\\boldsymbol{X}$.\n", + "\n", + "**Small question**: Do you think the example we have at hand here (the nuclear binding energies) can lead to problems in inverting the matrix $\\boldsymbol{X}^T\\boldsymbol{X}$? What kind of problems can we expect?" + ] + }, + { + "cell_type": "markdown", + "id": "f182ccd2", + "metadata": { + "editable": true + }, + "source": [ + "## Some useful matrix and vector expressions\n", + "\n", + "See the handwritten notes at \n", + "\n", + "These notes will be discussed during one of the lectures." + ] + }, + { + "cell_type": "markdown", + "id": "3e23c6ac", + "metadata": { + "editable": true + }, + "source": [ + "## Interpretations and optimizing our parameters\n", + "The residuals $\\boldsymbol{\\epsilon}$ are in turn given by" + ] + }, + { + "cell_type": "markdown", + "id": "2a10b1fa", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\epsilon} = \\boldsymbol{y}-\\boldsymbol{\\tilde{y}} = \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "753ac3da", + "metadata": { + "editable": true + }, + "source": [ + "and with" + ] + }, + { + "cell_type": "markdown", + "id": "abdc427a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)= 0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a959437a", + "metadata": { + "editable": true + }, + "source": [ + "we have" + ] + }, + { + "cell_type": "markdown", + "id": "764e589a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{X}^T\\boldsymbol{\\epsilon}=\\boldsymbol{X}^T\\left( \\boldsymbol{y}-\\boldsymbol{X}\\boldsymbol{\\beta}\\right)= 0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "00e50636", + "metadata": { + "editable": true + }, + "source": [ + "meaning that the solution for $\\boldsymbol{\\beta}$ is the one which minimizes the residuals. Later we will link this with the maximum likelihood approach.\n", + "\n", + "Let us now return to our nuclear binding energies and simply code the above equations." + ] + }, + { + "cell_type": "markdown", + "id": "7f1fb52c", + "metadata": { + "editable": true + }, + "source": [ + "## Own code for Ordinary Least Squares\n", + "\n", + "It is rather straightforward to implement the matrix inversion and obtain the parameters $\\boldsymbol{\\beta}$. After having defined the matrix $\\boldsymbol{X}$ we simply need to \n", + "write" + ] + }, + { + "cell_type": "code", + "execution_count": 37, + "id": "6e7ef8e3", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# matrix inversion to find beta\n", + "beta = np.linalg.inv(X.T.dot(X)).dot(X.T).dot(Energies)\n", + "# and then make the prediction\n", + "ytilde = X @ beta" + ] + }, + { + "cell_type": "markdown", + "id": "e8de20b3", + "metadata": { + "editable": true + }, + "source": [ + "Alternatively, you can use the least squares functionality in **Numpy** as" + ] + }, + { + "cell_type": "code", + "execution_count": 38, + "id": "890c0e17", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "fit = np.linalg.lstsq(X, Energies, rcond =None)[0]\n", + "ytildenp = np.dot(fit,X.T)" + ] + }, + { + "cell_type": "markdown", + "id": "fa46eacd", + "metadata": { + "editable": true + }, + "source": [ + "And finally we plot our fit with and compare with data" + ] + }, + { + "cell_type": "code", + "execution_count": 39, + "id": "dd7bc08b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "Masses['Eapprox'] = ytilde\n", + "# Generate a plot comparing the experimental with the fitted values values.\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$A = N + Z$')\n", + "ax.set_ylabel(r'$E_\\mathrm{bind}\\,/\\mathrm{MeV}$')\n", + "ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2,\n", + " label='Ame2016')\n", + "ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m',\n", + " label='Fit')\n", + "ax.legend()\n", + "save_fig(\"Masses2016OLS\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "f1dac4cb", + "metadata": { + "editable": true + }, + "source": [ + "## Adding error analysis and training set up\n", + "\n", + "We can easily test our fit by computing the $R2$ score that we discussed in connection with the functionality of **Scikit-Learn** in the introductory slides.\n", + "Since we are not using **Scikit-Learn** here we can define our own $R2$ function as" + ] + }, + { + "cell_type": "code", + "execution_count": 40, + "id": "8253a870", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def R2(y_data, y_model):\n", + " return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)" + ] + }, + { + "cell_type": "markdown", + "id": "25175ebc", + "metadata": { + "editable": true + }, + "source": [ + "and we would be using it as" + ] + }, + { + "cell_type": "code", + "execution_count": 41, + "id": "4106532b", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "print(R2(Energies,ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "3ae3cbe1", + "metadata": { + "editable": true + }, + "source": [ + "We can easily add our **MSE** score as" + ] + }, + { + "cell_type": "code", + "execution_count": 42, + "id": "9cbc337d", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def MSE(y_data,y_model):\n", + " n = np.size(y_model)\n", + " return np.sum((y_data-y_model)**2)/n\n", + "\n", + "print(MSE(Energies,ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "1f984da0", + "metadata": { + "editable": true + }, + "source": [ + "and finally the relative error as" + ] + }, + { + "cell_type": "code", + "execution_count": 43, + "id": "91777701", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "def RelativeError(y_data,y_model):\n", + " return abs((y_data-y_model)/y_data)\n", + "print(RelativeError(Energies, ytilde))" + ] + }, + { + "cell_type": "markdown", + "id": "f504b559", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "Normally, the response (dependent or outcome) variable $y_i$ is the\n", + "outcome of a numerical experiment or another type of experiment and is\n", + "thus only an approximation to the true value. It is then always\n", + "accompanied by an error estimate, often limited to a statistical error\n", + "estimate given by the standard deviation discussed earlier. In the\n", + "discussion here we will treat $y_i$ as our exact value for the\n", + "response variable.\n", + "\n", + "Introducing the standard deviation $\\sigma_i$ for each measurement\n", + "$y_i$, we define now the $\\chi^2$ function (omitting the $1/n$ term)\n", + "as" + ] + }, + { + "cell_type": "markdown", + "id": "67440280", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\chi^2(\\boldsymbol{\\beta})=\\frac{1}{n}\\sum_{i=0}^{n-1}\\frac{\\left(y_i-\\tilde{y}_i\\right)^2}{\\sigma_i^2}=\\frac{1}{n}\\left\\{\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)^T\\frac{1}{\\boldsymbol{\\Sigma^2}}\\left(\\boldsymbol{y}-\\boldsymbol{\\tilde{y}}\\right)\\right\\},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c5190684", + "metadata": { + "editable": true + }, + "source": [ + "where the matrix $\\boldsymbol{\\Sigma}$ is a diagonal matrix with $\\sigma_i$ as matrix elements." + ] + }, + { + "cell_type": "markdown", + "id": "1b878f0e", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "In order to find the parameters $\\beta_i$ we will then minimize the spread of $\\chi^2(\\boldsymbol{\\beta})$ by requiring" + ] + }, + { + "cell_type": "markdown", + "id": "8ab0a4a3", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_j} = \\frac{\\partial }{\\partial \\beta_j}\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(\\frac{y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}}{\\sigma_i}\\right)^2\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e9106c8d", + "metadata": { + "editable": true + }, + "source": [ + "which results in" + ] + }, + { + "cell_type": "markdown", + "id": "03c4e9d7", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_j} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}\\frac{x_{ij}}{\\sigma_i}\\left(\\frac{y_i-\\beta_0x_{i,0}-\\beta_1x_{i,1}-\\beta_2x_{i,2}-\\dots-\\beta_{n-1}x_{i,n-1}}{\\sigma_i}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "0e4abfb1", + "metadata": { + "editable": true + }, + "source": [ + "or in a matrix-vector form as" + ] + }, + { + "cell_type": "markdown", + "id": "45af97a4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{A}^T\\left( \\boldsymbol{b}-\\boldsymbol{A}\\boldsymbol{\\beta}\\right).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "9ff2fda9", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the matrix $\\boldsymbol{A} =\\boldsymbol{X}/\\boldsymbol{\\Sigma}$ with matrix elements $a_{ij} = x_{ij}/\\sigma_i$ and the vector $\\boldsymbol{b}$ with elements $b_i = y_i/\\sigma_i$." + ] + }, + { + "cell_type": "markdown", + "id": "c1593fa5", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "We can rewrite" + ] + }, + { + "cell_type": "markdown", + "id": "d34ea7db", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\boldsymbol{\\beta}} = 0 = \\boldsymbol{A}^T\\left( \\boldsymbol{b}-\\boldsymbol{A}\\boldsymbol{\\beta}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "699612be", + "metadata": { + "editable": true + }, + "source": [ + "as" + ] + }, + { + "cell_type": "markdown", + "id": "88ae1420", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{A}^T\\boldsymbol{b} = \\boldsymbol{A}^T\\boldsymbol{A}\\boldsymbol{\\beta},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b827d1c6", + "metadata": { + "editable": true + }, + "source": [ + "and if the matrix $\\boldsymbol{A}^T\\boldsymbol{A}$ is invertible we have the solution" + ] + }, + { + "cell_type": "markdown", + "id": "61900168", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{\\beta} =\\left(\\boldsymbol{A}^T\\boldsymbol{A}\\right)^{-1}\\boldsymbol{A}^T\\boldsymbol{b}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "eaab94fc", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "If we then introduce the matrix" + ] + }, + { + "cell_type": "markdown", + "id": "c8f542a4", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\boldsymbol{H} = \\left(\\boldsymbol{A}^T\\boldsymbol{A}\\right)^{-1},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c6e93dd5", + "metadata": { + "editable": true + }, + "source": [ + "we have then the following expression for the parameters $\\beta_j$ (the matrix elements of $\\boldsymbol{H}$ are $h_{ij}$)" + ] + }, + { + "cell_type": "markdown", + "id": "977ca552", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_j = \\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}\\frac{y_i}{\\sigma_i}\\frac{x_{ik}}{\\sigma_i} = \\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}b_ia_{ik}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ff3c6b13", + "metadata": { + "editable": true + }, + "source": [ + "We state without proof the expression for the uncertainty in the parameters $\\beta_j$ as (we leave this as an exercise)" + ] + }, + { + "cell_type": "markdown", + "id": "ebbb06cf", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma^2(\\beta_j) = \\sum_{i=0}^{n-1}\\sigma_i^2\\left( \\frac{\\partial \\beta_j}{\\partial y_i}\\right)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "ed6f641e", + "metadata": { + "editable": true + }, + "source": [ + "resulting in" + ] + }, + { + "cell_type": "markdown", + "id": "94c50447", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\sigma^2(\\beta_j) = \\left(\\sum_{k=0}^{p-1}h_{jk}\\sum_{i=0}^{n-1}a_{ik}\\right)\\left(\\sum_{l=0}^{p-1}h_{jl}\\sum_{m=0}^{n-1}a_{ml}\\right) = h_{jj}!\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "caa52732", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "The first step here is to approximate the function $y$ with a first-order polynomial, that is we write" + ] + }, + { + "cell_type": "markdown", + "id": "721f3130", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "y=y(x) \\rightarrow y(x_i) \\approx \\beta_0+\\beta_1 x_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "1ecec855", + "metadata": { + "editable": true + }, + "source": [ + "By computing the derivatives of $\\chi^2$ with respect to $\\beta_0$ and $\\beta_1$ show that these are given by" + ] + }, + { + "cell_type": "markdown", + "id": "9f72bcd2", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_0} = -2\\left[ \\frac{1}{n}\\sum_{i=0}^{n-1}\\left(\\frac{y_i-\\beta_0-\\beta_1x_{i}}{\\sigma_i^2}\\right)\\right]=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "952c0807", + "metadata": { + "editable": true + }, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "db7a0783", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\frac{\\partial \\chi^2(\\boldsymbol{\\beta})}{\\partial \\beta_1} = -\\frac{2}{n}\\left[ \\sum_{i=0}^{n-1}x_i\\left(\\frac{y_i-\\beta_0-\\beta_1x_{i}}{\\sigma_i^2}\\right)\\right]=0.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "03a57144", + "metadata": { + "editable": true + }, + "source": [ + "## The $\\chi^2$ function\n", + "\n", + "For a linear fit (a first-order polynomial) we don't need to invert a matrix!! \n", + "Defining" + ] + }, + { + "cell_type": "markdown", + "id": "22e8525a", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma = \\sum_{i=0}^{n-1}\\frac{1}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e7dc4b83", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_x = \\sum_{i=0}^{n-1}\\frac{x_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "5ceab08b", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_y = \\sum_{i=0}^{n-1}\\left(\\frac{y_i}{\\sigma_i^2}\\right),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "61608838", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_{xx} = \\sum_{i=0}^{n-1}\\frac{x_ix_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "59fd7508", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\gamma_{xy} = \\sum_{i=0}^{n-1}\\frac{y_ix_{i}}{\\sigma_i^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "caca22c4", + "metadata": { + "editable": true + }, + "source": [ + "we obtain" + ] + }, + { + "cell_type": "markdown", + "id": "b459c22c", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_0 = \\frac{\\gamma_{xx}\\gamma_y-\\gamma_x\\gamma_y}{\\gamma\\gamma_{xx}-\\gamma_x^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3715e857", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\beta_1 = \\frac{\\gamma_{xy}\\gamma-\\gamma_x\\gamma_y}{\\gamma\\gamma_{xx}-\\gamma_x^2}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3ad3eeb2", + "metadata": { + "editable": true + }, + "source": [ + "This approach (different linear and non-linear regression) suffers\n", + "often from both being underdetermined and overdetermined in the\n", + "unknown coefficients $\\beta_i$. A better approach is to use the\n", + "Singular Value Decomposition (SVD) method discussed next week." + ] + }, + { + "cell_type": "markdown", + "id": "d8b193ef", + "metadata": { + "editable": true + }, + "source": [ + "## Fitting an Equation of State for Dense Nuclear Matter\n", + "\n", + "Before we continue, let us introduce yet another example. We are going to fit the\n", + "nuclear equation of state using results from many-body calculations.\n", + "The equation of state we have made available here, as function of\n", + "density, has been derived using modern nucleon-nucleon potentials with\n", + "[the addition of three-body\n", + "forces](https://www.sciencedirect.com/science/article/pii/S0370157399001106). This\n", + "time the file is presented as a standard **csv** file.\n", + "\n", + "The beginning of the Python code here is similar to what you have seen\n", + "before, with the same initializations and declarations. We use also\n", + "**pandas** again, rather extensively in order to organize our data.\n", + "\n", + "The difference now is that we use **Scikit-Learn's** regression tools\n", + "instead of our own matrix inversion implementation. Furthermore, we\n", + "sneak in **Ridge** regression (to be discussed below) which includes a\n", + "hyperparameter $\\lambda$, also to be explained below." + ] + }, + { + "cell_type": "markdown", + "id": "11007882", + "metadata": { + "editable": true + }, + "source": [ + "## The code" + ] + }, + { + "cell_type": "code", + "execution_count": 44, + "id": "ee8cdde5", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "# Common imports\n", + "import os\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "import matplotlib.pyplot as plt\n", + "import sklearn.linear_model as skl\n", + "from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "infile = open(data_path(\"EoS.csv\"),'r')\n", + "\n", + "# Read the EoS data as csv file and organize the data into two arrays with density and energies\n", + "EoS = pd.read_csv(infile, names=('Density', 'Energy'))\n", + "EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')\n", + "EoS = EoS.dropna()\n", + "Energies = EoS['Energy']\n", + "Density = EoS['Density']\n", + "# The design matrix now as function of various polytrops\n", + "X = np.zeros((len(Density),4))\n", + "X[:,3] = Density**(4.0/3.0)\n", + "X[:,2] = Density\n", + "X[:,1] = Density**(2.0/3.0)\n", + "X[:,0] = 1\n", + "\n", + "# We use now Scikit-Learn's linear regressor and ridge regressor\n", + "# OLS part\n", + "clf = skl.LinearRegression().fit(X, Energies)\n", + "ytilde = clf.predict(X)\n", + "EoS['Eols'] = ytilde\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, ytilde))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, ytilde))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, ytilde))\n", + "print(clf.coef_, clf.intercept_)\n", + "\n", + "# The Ridge regression with a hyperparameter lambda = 0.1\n", + "_lambda = 0.1\n", + "clf_ridge = skl.Ridge(alpha=_lambda).fit(X, Energies)\n", + "yridge = clf_ridge.predict(X)\n", + "EoS['Eridge'] = yridge\n", + "# The mean squared error \n", + "print(\"Mean squared error: %.2f\" % mean_squared_error(Energies, yridge))\n", + "# Explained variance score: 1 is perfect prediction \n", + "print('Variance score: %.2f' % r2_score(Energies, yridge))\n", + "# Mean absolute error \n", + "print('Mean absolute error: %.2f' % mean_absolute_error(Energies, yridge))\n", + "print(clf_ridge.coef_, clf_ridge.intercept_)\n", + "\n", + "fig, ax = plt.subplots()\n", + "ax.set_xlabel(r'$\\rho[\\mathrm{fm}^{-3}]$')\n", + "ax.set_ylabel(r'Energy per particle')\n", + "ax.plot(EoS['Density'], EoS['Energy'], alpha=0.7, lw=2,\n", + " label='Theoretical data')\n", + "ax.plot(EoS['Density'], EoS['Eols'], alpha=0.7, lw=2, c='m',\n", + " label='OLS')\n", + "ax.plot(EoS['Density'], EoS['Eridge'], alpha=0.7, lw=2, c='g',\n", + " label='Ridge $\\lambda = 0.1$')\n", + "ax.legend()\n", + "save_fig(\"EoSfitting\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "1ea53cff", + "metadata": { + "editable": true + }, + "source": [ + "The above simple polynomial in density $\\rho$ gives an excellent fit\n", + "to the data. \n", + "\n", + "We note also that there is a small deviation between the\n", + "standard OLS and the Ridge regression at higher densities. We discuss this in more detail\n", + "below." + ] + }, + { + "cell_type": "markdown", + "id": "4d725567", + "metadata": { + "editable": true + }, + "source": [ + "## Splitting our Data in Training and Test data\n", + "\n", + "It is normal in essentially all Machine Learning studies to split the\n", + "data in a training set and a test set (sometimes also an additional\n", + "validation set). **Scikit-Learn** has an own function for this. There\n", + "is no explicit recipe for how much data should be included as training\n", + "data and say test data. An accepted rule of thumb is to use\n", + "approximately $2/3$ to $4/5$ of the data as training data. We will\n", + "postpone a discussion of this splitting to the end of these notes and\n", + "our discussion of the so-called **bias-variance** tradeoff. Here we\n", + "limit ourselves to repeat the above equation of state fitting example\n", + "but now splitting the data into a training set and a test set." + ] + }, + { + "cell_type": "code", + "execution_count": 45, + "id": "aa6446c1", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "import os\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.model_selection import train_test_split\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')\n", + "\n", + "def R2(y_data, y_model):\n", + " return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2)\n", + "def MSE(y_data,y_model):\n", + " n = np.size(y_model)\n", + " return np.sum((y_data-y_model)**2)/n\n", + "\n", + "infile = open(data_path(\"EoS.csv\"),'r')\n", + "\n", + "# Read the EoS data as csv file and organized into two arrays with density and energies\n", + "EoS = pd.read_csv(infile, names=('Density', 'Energy'))\n", + "EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce')\n", + "EoS = EoS.dropna()\n", + "Energies = EoS['Energy']\n", + "Density = EoS['Density']\n", + "# The design matrix now as function of various polytrops\n", + "X = np.zeros((len(Density),5))\n", + "X[:,0] = 1\n", + "X[:,1] = Density**(2.0/3.0)\n", + "X[:,2] = Density\n", + "X[:,3] = Density**(4.0/3.0)\n", + "X[:,4] = Density**(5.0/3.0)\n", + "# We split the data in test and training data\n", + "X_train, X_test, y_train, y_test = train_test_split(X, Energies, test_size=0.2)\n", + "# matrix inversion to find beta\n", + "beta = np.linalg.inv(X_train.T.dot(X_train)).dot(X_train.T).dot(y_train)\n", + "# and then make the prediction\n", + "ytilde = X_train @ beta\n", + "print(\"Training R2\")\n", + "print(R2(y_train,ytilde))\n", + "print(\"Training MSE\")\n", + "print(MSE(y_train,ytilde))\n", + "ypredict = X_test @ beta\n", + "print(\"Test R2\")\n", + "print(R2(y_test,ypredict))\n", + "print(\"Test MSE\")\n", + "print(MSE(y_test,ypredict))" + ] + }, + { + "cell_type": "markdown", + "id": "7b772723", + "metadata": { + "editable": true + }, + "source": [ + "## Exercises\n", + "\n", + "Here are three possible exercises for week 34" + ] + }, + { + "cell_type": "markdown", + "id": "3cf1d40f", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 1: Setting up various Python environments\n", + "\n", + "The first exercise here is of a mere technical art. We want you to have \n", + "* git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). \n", + "\n", + "* Install various Python packages\n", + "\n", + "We will make extensive use of Python as programming language and its\n", + "myriad of available libraries. You will find\n", + "IPython/Jupyter notebooks invaluable in your work. You can run **R**\n", + "codes in the Jupyter/IPython notebooks, with the immediate benefit of\n", + "visualizing your data. You can also use compiled languages like C++,\n", + "Rust, Fortran etc if you prefer. The focus in these lectures will be\n", + "on Python.\n", + "\n", + "If you have Python installed (we recommend Python3) and you feel\n", + "pretty familiar with installing different packages, we recommend that\n", + "you install the following Python packages via **pip** as \n", + "\n", + "1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow \n", + "\n", + "For **Tensorflow**, we recommend following the instructions in the text of \n", + "[Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do)\n", + "\n", + "We will come back to **tensorflow** later. \n", + "\n", + "For Python3, replace **pip** with **pip3**.\n", + "\n", + "For OSX users we recommend, after having installed Xcode, to\n", + "install **brew**. Brew allows for a seamless installation of additional\n", + "software via for example \n", + "\n", + "1. brew install python3\n", + "\n", + "For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution,\n", + "you can use **pip** as well and simply install Python as \n", + "\n", + "1. sudo apt-get install python3 (or python for Python2.7)\n", + "\n", + "If you don't want to perform these operations separately and venture\n", + "into the hassle of exploring how to set up dependencies and paths, we\n", + "recommend two widely used distrubutions which set up all relevant\n", + "dependencies for Python, namely \n", + "\n", + "* [Anaconda](https://docs.anaconda.com/), \n", + "\n", + "which is an open source\n", + "distribution of the Python and R programming languages for large-scale\n", + "data processing, predictive analytics, and scientific computing, that\n", + "aims to simplify package management and deployment. Package versions\n", + "are managed by the package management system **conda**. \n", + "\n", + "* [Enthought canopy](https://www.enthought.com/product/canopy/) \n", + "\n", + "is a Python\n", + "distribution for scientific and analytic computing distribution and\n", + "analysis environment, available for free and under a commercial\n", + "license.\n", + "\n", + "We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment." + ] + }, + { + "cell_type": "markdown", + "id": "59288b2d", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 2: making your own data and exploring scikit-learn\n", + "\n", + "We will generate our own dataset for a function $y(x)$ where $x \\in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\\cal {N}(0,1)$.\n", + "The following simple Python instructions define our $x$ and $y$ values (with 100 data points)." + ] + }, + { + "cell_type": "code", + "execution_count": 46, + "id": "f06dd2f8", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "x = np.random.rand(100,1)\n", + "y = 2.0+5*x*x+0.1*np.random.randn(100,1)" + ] + }, + { + "cell_type": "markdown", + "id": "21171361", + "metadata": { + "editable": true + }, + "source": [ + "1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. \n", + "\n", + "2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. \n", + "\n", + "3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as" + ] + }, + { + "cell_type": "markdown", + "id": "0389374f", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "MSE(\\boldsymbol{y},\\boldsymbol{\\tilde{y}}) = \\frac{1}{n}\n", + "\\sum_{i=0}^{n-1}(y_i-\\tilde{y}_i)^2,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e85ae9f5", + "metadata": { + "editable": true + }, + "source": [ + "and the $R^2$ score function.\n", + "If $\\tilde{\\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as" + ] + }, + { + "cell_type": "markdown", + "id": "6eac8058", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "R^2(\\boldsymbol{y}, \\tilde{\\boldsymbol{y}}) = 1 - \\frac{\\sum_{i=0}^{n - 1} (y_i - \\tilde{y}_i)^2}{\\sum_{i=0}^{n - 1} (y_i - \\bar{y})^2},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c9d88838", + "metadata": { + "editable": true + }, + "source": [ + "where we have defined the mean value of $\\boldsymbol{y}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "4a767648", + "metadata": { + "editable": true + }, + "source": [ + "$$\n", + "\\bar{y} = \\frac{1}{n} \\sum_{i=0}^{n - 1} y_i.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "3b1a7735", + "metadata": { + "editable": true + }, + "source": [ + "You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. \n", + "Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits." + ] + }, + { + "cell_type": "markdown", + "id": "8c8ad214", + "metadata": { + "editable": true + }, + "source": [ + "## Exercise 3: Split data in test and training data\n", + "\n", + "In this exercise we want you to to compute the MSE for the training\n", + "data and the test data as function of the complexity of a polynomial,\n", + "that is the degree of a given polynomial.\n", + "\n", + "The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf).\n", + "\n", + "Our data is defined by $x\\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis." + ] + }, + { + "cell_type": "code", + "execution_count": 47, + "id": "43caea5f", + "metadata": { + "collapsed": false, + "editable": true + }, + "outputs": [], + "source": [ + "np.random.seed()\n", + "n = 100\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)" + ] + }, + { + "cell_type": "markdown", + "id": "9c604beb", + "metadata": { + "editable": true + }, + "source": [ + "where $y$ is the function we want to fit with a given polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "519328d9", + "metadata": { + "editable": true + }, + "source": [ + "**a)**\n", + "Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data." + ] + }, + { + "cell_type": "markdown", + "id": "3fdc48c8", + "metadata": { + "editable": true + }, + "source": [ + "**b)**\n", + "Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**)\n", + "and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial." + ] + }, + { + "cell_type": "markdown", + "id": "e6774cc5", + "metadata": { + "editable": true + }, + "source": [ + "**c)**\n", + "Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?" + ] + } + ], + "metadata": { + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.10" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} \ No newline at end of file diff --git a/doc/LectureNotes/_build/jupyter_execute/week34.py b/doc/LectureNotes/_build/jupyter_execute/week34.py new file mode 100644 index 000000000..80c831eda --- /dev/null +++ b/doc/LectureNotes/_build/jupyter_execute/week34.py @@ -0,0 +1,2589 @@ +#!/usr/bin/env python +# coding: utf-8 + +# +# + +# # Week 34: Introduction to the course, Logistics and Practicalities +# **Morten Hjorth-Jensen**, Department of Physics and Center for Computing in Science Education, University of Oslo, Norway and Department of Physics and Astronomy and Facility for Rare Isotope Beams, Michigan State University, USA +# +# Date: **Week 34, August 21-25, 2023** + +# ## Overview of first week +# +# 1. The sessions on Tuesdays and Wednesdays last four hours for each group (four groups in total) and will include lectures in a flipped mode (promoting active learning) and work on exercices and projects. +# +# 2. The sessions will begin with lectures, discussions, questions and answers about the material to be covered every week. Videos and teaching material will be announced in due time. +# +# 3. There are four groups: +# +# * Tuesdays 815am-12pm and 1215pm-4pm +# +# * Wednesdays 815am-12pm and 1215pm-4pm. +# +# 4. On Thursdays we have a regular lecture. These lectures start at 1215pm and end at 2pm and serve the aims of giving an overview over various topics. These lectures will also be recorded. +# +# The labs are also available till 6pm Tuesdays and Wednesdays. Videos and learning material with reading suggestions will be made available before each week starts. + +# ## Schedule first week +# +# * August 22: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group. +# +# * August 23: Presentation of the course, aims and content. Introduction to software and repetition of Python Programming, linear algebra and basic elements of statistics. Please select group. +# +# * August 24: Lecture: Linear regression, examples and theory + +# ## Lectures and ComputerLab +# +# * The sessions on Tuesdays and Wednesdays last four hours and will include partly lectures in a flipped mode (promoting active learning) and work on exercices and projects. +# +# * Thursdays: regular lectures (12.15pm-2pm) +# +# * Weekly reading assignments and videos needed to solve projects and exercises. +# +# * Weekly exercises. You can hand in exercises if you want and get an extra score, see below. +# +# * Detailed lecture notes, exercises, all programs presented, projects etc can be found at the homepage of the course. +# +# * Weekly plans and all other information are on the official website. This info will also be conveyed via weekly emails. +# +# * No final exam, three projects that are graded and have to be approved. + +# ## Communication channels +# +# * Chat and communications via +# +# * **Discord** channel will be added asap + +# ## Course Format +# +# * Three compulsory projects. Electronic reports only using [Canvas](https://www.uio.no/english/services/it/education/canvas/) to hand in projects and [git](https://git-scm.com/) as version control software and [GitHub](https://github.com/) for repository (or [GitLab](https://about.gitlab.com/)) of all your material. +# +# * Evaluation and grading: The three projects are graded and each counts 1/3 of the final mark. No final written or oral exam. +# +# a. For the last project each group/participant submits a proposal or works with suggested (by us) proposals for the project. +# +# b. If possible, we would like to organize the last project as a workshop where each group presents this to all other participants of the course +# +# c. Based on feedback etc, each group finalizes the report and submits for grading. +# +# * Python is the default programming language, but feel free to use C/C++, Julia and/or Fortran or other programming languages. All source codes discussed during the lectures can be found at the webpage and [github address](https://github.com/CompPhysics/MachineLearning/tree/master/doc/Programs) of the course. + +# ## Teachers +# +# * Morten Hjorth-Jensen, morten.hjorth-jensen@fys.uio.no +# +# * **Phone**: +47-48257387 +# +# * **Office**: Department of Physics, University of Oslo, Eastern wing, room FØ470 +# +# * **Office hours**: *Anytime*! Individual or group office hours can be arranged either in person or via zoom. Feel free to send an email for planning. +# +# * Ida Torkjellsdatter Storehaug, i.t.storehaug@fys.uio.no +# +# * Fahimeh Najafi, fahimeh.najafi@fys.uio.no +# +# * Mia-Katrin Ose Kvalsund, m.k.o.kvalsund@fys.uio.no +# +# * Karl Henrik Fredly, k.h.fredly@fys.uio.no +# +# * Adam Jakobsen, adam.jakobsen@fys.uio.no +# +# * Daniel Haas Beccatini Lima, d.h.b.lima@fys.uio.no + +# ## Deadlines for projects (tentative) +# +# 1. Project 1: October 9 (available September 4) graded with feedback) +# +# 2. Project 2: November 6 (available October 6, graded with feedback) +# +# 3. Project 3: December 11 (available November 10, graded with feedback) +# +# Extra Credit (not mandatory), weekly exercise assignments, 10 in total (due Friday same week), 10% additional score. The extra credit assignments are due each Friday and can be uploaed to **Canvas** in your preferred format (although we prefer jupyter-notebooks). First assignment is for week 35. Each weekly exercise set counts 1%. + +# ## Grading +# +# Grades are awarded on a scale from A to F, where A is the best grade and F is a fail. There are three projects which are graded and each project counts 1/3 of the final grade. The total score is thus the average from all three projects. +# +# The final number of points is based on the average of all projects and the grade follows the following table: +# +# * 92-100 points: A +# +# * 77-91 points: B +# +# * 58-76 points: C +# +# * 46-57 points: D +# +# * 40-45 points: E +# +# * 0-39 points: F-failed +# +# In addition you can get an extra 10% score for weekly assignments (10 in total and due each Friday). Each weekly assignment counts 1%. + +# ## Reading material +# +# The lecture notes are collected as a jupyter-book at . +# +# In addition to the lecture notes, we recommend the books of Bishop, Hastie et al, Murphy and Goodfellow et al. We will follow these texts closely and the weekly reading assignments refer to these texts. The text by Hastie et al is also widely used in the Machine Learning community. Finally, we also recommend the hands-on text by Geron, see next slide for links. + +# ## Textbooks +# +# * [Goodfellow, Bengio, and Courville (GBC), Deep Learning](https://www.deeplearningbook.org/) +# +# * Christopher M. Bishop (CB), Pattern Recognition and Machine Learning +# +# * [Hastie, Tibshirani, and Friedman (HTF), The Elements of Statistical Learning, Springer](https://www.springer.com/gp/book/9780387848570). +# +# * [Aurelien Geron (AG), Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/). This text is very useful since it contains many code examples and hands-on applications of all algorithms discussed in this course. +# +# * [Kevin Murphy (KM), Probabilistic Machine Learning, an Introduction](https://probml.github.io/pml-book/book1.html) + +# ## Reading suggestions week 34 +# +# This week: Refresh linear algebra, GBC chapters 1 and 2. HTF chapters 2 and 3. Install scikit-learn. See lecture notes for week 34 at (these notes). + +# ## Prerequisites +# +# Basic knowledge in programming and mathematics, with an emphasis on +# linear algebra. Knowledge of Python or/and C++ as programming +# languages is strongly recommended and experience with Jupiter notebook +# is recommended. Required courses are the equivalents to the University +# of Oslo mathematics courses MAT1100, MAT1110, MAT1120 and at least one +# of the corresponding computing and programming courses INF1000/INF1110 +# or MAT-INF1100/MAT-INF1100L/BIOS1100/KJM-INF1100. Most universities +# offer nowadays a basic programming course (often compulsory) where +# Python is the recurring programming language. + +# ## Topics covered in this course: Statistical analysis and optimization of data +# +# The course has two central parts +# +# 1. Statistical analysis and optimization of data +# +# 2. Machine learning +# +# These topics will be scattered thorughout the course and may not necessarily be taught separately. Rather, we will often take an approach (during the lectures and project/exercise sessions) where say elements from statistical data analysis are mixed with specific Machine Learning algorithms. + +# ## Statistical analysis and optimization of data +# +# We plan to cover the following topics: +# * Basic concepts, expectation values, variance, covariance, correlation functions and errors; +# +# * Simpler models, binomial distribution, the Poisson distribution, simple and multivariate normal distributions; +# +# * Central elements of Bayesian statistics and modeling; +# +# * Gradient methods for data optimization; +# +# * Monte Carlo methods, Markov chains, Gibbs sampling and Metropolis-Hastings sampling (tentative); +# +# * Estimation of errors and resampling techniques such as the cross-validation, blocking, bootstrapping and jackknife methods; +# +# * Principal Component Analysis (PCA) and its mathematical foundation; + +# ## Machine Learning +# +# * Pre deep-learning revolution (2008 approx) +# +# * Linear Regression and Logistic Regression, classification and regression problems; +# +# * Bayesian linear and logistic regression, kernel regression; +# +# * Decisions trees, Random Forests, Bagging and Boosting methods; +# +# * Support vector machines (only survey); +# +# * Unsupervised learning and dimensionality reduction, from PCA to clustering; +# +# * Deep learning +# +# * Neural networks and deep learning; +# +# * Convolutional neural networks; +# +# * Recurrent neural networks; +# +# * Autoencoders +# +# * Generative methods with an emphasis on Boltzmann Machines, Variational Autoencoders and Generalized Adversarial Networks; +# +# Hands-on demonstrations, exercises and projects aim at deepening your understanding of these topics. + +# ## Extremely useful tools, strongly recommended +# +# **and discussed at the lab sessions.** +# +# * GIT for version control, and GitHub or GitLab as repositories, highly recommended. This will be discussed during the first exercise session +# +# * Anaconda and other Python environments, see intro slides and links to programming resources at + +# ## Other courses on Data science and Machine Learning at UiO +# +# * [FYS5419 Quantum Computing and Quantum Machine Learning](https://www.uio.no/studier/emner/matnat/fys/FYS5419/index-eng.html) +# +# * [FYS5429 Advanced Machine Learning for the Physical Sciences](https://www.uio.no/studier/emner/matnat/fys/FYS5429/index-eng.html) +# +# * [STK2100 Machine learning and statistical methods for prediction and classification](http://www.uio.no/studier/emner/matnat/math/STK2100/index-eng.html). +# +# * [IN3050/4050 Introduction to Artificial Intelligence and Machine Learning](https://www.uio.no/studier/emner/matnat/ifi/IN3050/index-eng.html). Introductory course in machine learning and AI with an algorithmic approach. +# +# * [STK-INF3000/4000 Selected Topics in Data Science](http://www.uio.no/studier/emner/matnat/math/STK-INF3000/index-eng.html). The course provides insight into selected contemporary relevant topics within Data Science. +# +# * [IN4080 Natural Language Processing](https://www.uio.no/studier/emner/matnat/ifi/IN4080/index.html). Probabilistic and machine learning techniques applied to natural language processing. + +# ## Other courses on Data science and Machine Learning at UiO, contn +# +# * [STK-IN4300 Statistical learning methods in Data Science](https://www.uio.no/studier/emner/matnat/math/STK-IN4300/index-eng.html). An advanced introduction to statistical and machine learning. For students with a good mathematics and statistics background. +# +# * [IN3310/4310 Deep Learnig for Image Analysis](https://www.uio.no/studier/emner/matnat/ifi/IN4310/index.html) +# +# * [STK4051 Computational Statistics](https://www.uio.no/studier/emner/matnat/math/STK4051/index-eng.html) +# +# * [STK4021 Applied Bayesian Analysis and Numerical Methods](https://www.uio.no/studier/emner/matnat/math/STK4021/index-eng.html) + +# ## Learning outcomes +# +# This course aims at giving you insights and knowledge about many of +# the central algorithms used in Data Analysis and Machine Learning. +# The course is project based and through various numerical projects, +# normally three, you will be exposed to fundamental research problems +# in these fields, with the aim to reproduce state of the art scientific +# results. Both supervised and unsupervised methods will be covered. The +# emphasis is on a frequentist approach, although we will try to link it +# with a Bayesian approach as well. You will learn to develop and +# structure large codes for studying different cases where Machine +# Learning is applied to, get acquainted with computing facilities and +# learn to handle large scientific projects. A good scientific and +# ethical conduct is emphasized throughout the course. More +# specifically, after this course you will +# +# * Learn about basic data analysis, statistical analysis, Bayesian statistics, Monte Carlo sampling, data optimization and machine learning; +# +# * Be capable of extending the acquired knowledge to other systems and cases; +# +# * Have an understanding of central algorithms used in data analysis and machine learning; +# +# * Understand linear methods for regression and classification, from ordinary least squares, via Lasso and Ridge to Logistic regression; +# +# * Learn about neural networks and deep learning methods for supervised and unsupervised learning. Emphasis on feed forward neural networks, convolutional and recurrent neural networks; +# +# * Learn about about decision trees, random forests, bagging and boosting methods; +# +# * Learn about support vector machines and kernel transformations; +# +# * Reduction of data sets, from PCA to clustering; +# +# * Generative models +# +# * Work on numerical projects to illustrate the theory. The projects play a central role and you are expected to know modern programming languages like Python or C++ and/or Fortran (Fortran2003 or later) or Julia or other. + +# ## Introduction +# +# Our emphasis throughout this series of lectures +# is on understanding the mathematical aspects of +# different algorithms used in the fields of data analysis and machine learning. +# +# However, where possible we will emphasize the +# importance of using available software. We start thus with a hands-on +# and top-down approach to machine learning. The aim is thus to start with +# relevant data or data we have produced +# and use these to introduce statistical data analysis +# concepts and machine learning algorithms before we delve into the +# algorithms themselves. The examples we will use in the beginning, start with simple +# polynomials with random noise added. We will use the Python +# software package [Scikit-Learn](http://scikit-learn.org/stable/) and +# introduce various machine learning algorithms to make fits of +# the data and predictions. We move thereafter to more interesting +# cases such as data from say experiments (below we will look at experimental nuclear binding energies as an example). +# These are examples where we can easily set up the data and +# then use machine learning algorithms included in for example +# **Scikit-Learn**. +# +# These examples will serve us the purpose of getting +# started. Furthermore, they allow us to catch more than two birds with +# a stone. They will allow us to bring in some programming specific +# topics and tools as well as showing the power of various Python +# libraries for machine learning and statistical data analysis. +# +# Although we have projects where you write your own codes, we will also focus on two +# specific Python packages for Machine Learning, Scikit-Learn and +# Tensorflow with Keras (see below for links etc). Moreover, the examples we +# introduce will serve as inputs to many of our discussions later, as +# well as allowing you to set up models and produce your own data and +# get started with programming. + +# ## AI/ML and some statements you may have heard (and what do they mean?) +# +# 1. Fei-Fei Li on ImageNet: **map out the entire world of objects** ([The data that transformed AI research](https://cacm.acm.org/news/219702-the-data-that-transformed-ai-research-and-possibly-the-world/fulltext)) +# +# 2. Russell and Norvig in their popular textbook: **relevant to any intellectual task; it is truly a universal field** ([Artificial Intelligence, A modern approach](http://aima.cs.berkeley.edu/)) +# +# 3. Woody Bledsoe puts it more bluntly: **in the long run, AI is the only science** (quoted in Pamilla McCorduck, [Machines who think](https://www.pamelamccorduck.com/machines-who-think)) +# +# If you wish to have a critical read on AI/ML from a societal point of view, see [Kate Crawford's recent text Atlas of AI](https://www.katecrawford.net/) +# +# **Here: with AI/ML we intend a collection of machine learning methods with an emphasis on statistical learning and data analysis** + +# ## What is Machine Learning? +# +# Statistics, data science and machine learning form important fields of +# research in modern science. They describe how to learn and make +# predictions from data, as well as allowing us to extract important +# correlations about physical process and the underlying laws of motion +# in large data sets. The latter, big data sets, appear frequently in +# essentially all disciplines, from the traditional Science, Technology, +# Mathematics and Engineering fields to Life Science, Law, education +# research, the Humanities and the Social Sciences. +# +# It has become more +# and more common to see research projects on big data in for example +# the Social Sciences where extracting patterns from complicated survey +# data is one of many research directions. Having a solid grasp of data +# analysis and machine learning is thus becoming central to scientific +# computing in many fields, and competences and skills within the fields +# of machine learning and scientific computing are nowadays strongly +# requested by many potential employers. The latter cannot be +# overstated, familiarity with machine learning has almost become a +# prerequisite for many of the most exciting employment opportunities, +# whether they are in bioinformatics, life science, physics or finance, +# in the private or the public sector. This author has had several +# students or met students who have been hired recently based on their +# skills and competences in scientific computing and data science, often +# with marginal knowledge of machine learning. +# +# Machine learning is a subfield of computer science, and is closely +# related to computational statistics. It evolved from the study of +# pattern recognition in artificial intelligence (AI) research, and has +# made contributions to AI tasks like computer vision, natural language +# processing and speech recognition. Many of the methods we will study are also +# strongly rooted in basic mathematics and physics research. +# +# Ideally, machine learning represents the science of giving computers +# the ability to learn without being explicitly programmed. The idea is +# that there exist generic algorithms which can be used to find patterns +# in a broad class of data sets without having to write code +# specifically for each problem. The algorithm will build its own logic +# based on the data. You should however always keep in mind that +# machines and algorithms are to a large extent developed by humans. The +# insights and knowledge we have about a specific system, play a central +# role when we develop a specific machine learning algorithm. +# +# Machine learning is an extremely rich field, in spite of its young +# age. The increases we have seen during the last three decades in +# computational capabilities have been followed by developments of +# methods and techniques for analyzing and handling large date sets, +# relying heavily on statistics, computer science and mathematics. The +# field is rather new and developing rapidly. Popular software packages +# written in Python for machine learning like +# [Scikit-learn](http://scikit-learn.org/stable/), +# [Tensorflow](https://www.tensorflow.org/), +# [PyTorch](http://pytorch.org/) and [Keras](https://keras.io/), all +# freely available at their respective GitHub sites, encompass +# communities of developers in the thousands or more. And the number of +# code developers and contributors keeps increasing. Not all the +# algorithms and methods can be given a rigorous mathematical +# justification, opening up thereby large rooms for experimenting and +# trial and error and thereby exciting new developments. However, a +# solid command of linear algebra, multivariate theory, probability +# theory, statistical data analysis, understanding errors and Monte +# Carlo methods are central elements in a proper understanding of many +# of algorithms and methods we will discuss. + +# ## Types of Machine Learning +# +# The approaches to machine learning are many, but are often split into +# two main categories. In *supervised learning* we know the answer to a +# problem, and let the computer deduce the logic behind it. On the other +# hand, *unsupervised learning* is a method for finding patterns and +# relationship in data sets without any prior knowledge of the system. +# Some authours also operate with a third category, namely +# *reinforcement learning*. This is a paradigm of learning inspired by +# behavioral psychology, where learning is achieved by trial-and-error, +# solely from rewards and punishment. +# +# Another way to categorize machine learning tasks is to consider the +# desired output of a system. Some of the most common tasks are: +# +# * Classification: Outputs are divided into two or more classes. The goal is to produce a model that assigns inputs into one of these classes. An example is to identify digits based on pictures of hand-written ones. Classification is typically supervised learning. +# +# * Regression: Finding a functional relationship between an input data set and a reference data set. The goal is to construct a function that maps input data to continuous output values. +# +# * Clustering: Data are divided into groups with certain common traits, without knowing the different groups beforehand. It is thus a form of unsupervised learning. + +# ## Essential elements of ML +# +# The methods we cover have three main topics in common, irrespective of +# whether we deal with supervised or unsupervised learning. +# * The first ingredient is normally our data set (which can be subdivided into training, validation and test data). Many find the most difficult part of using Machine Learning to be the set up of your data in a meaningful way. +# +# * The second item is a model which is normally a function of some parameters. The model reflects our knowledge of the system (or lack thereof). As an example, if we know that our data show a behavior similar to what would be predicted by a polynomial, fitting our data to a polynomial of some degree would then determin our model. +# +# * The last ingredient is a so-called **cost/loss** function (or error or risk function) which allows us to present an estimate on how good our model is in reproducing the data it is supposed to train. + +# ## An optimization/minimization problem +# +# At the heart of basically all Machine Learning algorithms we will encounter so-called minimization or optimization algorithms. A large family of such methods are so-called **gradient methods**. + +# ## A Frequentist approach to data analysis +# +# When you hear phrases like **predictions and estimations** and +# **correlations and causations**, what do you think of? May be you think +# of the difference between classifying new data points and generating +# new data points. +# Or perhaps you consider that correlations represent some kind of symmetric statements like +# if $A$ is correlated with $B$, then $B$ is correlated with +# $A$. Causation on the other hand is directional, that is if $A$ causes $B$, $B$ does not +# necessarily cause $A$. +# +# These concepts are in some sense the difference between machine +# learning and statistics. In machine learning and prediction based +# tasks, we are often interested in developing algorithms that are +# capable of learning patterns from given data in an automated fashion, +# and then using these learned patterns to make predictions or +# assessments of newly given data. In many cases, our primary concern +# is the quality of the predictions or assessments, and we are less +# concerned about the underlying patterns that were learned in order +# to make these predictions. +# +# In machine learning we normally use [a so-called frequentist approach](https://en.wikipedia.org/wiki/Frequentist_inference), +# where the aim is to make predictions and find correlations. We focus +# less on for example extracting a probability distribution function (PDF). The PDF can be +# used in turn to make estimations and find causations such as given $A$ +# what is the likelihood of finding $B$. + +# ## What is a good model? +# +# In science and engineering we often end up in situations where we want to infer (or learn) a +# quantitative model $M$ for a given set of sample points $\boldsymbol{X} \in [x_1, x_2,\dots x_N]$. +# +# As we will see repeatedely in these lectures, we could try to fit these data points to a model given by a +# straight line, or if we wish to be more sophisticated to a more complex +# function. +# +# The reason for inferring such a model is that it +# serves many useful purposes. On the one hand, the model can reveal information +# encoded in the data or underlying mechanisms from which the data were generated. For instance, we could discover important +# corelations that relate interesting physics interpretations. +# +# In addition, it can simplify the representation of the given data set and help +# us in making predictions about future data samples. +# +# A first important consideration to keep in mind is that inferring the *correct* model +# for a given data set is an elusive, if not impossible, task. The fundamental difficulty +# is that if we are not specific about what we mean by a *correct* model, there +# could easily be many different models that fit the given data set *equally well*. + +# ## What is a good model? Can we define it? +# +# The central question is this: what leads us to say that a model is correct or +# optimal for a given data set? To make the model inference problem well posed, i.e., +# to guarantee that there is a unique optimal model for the given data, we need to +# impose additional assumptions or restrictions on the class of models considered. To +# this end, we should not be looking for just any model that can describe the data. +# Instead, we should look for a **model** $M$ that is the best among a restricted class +# of models. In addition, to make the model inference problem computationally +# tractable, we need to specify how restricted the class of models needs to be. A +# common strategy is to start +# with the simplest possible class of models that is just necessary to describe the data +# or solve the problem at hand. More precisely, the model class should be rich enough +# to contain at least one model that can fit the data to a desired accuracy and yet be +# restricted enough that it is relatively simple to find the best model for the given data. +# +# Thus, the most popular strategy is to start from the +# simplest class of models and increase the complexity of the models only when the +# simpler models become inadequate. For instance, if we work with a regression problem to fit a set of sample points, one +# may first try the simplest class of models, namely linear models, followed obviously by more complex models. +# +# How to evaluate which model fits best the data is something we will come back to over and over again in these sets of lectures. + +# ## Software and needed installations +# +# We will make extensive use of Python as programming language and its +# myriad of available libraries. You will find +# Jupyter notebooks invaluable in your work. You can run **R** +# codes in the Jupyter/IPython notebooks, with the immediate benefit of +# visualizing your data. You can also use compiled languages like C++, +# Rust, Julia, Fortran etc if you prefer. The focus in these lectures will be +# on Python. +# +# If you have Python installed (we strongly recommend Python3) and you feel +# pretty familiar with installing different packages, we recommend that +# you install the following Python packages via **pip** as +# +# 1. pip install numpy scipy matplotlib ipython scikit-learn mglearn sympy pandas pillow +# +# For Python3, replace **pip** with **pip3**. +# +# For OSX users we recommend, after having installed Xcode, to +# install **brew**. Brew allows for a seamless installation of additional +# software via for example +# +# 1. brew install python3 +# +# For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +# you can use **pip** as well and simply install Python as +# +# 1. sudo apt-get install python3 (or python for pyhton2.7) +# +# etc etc. + +# ## Python installers +# +# If you don't want to perform these operations separately and venture +# into the hassle of exploring how to set up dependencies and paths, we +# recommend two widely used distrubutions which set up all relevant +# dependencies for Python, namely +# +# * [Anaconda](https://docs.anaconda.com/), +# +# which is an open source +# distribution of the Python and R programming languages for large-scale +# data processing, predictive analytics, and scientific computing, that +# aims to simplify package management and deployment. Package versions +# are managed by the package management system **conda**. +# +# * [Enthought canopy](https://www.enthought.com/product/canopy/) +# +# is a Python +# distribution for scientific and analytic computing distribution and +# analysis environment, available for free and under a commercial +# license. +# +# Furthermore, [Google's Colab](https://colab.research.google.com/notebooks/welcome.ipynb) is a free Jupyter notebook environment that requires +# no setup and runs entirely in the cloud. Try it out! + +# ## Useful Python libraries +# Here we list several useful Python libraries we strongly recommend (if you use anaconda many of these are already there) +# +# * [NumPy](https://www.numpy.org/) is a highly popular library for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays +# +# * [The pandas](https://pandas.pydata.org/) library provides high-performance, easy-to-use data structures and data analysis tools +# +# * [Xarray](http://xarray.pydata.org/en/stable/) is a Python package that makes working with labelled multi-dimensional arrays simple, efficient, and fun! +# +# * [Scipy](https://www.scipy.org/) (pronounced “Sigh Pie”) is a Python-based ecosystem of open-source software for mathematics, science, and engineering. +# +# * [Matplotlib](https://matplotlib.org/) is a Python 2D plotting library which produces publication quality figures in a variety of hardcopy formats and interactive environments across platforms. +# +# * [Autograd](https://github.com/HIPS/autograd) can automatically differentiate native Python and Numpy code. It can handle a large subset of Python's features, including loops, ifs, recursion and closures, and it can even take derivatives of derivatives of derivatives +# +# * [JAX](https://jax.readthedocs.io/en/latest/index.html) has now more or less replaced **Autograd**. JAX is Autograd and XLA, brought together for high-performance numerical computing and machine learning research. It provides composable transformations of Python+NumPy programs: differentiate, vectorize, parallelize, Just-In-Time compile to GPU/TPU, and more. +# +# * [SymPy](https://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. +# +# * [scikit-learn](https://scikit-learn.org/stable/) has simple and efficient tools for machine learning, data mining and data analysis +# +# * [TensorFlow](https://www.tensorflow.org/) is a Python library for fast numerical computing created and released by Google +# +# * [Keras](https://keras.io/) is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano +# +# * And many more such as [pytorch](https://pytorch.org/), [Theano](https://pypi.org/project/Theano/) etc + +# ## Installing R, C++, cython or Julia +# +# You will also find it convenient to utilize **R**. We will mainly +# use Python during our lectures and in various projects and exercises. +# Those of you +# already familiar with **R** should feel free to continue using **R**, keeping +# however an eye on the parallel Python set ups. Similarly, if you are a +# Python afecionado, feel free to explore **R** as well. Jupyter(Julia, Python and R) /Ipython +# notebook allows you to run **R** codes and **Julia** codes interactively in your +# browser. The software library **R** is really tailored for statistical data analysis +# and allows for an easy usage of the tools and algorithms we will discuss in these +# lectures. +# +# To install **R** with Jupyter notebook +# [follow the link here](https://mpacer.org/maths/r-kernel-for-ipython-notebook) + +# ## Installing R, C++, cython, Numba etc +# +# For the C++ aficionados, Jupyter/IPython notebook allows you also to +# install C++ and run codes written in this language interactively in +# the browser. Since we will emphasize writing many of the algorithms +# yourself, you can thus opt for either Python or C++ (or Fortran or other compiled languages) as programming +# languages. +# +# To add more entropy, **cython** can also be used when running your +# notebooks. It means that Python with the jupyter notebook +# setup allows you to integrate widely popular softwares and tools for +# scientific computing. Similarly, the +# [Numba Python package](https://numba.pydata.org/) delivers increased performance +# capabilities with minimal rewrites of your codes. With its +# versatility, including symbolic operations, Python offers a unique +# computational environment. Your jupyter notebook can easily be +# converted into a nicely rendered **PDF** file or a Latex file for +# further processing. For example, convert to latex as + +# pycod jupyter nbconvert filename.ipynb --to latex +# + +# And to add more versatility, the Python package [SymPy](http://www.sympy.org/en/index.html) is a Python library for symbolic mathematics. It aims to become a full-featured computer algebra system (CAS) and is entirely written in Python. +# +# Finally, we recommend strongly using Autograd or JAX for automatic differentiation. + +# ## Numpy examples and Important Matrix and vector handling packages +# +# There are several central software libraries for linear algebra and eigenvalue problems. Several of the more +# popular ones have been wrapped into ofter software packages like those from the widely used text **Numerical Recipes**. The original source codes in many of the available packages are often taken from the widely used +# software package LAPACK, which follows two other popular packages +# developed in the 1970s, namely EISPACK and LINPACK. We describe them shortly here. +# +# * LINPACK: package for linear equations and least square problems. +# +# * LAPACK:package for solving symmetric, unsymmetric and generalized eigenvalue problems. From LAPACK's website it is possible to download for free all source codes from this library. Both C/C++ and Fortran versions are available. +# +# * BLAS (I, II and III): (Basic Linear Algebra Subprograms) are routines that provide standard building blocks for performing basic vector and matrix operations. Blas I is vector operations, II vector-matrix operations and III matrix-matrix operations. Highly parallelized and efficient codes, all available for download from . + +# ## Numpy and arrays +# [Numpy](http://www.numpy.org/) provides an easy way to handle arrays in Python. The standard way to import this library is as + +# In[1]: + + +import numpy as np + + +# Here follows a simple example where we set up an array of ten elements, all determined by random numbers drawn according to the normal distribution, + +# In[2]: + + +n = 10 +x = np.random.normal(size=n) +print(x) + + +# We defined a vector $x$ with $n=10$ elements with its values given by the Normal distribution $N(0,1)$. +# Another alternative is to declare a vector as follows + +# In[3]: + + +import numpy as np +x = np.array([1, 2, 3]) +print(x) + + +# Here we have defined a vector with three elements, with $x_0=1$, $x_1=2$ and $x_2=3$. Note that both Python and C++ +# start numbering array elements from $0$ and on. This means that a vector with $n$ elements has a sequence of entities $x_0, x_1, x_2, \dots, x_{n-1}$. We could also let (recommended) Numpy to compute the logarithms of a specific array as + +# In[4]: + + +import numpy as np +x = np.log(np.array([4, 7, 8])) +print(x) + + +# In the last example we used Numpy's unary function $np.log$. This function is +# highly tuned to compute array elements since the code is vectorized +# and does not require looping. We normaly recommend that you use the +# Numpy intrinsic functions instead of the corresponding **log** function +# from Python's **math** module. The looping is done explicitely by the +# **np.log** function. The alternative, and slower way to compute the +# logarithms of a vector would be to write + +# In[5]: + + +import numpy as np +from math import log +x = np.array([4, 7, 8]) +for i in range(0, len(x)): + x[i] = log(x[i]) +print(x) + + +# We note that our code is much longer already and we need to import the **log** function from the **math** module. +# The attentive reader will also notice that the output is $[1, 1, 2]$. Python interprets automagically our numbers as integers (like the **automatic** keyword in C++). To change this we could define our array elements to be double precision numbers as + +# In[6]: + + +import numpy as np +x = np.log(np.array([4, 7, 8], dtype = np.float64)) +print(x) + + +# or simply write them as double precision numbers (Python uses 64 bits as default for floating point type variables), that is + +# In[7]: + + +import numpy as np +x = np.log(np.array([4.0, 7.0, 8.0])) +print(x) + + +# To check the number of bytes (remember that one byte contains eight bits for double precision variables), you can use simple use the **itemsize** functionality (the array $x$ is actually an object which inherits the functionalities defined in Numpy) as + +# In[8]: + + +import numpy as np +x = np.log(np.array([4.0, 7.0, 8.0])) +print(x.itemsize) + + +# ## Matrices in Python +# +# Having defined vectors, we are now ready to try out matrices. We can +# define a $3 \times 3 $ real matrix $\boldsymbol{A}$ as (recall that we user +# lowercase letters for vectors and uppercase letters for matrices) + +# In[9]: + + +import numpy as np +A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ])) +print(A) + + +# If we use the **shape** function we would get $(3, 3)$ as output, that is verifying that our matrix is a $3\times 3$ matrix. We can slice the matrix and print for example the first column (Python organized matrix elements in a row-major order, see below) as + +# In[10]: + + +import numpy as np +A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ])) +# print the first column, row-major order and elements start with 0 +print(A[:,0]) + + +# We can continue this was by printing out other columns or rows. The example here prints out the second column + +# In[11]: + + +import numpy as np +A = np.log(np.array([ [4.0, 7.0, 8.0], [3.0, 10.0, 11.0], [4.0, 5.0, 7.0] ])) +# print the first column, row-major order and elements start with 0 +print(A[1,:]) + + +# Numpy contains many other functionalities that allow us to slice, subdivide etc etc arrays. We strongly recommend that you look up the [Numpy website for more details](http://www.numpy.org/). Useful functions when defining a matrix are the **np.zeros** function which declares a matrix of a given dimension and sets all elements to zero + +# In[12]: + + +import numpy as np +n = 10 +# define a matrix of dimension 10 x 10 and set all elements to zero +A = np.zeros( (n, n) ) +print(A) + + +# or initializing all elements to + +# In[13]: + + +import numpy as np +n = 10 +# define a matrix of dimension 10 x 10 and set all elements to one +A = np.ones( (n, n) ) +print(A) + + +# or as unitarily distributed random numbers (see the material on random number generators in the statistics part) + +# In[14]: + + +import numpy as np +n = 10 +# define a matrix of dimension 10 x 10 and set all elements to random numbers with x \in [0, 1] +A = np.random.rand(n, n) +print(A) + + +# As we will see throughout these lectures, there are several extremely useful functionalities in Numpy. +# As an example, consider the discussion of the covariance matrix. Suppose we have defined three vectors +# $\boldsymbol{x}, \boldsymbol{y}, \boldsymbol{z}$ with $n$ elements each. The covariance matrix is defined as + +# $$ +# \boldsymbol{\Sigma} = \begin{bmatrix} \sigma_{xx} & \sigma_{xy} & \sigma_{xz} \\ +# \sigma_{yx} & \sigma_{yy} & \sigma_{yz} \\ +# \sigma_{zx} & \sigma_{zy} & \sigma_{zz} +# \end{bmatrix}, +# $$ + +# where for example + +# $$ +# \sigma_{xy} =\frac{1}{n} \sum_{i=0}^{n-1}(x_i- \overline{x})(y_i- \overline{y}). +# $$ + +# The Numpy function **np.cov** calculates the covariance elements using the factor $1/(n-1)$ instead of $1/n$ since it assumes we do not have the exact mean values. +# The following simple function uses the **np.vstack** function which takes each vector of dimension $1\times n$ and produces a $3\times n$ matrix $\boldsymbol{W}$ + +# $$ +# \boldsymbol{W} = \begin{bmatrix} x_0 & x_1 & x_2 & \dots & x_{n-2} & x_{n-1} \\ +# y_0 & y_1 & y_2 & \dots & y_{n-2} & y_{n-1} \\ +# z_0 & z_1 & z_2 & \dots & z_{n-2} & z_{n-1} \\ +# \end{bmatrix}, +# $$ + +# which in turn is converted into into the $3\times 3$ covariance matrix +# $\boldsymbol{\Sigma}$ via the Numpy function **np.cov()**. We note that we can also calculate +# the mean value of each set of samples $\boldsymbol{x}$ etc using the Numpy +# function **np.mean(x)**. We can also extract the eigenvalues of the +# covariance matrix through the **np.linalg.eig()** function. + +# In[15]: + + +# Importing various packages +import numpy as np + +n = 100 +x = np.random.normal(size=n) +print(np.mean(x)) +y = 4+3*x+np.random.normal(size=n) +print(np.mean(y)) +z = x**3+np.random.normal(size=n) +print(np.mean(z)) +W = np.vstack((x, y, z)) +Sigma = np.cov(W) +print(Sigma) +Eigvals, Eigvecs = np.linalg.eig(Sigma) +print(Eigvals) + + +# In[16]: + + +get_ipython().run_line_magic('matplotlib', 'inline') + +import numpy as np +import matplotlib.pyplot as plt +from scipy import sparse +eye = np.eye(4) +print(eye) +sparse_mtx = sparse.csr_matrix(eye) +print(sparse_mtx) +x = np.linspace(-10,10,100) +y = np.sin(x) +plt.plot(x,y,marker='x') +plt.show() + + +# ## Meet the Pandas +# +# +# +# +#

Figure 1:

+# +# +# Another useful Python package is +# [pandas](https://pandas.pydata.org/), which is an open source library +# providing high-performance, easy-to-use data structures and data +# analysis tools for Python. **pandas** stands for panel data, a term borrowed from econometrics and is an efficient library for data analysis with an emphasis on tabular data. +# **pandas** has two major classes, the **DataFrame** class with two-dimensional data objects and tabular data organized in columns and the class **Series** with a focus on one-dimensional data objects. Both classes allow you to index data easily as we will see in the examples below. +# **pandas** allows you also to perform mathematical operations on the data, spanning from simple reshapings of vectors and matrices to statistical operations. +# +# The following simple example shows how we can, in an easy way make tables of our data. Here we define a data set which includes names, place of birth and date of birth, and displays the data in an easy to read way. We will see repeated use of **pandas**, in particular in connection with classification of data. + +# In[17]: + + +import pandas as pd +from IPython.display import display +data = {'First Name': ["Frodo", "Bilbo", "Aragorn II", "Samwise"], + 'Last Name': ["Baggins", "Baggins","Elessar","Gamgee"], + 'Place of birth': ["Shire", "Shire", "Eriador", "Shire"], + 'Date of Birth T.A.': [2968, 2890, 2931, 2980] + } +data_pandas = pd.DataFrame(data) +display(data_pandas) + + +# In the above we have imported **pandas** with the shorthand **pd**, the latter has become the standard way we import **pandas**. We make then a list of various variables +# and reorganize the aboves lists into a **DataFrame** and then print out a neat table with specific column labels as *Name*, *place of birth* and *date of birth*. +# Displaying these results, we see that the indices are given by the default numbers from zero to three. +# **pandas** is extremely flexible and we can easily change the above indices by defining a new type of indexing as + +# In[18]: + + +data_pandas = pd.DataFrame(data,index=['Frodo','Bilbo','Aragorn','Sam']) +display(data_pandas) + + +# Thereafter we display the content of the row which begins with the index **Aragorn** + +# In[19]: + + +display(data_pandas.loc['Aragorn']) + + +# We can easily append data to this, for example + +# In[20]: + + +new_hobbit = {'First Name': ["Peregrin"], + 'Last Name': ["Took"], + 'Place of birth': ["Shire"], + 'Date of Birth T.A.': [2990] + } +data_pandas=data_pandas.append(pd.DataFrame(new_hobbit, index=['Pippin'])) +display(data_pandas) + + +# Here are other examples where we use the **DataFrame** functionality to handle arrays, now with more interesting features for us, namely numbers. We set up a matrix +# of dimensionality $10\times 5$ and compute the mean value and standard deviation of each column. Similarly, we can perform mathematial operations like squaring the matrix elements and many other operations. + +# In[21]: + + +import numpy as np +import pandas as pd +from IPython.display import display +np.random.seed(100) +# setting up a 10 x 5 matrix +rows = 10 +cols = 5 +a = np.random.randn(rows,cols) +df = pd.DataFrame(a) +display(df) +print(df.mean()) +print(df.std()) +display(df**2) + + +# Thereafter we can select specific columns only and plot final results + +# In[22]: + + +df.columns = ['First', 'Second', 'Third', 'Fourth', 'Fifth'] +df.index = np.arange(10) + +display(df) +print(df['Second'].mean() ) +print(df.info()) +print(df.describe()) + +from pylab import plt, mpl +plt.style.use('seaborn') +mpl.rcParams['font.family'] = 'serif' + +df.cumsum().plot(lw=2.0, figsize=(10,6)) +plt.show() + + +df.plot.bar(figsize=(10,6), rot=15) +plt.show() + + +# We can produce a $4\times 4$ matrix + +# In[23]: + + +b = np.arange(16).reshape((4,4)) +print(b) +df1 = pd.DataFrame(b) +print(df1) + + +# and many other operations. +# +# The **Series** class is another important class included in +# **pandas**. You can view it as a specialization of **DataFrame** but where +# we have just a single column of data. It shares many of the same features as **DataFrame**. As with **DataFrame**, +# most operations are vectorized, achieving thereby a high performance when dealing with computations of arrays, in particular labeled arrays. +# As we will see below it leads also to a very concice code close to the mathematical operations we may be interested in. +# For multidimensional arrays, we recommend strongly [xarray](http://xarray.pydata.org/en/stable/). **xarray** has much of the same flexibility as **pandas**, but allows for the extension to higher dimensions than two. We will see examples later of the usage of both **pandas** and **xarray**. + +# ### Simple linear regression model using **scikit-learn** +# +# We start with perhaps our simplest possible example, using **Scikit-Learn** to perform linear regression analysis on a data set produced by us. +# +# What follows is a simple Python code where we have defined a function +# $y$ in terms of the variable $x$. Both are defined as vectors with $100$ entries. +# The numbers in the vector $\boldsymbol{x}$ are given +# by random numbers generated with a uniform distribution with entries +# $x_i \in [0,1]$ (more about probability distribution functions +# later). These values are then used to define a function $y(x)$ +# (tabulated again as a vector) with a linear dependence on $x$ plus a +# random noise added via the normal distribution. +# +# The Numpy functions are imported used the **import numpy as np** +# statement and the random number generator for the uniform distribution +# is called using the function **np.random.rand()**, where we specificy +# that we want $100$ random variables. Using Numpy we define +# automatically an array with the specified number of elements, $100$ in +# our case. With the Numpy function **randn()** we can compute random +# numbers with the normal distribution (mean value $\mu$ equal to zero and +# variance $\sigma^2$ set to one) and produce the values of $y$ assuming a linear +# dependence as function of $x$ + +# $$ +# y = 2x+N(0,1), +# $$ + +# where $N(0,1)$ represents random numbers generated by the normal +# distribution. From **Scikit-Learn** we import then the +# **LinearRegression** functionality and make a prediction $\tilde{y} = +# \alpha + \beta x$ using the function **fit(x,y)**. We call the set of +# data $(\boldsymbol{x},\boldsymbol{y})$ for our training data. The Python package +# **scikit-learn** has also a functionality which extracts the above +# fitting parameters $\alpha$ and $\beta$ (see below). Later we will +# distinguish between training data and test data. +# +# For plotting we use the Python package +# [matplotlib](https://matplotlib.org/) which produces publication +# quality figures. Feel free to explore the extensive +# [gallery](https://matplotlib.org/gallery/index.html) of examples. In +# this example we plot our original values of $x$ and $y$ as well as the +# prediction **ypredict** ($\tilde{y}$), which attempts at fitting our +# data with a straight line. +# +# The Python code follows here. + +# In[24]: + + +# Importing various packages +import numpy as np +import matplotlib.pyplot as plt +from sklearn.linear_model import LinearRegression + +x = np.random.rand(100,1) +y = 2*x+np.random.randn(100,1) +linreg = LinearRegression() +linreg.fit(x,y) +xnew = np.array([[0],[1]]) +ypredict = linreg.predict(xnew) + +plt.plot(xnew, ypredict, "r-") +plt.plot(x, y ,'ro') +plt.axis([0,1.0,0, 5.0]) +plt.xlabel(r'$x$') +plt.ylabel(r'$y$') +plt.title(r'Simple Linear Regression') +plt.show() + + +# This example serves several aims. It allows us to demonstrate several +# aspects of data analysis and later machine learning algorithms. The +# immediate visualization shows that our linear fit is not +# impressive. It goes through the data points, but there are many +# outliers which are not reproduced by our linear regression. We could +# now play around with this small program and change for example the +# factor in front of $x$ and the normal distribution. Try to change the +# function $y$ to + +# $$ +# y = 10x+0.01 \times N(0,1), +# $$ + +# where $x$ is defined as before. Does the fit look better? Indeed, by +# reducing the role of the noise given by the normal distribution we see immediately that +# our linear prediction seemingly reproduces better the training +# set. However, this testing 'by the eye' is obviouly not satisfactory in the +# long run. Here we have only defined the training data and our model, and +# have not discussed a more rigorous approach to the **cost** function. +# +# We need more rigorous criteria in defining whether we have succeeded or +# not in modeling our training data. You will be surprised to see that +# many scientists seldomly venture beyond this 'by the eye' approach. A +# standard approach for the *cost* function is the so-called $\chi^2$ +# function (a variant of the mean-squared error (MSE)) + +# $$ +# \chi^2 = \frac{1}{n} +# \sum_{i=0}^{n-1}\frac{(y_i-\tilde{y}_i)^2}{\sigma_i^2}, +# $$ + +# where $\sigma_i^2$ is the variance (to be defined later) of the entry +# $y_i$. We may not know the explicit value of $\sigma_i^2$, it serves +# however the aim of scaling the equations and make the cost function +# dimensionless. +# +# Minimizing the cost function is a central aspect of +# our discussions to come. Finding its minima as function of the model +# parameters ($\alpha$ and $\beta$ in our case) will be a recurring +# theme in these series of lectures. Essentially all machine learning +# algorithms we will discuss center around the minimization of the +# chosen cost function. This depends in turn on our specific +# model for describing the data, a typical situation in supervised +# learning. Automatizing the search for the minima of the cost function is a +# central ingredient in all algorithms. Typical methods which are +# employed are various variants of **gradient** methods. These will be +# discussed in more detail later. Again, you'll be surprised to hear that +# many practitioners minimize the above function ''by the eye', popularly dubbed as +# 'chi by the eye'. That is, change a parameter and see (visually and numerically) that +# the $\chi^2$ function becomes smaller. +# +# There are many ways to define the cost function. A simpler approach is to look at the relative difference between the training data and the predicted data, that is we define +# the relative error (why would we prefer the MSE instead of the relative error?) as + +# $$ +# \epsilon_{\mathrm{relative}}= \frac{\vert \boldsymbol{y} -\boldsymbol{\tilde{y}}\vert}{\vert \boldsymbol{y}\vert}. +# $$ + +# The squared cost function results in an arithmetic mean-unbiased +# estimator, and the absolute-value cost function results in a +# median-unbiased estimator (in the one-dimensional case, and a +# geometric median-unbiased estimator for the multi-dimensional +# case). The squared cost function has the disadvantage that it has the tendency +# to be dominated by outliers. +# +# We can modify easily the above Python code and plot the relative error instead + +# In[25]: + + +import numpy as np +import matplotlib.pyplot as plt +from sklearn.linear_model import LinearRegression + +x = np.random.rand(100,1) +y = 5*x+0.01*np.random.randn(100,1) +linreg = LinearRegression() +linreg.fit(x,y) +ypredict = linreg.predict(x) + +plt.plot(x, np.abs(ypredict-y)/abs(y), "ro") +plt.axis([0,1.0,0.0, 0.5]) +plt.xlabel(r'$x$') +plt.ylabel(r'$\epsilon_{\mathrm{relative}}$') +plt.title(r'Relative error') +plt.show() + + +# Depending on the parameter in front of the normal distribution, we may +# have a small or larger relative error. Try to play around with +# different training data sets and study (graphically) the value of the +# relative error. +# +# As mentioned above, **Scikit-Learn** has an impressive functionality. +# We can for example extract the values of $\alpha$ and $\beta$ and +# their error estimates, or the variance and standard deviation and many +# other properties from the statistical data analysis. +# +# Here we show an +# example of the functionality of **Scikit-Learn**. + +# In[26]: + + +import numpy as np +import matplotlib.pyplot as plt +from sklearn.linear_model import LinearRegression +from sklearn.metrics import mean_squared_error, r2_score, mean_squared_log_error, mean_absolute_error + +x = np.random.rand(100,1) +y = 2.0+ 5*x+0.5*np.random.randn(100,1) +linreg = LinearRegression() +linreg.fit(x,y) +ypredict = linreg.predict(x) +print('The intercept alpha: \n', linreg.intercept_) +print('Coefficient beta : \n', linreg.coef_) +# The mean squared error +print("Mean squared error: %.2f" % mean_squared_error(y, ypredict)) +# Explained variance score: 1 is perfect prediction +print('Variance score: %.2f' % r2_score(y, ypredict)) +# Mean squared log error +print('Mean squared log error: %.2f' % mean_squared_log_error(y, ypredict) ) +# Mean absolute error +print('Mean absolute error: %.2f' % mean_absolute_error(y, ypredict)) +plt.plot(x, ypredict, "r-") +plt.plot(x, y ,'ro') +plt.axis([0.0,1.0,1.5, 7.0]) +plt.xlabel(r'$x$') +plt.ylabel(r'$y$') +plt.title(r'Linear Regression fit ') +plt.show() + + +# The function **coef** gives us the parameter $\beta$ of our fit while **intercept** yields +# $\alpha$. Depending on the constant in front of the normal distribution, we get values near or far from $\alpha =2$ and $\beta =5$. Try to play around with different parameters in front of the normal distribution. The function **meansquarederror** gives us the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error or loss defined as + +# $$ +# MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +# \sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +# $$ + +# The smaller the value, the better the fit. Ideally we would like to +# have an MSE equal zero. The attentive reader has probably recognized +# this function as being similar to the $\chi^2$ function defined above. +# +# The **r2score** function computes $R^2$, the coefficient of +# determination. It provides a measure of how well future samples are +# likely to be predicted by the model. Best possible score is 1.0 and it +# can be negative (because the model can be arbitrarily worse). A +# constant model that always predicts the expected value of $\boldsymbol{y}$, +# disregarding the input features, would get a $R^2$ score of $0.0$. +# +# If $\tilde{\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as + +# $$ +# R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +# $$ + +# where we have defined the mean value of $\boldsymbol{y}$ as + +# $$ +# \bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +# $$ + +# Another quantity taht we will meet again in our discussions of regression analysis is +# the mean absolute error (MAE), a risk metric corresponding to the expected value of the absolute error loss or what we call the $l1$-norm loss. In our discussion above we presented the relative error. +# The MAE is defined as follows + +# $$ +# \text{MAE}(\boldsymbol{y}, \boldsymbol{\tilde{y}}) = \frac{1}{n} \sum_{i=0}^{n-1} \left| y_i - \tilde{y}_i \right|. +# $$ + +# We present the +# squared logarithmic (quadratic) error + +# $$ +# \text{MSLE}(\boldsymbol{y}, \boldsymbol{\tilde{y}}) = \frac{1}{n} \sum_{i=0}^{n - 1} (\log_e (1 + y_i) - \log_e (1 + \tilde{y}_i) )^2, +# $$ + +# where $\log_e (x)$ stands for the natural logarithm of $x$. This error +# estimate is best to use when targets having exponential growth, such +# as population counts, average sales of a commodity over a span of +# years etc. +# +# Finally, another cost function is the Huber cost function used in robust regression. +# +# The rationale behind this possible cost function is its reduced +# sensitivity to outliers in the data set. In our discussions on +# dimensionality reduction and normalization of data we will meet other +# ways of dealing with outliers. +# +# The Huber cost function is defined as + +# $$ +# H_{\delta}(\boldsymbol{a})=\left\{\begin{array}{cc}\frac{1}{2} \boldsymbol{a}^{2}& \text{for }|\boldsymbol{a}|\leq \delta\\ \delta (|\boldsymbol{a}|-\frac{1}{2}\delta ),&\text{otherwise}.\end{array}\right. +# $$ + +# Here $\boldsymbol{a}=\boldsymbol{y} - \boldsymbol{\tilde{y}}$. +# +# We will discuss in more detail these and other functions in the +# various lectures and lab sessions. + +# ### To our real data: nuclear binding energies. Brief reminder on masses and binding energies +# +# Let us now dive into nuclear physics and remind ourselves briefly about some basic features about binding +# energies. A basic quantity which can be measured for the ground +# states of nuclei is the atomic mass $M(N, Z)$ of the neutral atom with +# atomic mass number $A$ and charge $Z$. The number of neutrons is $N$. There are indeed several sophisticated experiments worldwide which allow us to measure this quantity to high precision (parts per million even). +# +# Atomic masses are usually tabulated in terms of the mass excess defined by + +# $$ +# \Delta M(N, Z) = M(N, Z) - uA, +# $$ + +# where $u$ is the Atomic Mass Unit + +# $$ +# u = M(^{12}\mathrm{C})/12 = 931.4940954(57) \hspace{0.1cm} \mathrm{MeV}/c^2. +# $$ + +# The nucleon masses are + +# $$ +# m_p = 1.00727646693(9)u, +# $$ + +# and + +# $$ +# m_n = 939.56536(8)\hspace{0.1cm} \mathrm{MeV}/c^2 = 1.0086649156(6)u. +# $$ + +# In the [2016 mass evaluation of by W.J.Huang, G.Audi, M.Wang, F.G.Kondev, S.Naimi and X.Xu](http://nuclearmasses.org/resources_folder/Wang_2017_Chinese_Phys_C_41_030003.pdf) +# there are data on masses and decays of 3437 nuclei. +# +# The nuclear binding energy is defined as the energy required to break +# up a given nucleus into its constituent parts of $N$ neutrons and $Z$ +# protons. In terms of the atomic masses $M(N, Z)$ the binding energy is +# defined by + +# $$ +# BE(N, Z) = ZM_H c^2 + Nm_n c^2 - M(N, Z)c^2 , +# $$ + +# where $M_H$ is the mass of the hydrogen atom and $m_n$ is the mass of the neutron. +# In terms of the mass excess the binding energy is given by + +# $$ +# BE(N, Z) = Z\Delta_H c^2 + N\Delta_n c^2 -\Delta(N, Z)c^2 , +# $$ + +# where $\Delta_H c^2 = 7.2890$ MeV and $\Delta_n c^2 = 8.0713$ MeV. +# +# A popular and physically intuitive model which can be used to parametrize +# the experimental binding energies as function of $A$, is the so-called +# **liquid drop model**. The ansatz is based on the following expression + +# $$ +# BE(N,Z) = a_1A-a_2A^{2/3}-a_3\frac{Z^2}{A^{1/3}}-a_4\frac{(N-Z)^2}{A}, +# $$ + +# where $A$ stands for the number of nucleons and the $a_i$s are parameters which are determined by a fit +# to the experimental data. +# +# To arrive at the above expression we have assumed that we can make the following assumptions: +# +# * There is a volume term $a_1A$ proportional with the number of nucleons (the energy is also an extensive quantity). When an assembly of nucleons of the same size is packed together into the smallest volume, each interior nucleon has a certain number of other nucleons in contact with it. This contribution is proportional to the volume. +# +# * There is a surface energy term $a_2A^{2/3}$. The assumption here is that a nucleon at the surface of a nucleus interacts with fewer other nucleons than one in the interior of the nucleus and hence its binding energy is less. This surface energy term takes that into account and is therefore negative and is proportional to the surface area. +# +# * There is a Coulomb energy term $a_3\frac{Z^2}{A^{1/3}}$. The electric repulsion between each pair of protons in a nucleus yields less binding. +# +# * There is an asymmetry term $a_4\frac{(N-Z)^2}{A}$. This term is associated with the Pauli exclusion principle and reflects the fact that the proton-neutron interaction is more attractive on the average than the neutron-neutron and proton-proton interactions. +# +# We could also add a so-called pairing term, which is a correction term that +# arises from the tendency of proton pairs and neutron pairs to +# occur. An even number of particles is more stable than an odd number. + +# ### Organizing our data +# +# Let us start with reading and organizing our data. +# We start with the compilation of masses and binding energies from 2016. +# After having downloaded this file to our own computer, we are now ready to read the file and start structuring our data. +# +# We start with preparing folders for storing our calculations and the data file over masses and binding energies. We import also various modules that we will find useful in order to present various Machine Learning methods. Here we focus mainly on the functionality of **scikit-learn**. + +# In[27]: + + +# Common imports +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +import sklearn.linear_model as skl +from sklearn.model_selection import train_test_split +from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error +import os + +# Where to save the figures and data files +PROJECT_ROOT_DIR = "Results" +FIGURE_ID = "Results/FigureFiles" +DATA_ID = "DataFiles/" + +if not os.path.exists(PROJECT_ROOT_DIR): + os.mkdir(PROJECT_ROOT_DIR) + +if not os.path.exists(FIGURE_ID): + os.makedirs(FIGURE_ID) + +if not os.path.exists(DATA_ID): + os.makedirs(DATA_ID) + +def image_path(fig_id): + return os.path.join(FIGURE_ID, fig_id) + +def data_path(dat_id): + return os.path.join(DATA_ID, dat_id) + +def save_fig(fig_id): + plt.savefig(image_path(fig_id) + ".png", format='png') + +infile = open(data_path("MassEval2016.dat"),'r') + + +# Before we proceed, we define also a function for making our plots. You can obviously avoid this and simply set up various **matplotlib** commands every time you need them. You may however find it convenient to collect all such commands in one function and simply call this function. + +# In[28]: + + +from pylab import plt, mpl +plt.style.use('seaborn') +mpl.rcParams['font.family'] = 'serif' + +def MakePlot(x,y, styles, labels, axlabels): + plt.figure(figsize=(10,6)) + for i in range(len(x)): + plt.plot(x[i], y[i], styles[i], label = labels[i]) + plt.xlabel(axlabels[0]) + plt.ylabel(axlabels[1]) + plt.legend(loc=0) + + +# Our next step is to read the data on experimental binding energies and +# reorganize them as functions of the mass number $A$, the number of +# protons $Z$ and neutrons $N$ using **pandas**. Before we do this it is +# always useful (unless you have a binary file or other types of compressed +# data) to actually open the file and simply take a look at it! +# +# In particular, the program that outputs the final nuclear masses is written in Fortran with a specific format. It means that we need to figure out the format and which columns contain the data we are interested in. Pandas comes with a function that reads formatted output. After having admired the file, we are now ready to start massaging it with **pandas**. The file begins with some basic format information. + +# In[29]: + + +""" +This is taken from the data file of the mass 2016 evaluation. +All files are 3436 lines long with 124 character per line. + Headers are 39 lines long. + col 1 : Fortran character control: 1 = page feed 0 = line feed + format : a1,i3,i5,i5,i5,1x,a3,a4,1x,f13.5,f11.5,f11.3,f9.3,1x,a2,f11.3,f9.3,1x,i3,1x,f12.5,f11.5 + These formats are reflected in the pandas widths variable below, see the statement + widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), + Pandas has also a variable header, with length 39 in this case. +""" + + +# The data we are interested in are in columns 2, 3, 4 and 11, giving us +# the number of neutrons, protons, mass numbers and binding energies, +# respectively. We add also for the sake of completeness the element name. The data are in fixed-width formatted lines and we will +# covert them into the **pandas** DataFrame structure. + +# In[30]: + + +# Read the experimental data with Pandas +Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11), + names=('N', 'Z', 'A', 'Element', 'Ebinding'), + widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), + header=39, + index_col=False) + +# Extrapolated values are indicated by '#' in place of the decimal place, so +# the Ebinding column won't be numeric. Coerce to float and drop these entries. +Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce') +Masses = Masses.dropna() +# Convert from keV to MeV. +Masses['Ebinding'] /= 1000 + +# Group the DataFrame by nucleon number, A. +Masses = Masses.groupby('A') +# Find the rows of the grouped DataFrame with the maximum binding energy. +Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()]) + + +# We have now read in the data, grouped them according to the variables we are interested in. +# We see how easy it is to reorganize the data using **pandas**. If we +# were to do these operations in C/C++ or Fortran, we would have had to +# write various functions/subroutines which perform the above +# reorganizations for us. Having reorganized the data, we can now start +# to make some simple fits using both the functionalities in **numpy** and +# **Scikit-Learn** afterwards. +# +# Now we define five variables which contain +# the number of nucleons $A$, the number of protons $Z$ and the number of neutrons $N$, the element name and finally the energies themselves. + +# In[31]: + + +A = Masses['A'] +Z = Masses['Z'] +N = Masses['N'] +Element = Masses['Element'] +Energies = Masses['Ebinding'] +print(Masses) + + +# The next step, and we will define this mathematically later, is to set up the so-called **design matrix**. We will throughout call this matrix $\boldsymbol{X}$. +# It has dimensionality $p\times n$, where $n$ is the number of data points and $p$ are the so-called predictors. In our case here they are given by the number of polynomials in $A$ we wish to include in the fit. + +# In[32]: + + +# Now we set up the design matrix X +X = np.zeros((len(A),5)) +X[:,0] = 1 +X[:,1] = A +X[:,2] = A**(2.0/3.0) +X[:,3] = A**(-1.0/3.0) +X[:,4] = A**(-1.0) + + +# With **scikitlearn** we are now ready to use linear regression and fit our data. + +# In[33]: + + +clf = skl.LinearRegression().fit(X, Energies) +fity = clf.predict(X) + + +# Pretty simple! +# Now we can print measures of how our fit is doing, the coefficients from the fits and plot the final fit together with our data. + +# In[34]: + + +# The mean squared error +print("Mean squared error: %.2f" % mean_squared_error(Energies, fity)) +# Explained variance score: 1 is perfect prediction +print('Variance score: %.2f' % r2_score(Energies, fity)) +# Mean absolute error +print('Mean absolute error: %.2f' % mean_absolute_error(Energies, fity)) +print(clf.coef_, clf.intercept_) + +Masses['Eapprox'] = fity +# Generate a plot comparing the experimental with the fitted values values. +fig, ax = plt.subplots() +ax.set_xlabel(r'$A = N + Z$') +ax.set_ylabel(r'$E_\mathrm{bind}\,/\mathrm{MeV}$') +ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2, + label='Ame2016') +ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m', + label='Fit') +ax.legend() +save_fig("Masses2016") +plt.show() + + +# ### And what about using neural networks? +# +# The **seaborn** package allows us to visualize data in an efficient way. Note that we use **scikit-learn**'s multi-layer perceptron (or feed forward neural network) +# functionality. + +# In[35]: + + +from sklearn.neural_network import MLPRegressor +from sklearn.metrics import accuracy_score +import seaborn as sns + +X_train = X +Y_train = Energies +n_hidden_neurons = 100 +epochs = 100 +# store models for later use +eta_vals = np.logspace(-5, 1, 7) +lmbd_vals = np.logspace(-5, 1, 7) +# store the models for later use +DNN_scikit = np.zeros((len(eta_vals), len(lmbd_vals)), dtype=object) +train_accuracy = np.zeros((len(eta_vals), len(lmbd_vals))) +sns.set() +for i, eta in enumerate(eta_vals): + for j, lmbd in enumerate(lmbd_vals): + dnn = MLPRegressor(hidden_layer_sizes=(n_hidden_neurons), activation='logistic', + alpha=lmbd, learning_rate_init=eta, max_iter=epochs) + dnn.fit(X_train, Y_train) + DNN_scikit[i][j] = dnn + train_accuracy[i][j] = dnn.score(X_train, Y_train) + +fig, ax = plt.subplots(figsize = (10, 10)) +sns.heatmap(train_accuracy, annot=True, ax=ax, cmap="viridis") +ax.set_title("Training Accuracy") +ax.set_ylabel("$\eta$") +ax.set_xlabel("$\lambda$") +plt.show() + + +# ## A first summary +# +# The aim behind these introductory words was to present to you various +# Python libraries and their functionalities, in particular libraries like +# **numpy**, **pandas**, **xarray** and **matplotlib** and other that make our life much easier +# in handling various data sets and visualizing data. +# +# Furthermore, +# **Scikit-Learn** allows us with few lines of code to implement popular +# Machine Learning algorithms for supervised learning. Later we will meet **Tensorflow**, a powerful library for deep learning. +# Now it is time to dive more into the details of various methods. We will start with linear regression and try to take a deeper look at what it entails. + +# ## Why Linear Regression (aka Ordinary Least Squares and family) +# +# Fitting a continuous function with linear parameterization in terms of the parameters $\boldsymbol{\beta}$. +# * Method of choice for fitting a continuous function! +# +# * Gives an excellent introduction to central Machine Learning features with **understandable pedagogical** links to other methods like **Neural Networks**, **Support Vector Machines** etc +# +# * Analytical expression for the fitting parameters $\boldsymbol{\beta}$ +# +# * Analytical expressions for statistical propertiers like mean values, variances, confidence intervals and more +# +# * Analytical relation with probabilistic interpretations +# +# * Easy to introduce basic concepts like bias-variance tradeoff, cross-validation, resampling and regularization techniques and many other ML topics +# +# * Easy to code! And links well with classification problems and logistic regression and neural networks +# +# * Allows for **easy** hands-on understanding of gradient descent methods +# +# * and many more features +# +# For more discussions of Ridge and Lasso regression, [Wessel van Wieringen's](https://arxiv.org/abs/1509.09169) article is highly recommended. +# Similarly, [Mehta et al's article](https://arxiv.org/abs/1803.08823) is also recommended. + +# ## Regression analysis, overarching aims +# +# Regression modeling deals with the description of the sampling distribution of a given random variable $y$ and how it varies as function of another variable or a set of such variables $\boldsymbol{x} =[x_0, x_1,\dots, x_{n-1}]^T$. +# The first variable is called the **dependent**, the **outcome** or the **response** variable while the set of variables $\boldsymbol{x}$ is called the independent variable, or the predictor variable or the explanatory variable, or simply just the **inputs**. +# +# A regression model aims at finding a likelihood function $p(\boldsymbol{y}\vert \boldsymbol{x})$ or in the more traditional sense a function $\boldsymbol{y}(\boldsymbol{x})$, that is the conditional distribution for $\boldsymbol{y}$ with a given $\boldsymbol{x}$. The estimation of $p(\boldsymbol{y}\vert \boldsymbol{x})$ is made using a data set with +# * $n$ cases $i = 0, 1, 2, \dots, n-1$ +# +# * Response (target, dependent or outcome) variable $y_i$ with $i = 0, 1, 2, \dots, n-1$ +# +# * $p$ so-called explanatory (independent or predictor or feature) variables $\boldsymbol{x}_i=[x_{i0}, x_{i1}, \dots, x_{ip-1}]$ with $i = 0, 1, 2, \dots, n-1$ and explanatory variables running from $0$ to $p-1$. See below for more explicit examples. +# +# The goal of the regression analysis is to extract/exploit relationship between $\boldsymbol{y}$ and $\boldsymbol{x}$ in order to infer specific dependencies, approximations to the likelihood functions, functional relationships and to make predictions, making fits and many other things. + +# ## Regression analysis, overarching aims II +# +# Consider an experiment in which $p$ characteristics/features of $n$ samples are +# measured. The data from this experiment, for various explanatory variables $p$ are normally represented by a matrix +# $\mathbf{X}$. +# +# The matrix $\mathbf{X}$ is called the *design +# matrix*. Additional information of the samples is available in the +# form of $\boldsymbol{y}$ (also as above). The variable $\boldsymbol{y}$ is +# generally referred to as the *response variable*. The aim of +# regression analysis is to explain $\boldsymbol{y}$ in terms of +# $\boldsymbol{X}$ through a functional relationship like $y_i = +# f(\mathbf{X}_{i,\ast})$. When no prior knowledge on the form of +# $f(\cdot)$ is available, it is common to assume a linear relationship +# between $\boldsymbol{X}$ and $\boldsymbol{y}$. This assumption gives rise to +# the *linear regression model* where $\boldsymbol{\beta} = [\beta_0, \ldots, +# \beta_{p-1}]^{T}$ are the *regression parameters*. +# +# Linear regression gives us a set of analytical equations for the parameters $\beta_j$. + +# ## Examples +# In order to understand the relation among the predictors (or features or properties) $p$, the set of data $n$ and the target (outcome, output etc) $\boldsymbol{y}$, +# consider the model we discussed for describing nuclear binding energies. +# +# There we assumed that we could parametrize the data using a polynomial approximation based on the liquid drop model. +# Assuming + +# $$ +# BE(A) = a_0+a_1A+a_2A^{2/3}+a_3A^{-1/3}+a_4A^{-1}, +# $$ + +# we have five predictors, that is the intercept, the $A$ dependent term, the $A^{2/3}$ term and the $A^{-1/3}$ and $A^{-1}$ terms. +# This gives $p=0,1,2,3,4$. Furthermore we have $n$ entries for each predictor. It means that our design matrix is a +# $p\times n$ matrix $\boldsymbol{X}$. +# +# Here the predictors are based on a model we have made. A popular data set which is widely encountered in ML applications is the +# so-called [credit card default data from Taiwan](https://www.sciencedirect.com/science/article/pii/S0957417407006719?via%3Dihub). The data set contains data on $n=30000$ credit card holders with predictors like gender, marital status, age, profession, education, etc. In total there are $24$ such predictors or attributes leading to a design matrix of dimensionality $24 \times 30000$. This is however a classification problem and we will come back to it when we discuss Logistic Regression. + +# ## General linear models and linear algebra +# Before we proceed let us study a case where we aim at fitting a set of data $\boldsymbol{y}=[y_0,y_1,\dots,y_{n-1}]$. We could think of these data as a result of an experiment or a complicated numerical experiment. These data are functions of a series of variables $\boldsymbol{x}=[x_0,x_1,\dots,x_{n-1}]$, that is $y_i = y(x_i)$ with $i=0,1,2,\dots,n-1$. The variables $x_i$ could represent physical quantities like time, temperature, position etc. We assume that $y(x)$ is a smooth function. +# +# Since obtaining these data points may not be trivial, we want to use these data to fit a function which can allow us to make predictions for values of $y$ which are not in the present set. The perhaps simplest approach is to assume we can parametrize our function in terms of a polynomial of degree $n-1$ with $n$ points, that is + +# $$ +# y=y(x) \rightarrow y(x_i)=\tilde{y}_i+\epsilon_i=\sum_{j=0}^{n-1} \beta_j x_i^j+\epsilon_i, +# $$ + +# where $\epsilon_i$ is the error in our approximation. + +# ## Rewriting the fitting procedure as a linear algebra problem +# For every set of values $y_i,x_i$ we have thus the corresponding set of equations + +# $$ +# \begin{align*} +# y_0&=\beta_0+\beta_1x_0^1+\beta_2x_0^2+\dots+\beta_{n-1}x_0^{n-1}+\epsilon_0\\ +# y_1&=\beta_0+\beta_1x_1^1+\beta_2x_1^2+\dots+\beta_{n-1}x_1^{n-1}+\epsilon_1\\ +# y_2&=\beta_0+\beta_1x_2^1+\beta_2x_2^2+\dots+\beta_{n-1}x_2^{n-1}+\epsilon_2\\ +# \dots & \dots \\ +# y_{n-1}&=\beta_0+\beta_1x_{n-1}^1+\beta_2x_{n-1}^2+\dots+\beta_{n-1}x_{n-1}^{n-1}+\epsilon_{n-1}.\\ +# \end{align*} +# $$ + +# ## Rewriting the fitting procedure as a linear algebra problem, more details +# Defining the vectors + +# $$ +# \boldsymbol{y} = [y_0,y_1, y_2,\dots, y_{n-1}]^T, +# $$ + +# and + +# $$ +# \boldsymbol{\beta} = [\beta_0,\beta_1, \beta_2,\dots, \beta_{n-1}]^T, +# $$ + +# and + +# $$ +# \boldsymbol{\epsilon} = [\epsilon_0,\epsilon_1, \epsilon_2,\dots, \epsilon_{n-1}]^T, +# $$ + +# and the design matrix + +# $$ +# \boldsymbol{X}= +# \begin{bmatrix} +# 1& x_{0}^1 &x_{0}^2& \dots & \dots &x_{0}^{n-1}\\ +# 1& x_{1}^1 &x_{1}^2& \dots & \dots &x_{1}^{n-1}\\ +# 1& x_{2}^1 &x_{2}^2& \dots & \dots &x_{2}^{n-1}\\ +# \dots& \dots &\dots& \dots & \dots &\dots\\ +# 1& x_{n-1}^1 &x_{n-1}^2& \dots & \dots &x_{n-1}^{n-1}\\ +# \end{bmatrix} +# $$ + +# we can rewrite our equations as + +# $$ +# \boldsymbol{y} = \boldsymbol{X}\boldsymbol{\beta}+\boldsymbol{\epsilon}. +# $$ + +# The above design matrix is called a [Vandermonde matrix](https://en.wikipedia.org/wiki/Vandermonde_matrix). + +# ## Generalizing the fitting procedure as a linear algebra problem +# +# We are obviously not limited to the above polynomial expansions. We +# could replace the various powers of $x$ with elements of Fourier +# series or instead of $x_i^j$ we could have $\cos{(j x_i)}$ or $\sin{(j +# x_i)}$, or time series or other orthogonal functions. For every set +# of values $y_i,x_i$ we can then generalize the equations to + +# $$ +# \begin{align*} +# y_0&=\beta_0x_{00}+\beta_1x_{01}+\beta_2x_{02}+\dots+\beta_{n-1}x_{0n-1}+\epsilon_0\\ +# y_1&=\beta_0x_{10}+\beta_1x_{11}+\beta_2x_{12}+\dots+\beta_{n-1}x_{1n-1}+\epsilon_1\\ +# y_2&=\beta_0x_{20}+\beta_1x_{21}+\beta_2x_{22}+\dots+\beta_{n-1}x_{2n-1}+\epsilon_2\\ +# \dots & \dots \\ +# y_{i}&=\beta_0x_{i0}+\beta_1x_{i1}+\beta_2x_{i2}+\dots+\beta_{n-1}x_{in-1}+\epsilon_i\\ +# \dots & \dots \\ +# y_{n-1}&=\beta_0x_{n-1,0}+\beta_1x_{n-1,2}+\beta_2x_{n-1,2}+\dots+\beta_{n-1}x_{n-1,n-1}+\epsilon_{n-1}.\\ +# \end{align*} +# $$ + +# **Note that we have $p=n$ here. The matrix is symmetric. This is generally not the case!** + +# ## Generalizing the fitting procedure as a linear algebra problem +# We redefine in turn the matrix $\boldsymbol{X}$ as + +# $$ +# \boldsymbol{X}= +# \begin{bmatrix} +# x_{00}& x_{01} &x_{02}& \dots & \dots &x_{0,n-1}\\ +# x_{10}& x_{11} &x_{12}& \dots & \dots &x_{1,n-1}\\ +# x_{20}& x_{21} &x_{22}& \dots & \dots &x_{2,n-1}\\ +# \dots& \dots &\dots& \dots & \dots &\dots\\ +# x_{n-1,0}& x_{n-1,1} &x_{n-1,2}& \dots & \dots &x_{n-1,n-1}\\ +# \end{bmatrix} +# $$ + +# and without loss of generality we rewrite again our equations as + +# $$ +# \boldsymbol{y} = \boldsymbol{X}\boldsymbol{\beta}+\boldsymbol{\epsilon}. +# $$ + +# The left-hand side of this equation is kwown. Our error vector $\boldsymbol{\epsilon}$ and the parameter vector $\boldsymbol{\beta}$ are our unknow quantities. How can we obtain the optimal set of $\beta_i$ values? + +# ## Optimizing our parameters +# We have defined the matrix $\boldsymbol{X}$ via the equations + +# $$ +# \begin{align*} +# y_0&=\beta_0x_{00}+\beta_1x_{01}+\beta_2x_{02}+\dots+\beta_{n-1}x_{0n-1}+\epsilon_0\\ +# y_1&=\beta_0x_{10}+\beta_1x_{11}+\beta_2x_{12}+\dots+\beta_{n-1}x_{1n-1}+\epsilon_1\\ +# y_2&=\beta_0x_{20}+\beta_1x_{21}+\beta_2x_{22}+\dots+\beta_{n-1}x_{2n-1}+\epsilon_1\\ +# \dots & \dots \\ +# y_{i}&=\beta_0x_{i0}+\beta_1x_{i1}+\beta_2x_{i2}+\dots+\beta_{n-1}x_{in-1}+\epsilon_1\\ +# \dots & \dots \\ +# y_{n-1}&=\beta_0x_{n-1,0}+\beta_1x_{n-1,2}+\beta_2x_{n-1,2}+\dots+\beta_{n-1}x_{n-1,n-1}+\epsilon_{n-1}.\\ +# \end{align*} +# $$ + +# As we noted above, we stayed with a system with the design matrix +# $\boldsymbol{X}\in {\mathbb{R}}^{n\times n}$, that is we have $p=n$. For reasons to come later (algorithmic arguments) we will hereafter define +# our matrix as $\boldsymbol{X}\in {\mathbb{R}}^{n\times p}$, with the predictors refering to the column numbers and the entries $n$ being the row elements. + +# ## Our model for the nuclear binding energies +# +# In our [introductory notes](https://compphysics.github.io/MachineLearning/doc/pub/How2ReadData/html/How2ReadData.html) we looked at the so-called [liquid drop model](https://en.wikipedia.org/wiki/Semi-empirical_mass_formula). Let us remind ourselves about what we did by looking at the code. +# +# We restate the parts of the code we are most interested in. + +# In[36]: + + +# Common imports +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from IPython.display import display +import os + +# Where to save the figures and data files +PROJECT_ROOT_DIR = "Results" +FIGURE_ID = "Results/FigureFiles" +DATA_ID = "DataFiles/" + +if not os.path.exists(PROJECT_ROOT_DIR): + os.mkdir(PROJECT_ROOT_DIR) + +if not os.path.exists(FIGURE_ID): + os.makedirs(FIGURE_ID) + +if not os.path.exists(DATA_ID): + os.makedirs(DATA_ID) + +def image_path(fig_id): + return os.path.join(FIGURE_ID, fig_id) + +def data_path(dat_id): + return os.path.join(DATA_ID, dat_id) + +def save_fig(fig_id): + plt.savefig(image_path(fig_id) + ".png", format='png') + +infile = open(data_path("MassEval2016.dat"),'r') + + +# Read the experimental data with Pandas +Masses = pd.read_fwf(infile, usecols=(2,3,4,6,11), + names=('N', 'Z', 'A', 'Element', 'Ebinding'), + widths=(1,3,5,5,5,1,3,4,1,13,11,11,9,1,2,11,9,1,3,1,12,11,1), + header=39, + index_col=False) + +# Extrapolated values are indicated by '#' in place of the decimal place, so +# the Ebinding column won't be numeric. Coerce to float and drop these entries. +Masses['Ebinding'] = pd.to_numeric(Masses['Ebinding'], errors='coerce') +Masses = Masses.dropna() +# Convert from keV to MeV. +Masses['Ebinding'] /= 1000 + +# Group the DataFrame by nucleon number, A. +Masses = Masses.groupby('A') +# Find the rows of the grouped DataFrame with the maximum binding energy. +Masses = Masses.apply(lambda t: t[t.Ebinding==t.Ebinding.max()]) +A = Masses['A'] +Z = Masses['Z'] +N = Masses['N'] +Element = Masses['Element'] +Energies = Masses['Ebinding'] + +# Now we set up the design matrix X +X = np.zeros((len(A),5)) +X[:,0] = 1 +X[:,1] = A +X[:,2] = A**(2.0/3.0) +X[:,3] = A**(-1.0/3.0) +X[:,4] = A**(-1.0) +# Then nice printout using pandas +DesignMatrix = pd.DataFrame(X) +DesignMatrix.index = A +DesignMatrix.columns = ['1', 'A', 'A^(2/3)', 'A^(-1/3)', '1/A'] +display(DesignMatrix) + + +# With $\boldsymbol{\beta}\in {\mathbb{R}}^{p\times 1}$, it means that we will hereafter write our equations for the approximation as + +# $$ +# \boldsymbol{\tilde{y}}= \boldsymbol{X}\boldsymbol{\beta}, +# $$ + +# throughout these lectures. + +# ## Optimizing our parameters, more details +# With the above we use the design matrix to define the approximation $\boldsymbol{\tilde{y}}$ via the unknown quantity $\boldsymbol{\beta}$ as + +# $$ +# \boldsymbol{\tilde{y}}= \boldsymbol{X}\boldsymbol{\beta}, +# $$ + +# and in order to find the optimal parameters $\beta_i$ instead of solving the above linear algebra problem, we define a function which gives a measure of the spread between the values $y_i$ (which represent hopefully the exact values) and the parameterized values $\tilde{y}_i$, namely + +# $$ +# C(\boldsymbol{\beta})=\frac{1}{n}\sum_{i=0}^{n-1}\left(y_i-\tilde{y}_i\right)^2=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)^T\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)\right\}, +# $$ + +# or using the matrix $\boldsymbol{X}$ and in a more compact matrix-vector notation as + +# $$ +# C(\boldsymbol{\beta})=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}. +# $$ + +# This function is one possible way to define the so-called cost function. +# +# It is also common to define +# the function $C$ as + +# $$ +# C(\boldsymbol{\beta})=\frac{1}{2n}\sum_{i=0}^{n-1}\left(y_i-\tilde{y}_i\right)^2, +# $$ + +# since when taking the first derivative with respect to the unknown parameters $\beta$, the factor of $2$ cancels out. + +# ## Interpretations and optimizing our parameters +# +# The function + +# $$ +# C(\boldsymbol{\beta})=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}, +# $$ + +# can be linked to the variance of the quantity $y_i$ if we interpret the latter as the mean value. +# When linking (see the discussion below) with the maximum likelihood approach below, we will indeed interpret $y_i$ as a mean value + +# $$ +# y_{i}=\langle y_i \rangle = \beta_0x_{i,0}+\beta_1x_{i,1}+\beta_2x_{i,2}+\dots+\beta_{n-1}x_{i,n-1}+\epsilon_i, +# $$ + +# where $\langle y_i \rangle$ is the mean value. Keep in mind also that +# till now we have treated $y_i$ as the exact value. Normally, the +# response (dependent or outcome) variable $y_i$ the outcome of a +# numerical experiment or another type of experiment and is thus only an +# approximation to the true value. It is then always accompanied by an +# error estimate, often limited to a statistical error estimate given by +# the standard deviation discussed earlier. In the discussion here we +# will treat $y_i$ as our exact value for the response variable. +# +# In order to find the parameters $\beta_i$ we will then minimize the spread of $C(\boldsymbol{\beta})$, that is we are going to solve the problem + +# $$ +# {\displaystyle \min_{\boldsymbol{\beta}\in +# {\mathbb{R}}^{p}}}\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)^T\left(\boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)\right\}. +# $$ + +# In practical terms it means we will require + +# $$ +# \frac{\partial C(\boldsymbol{\beta})}{\partial \beta_j} = \frac{\partial }{\partial \beta_j}\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}\right)^2\right]=0, +# $$ + +# which results in + +# $$ +# \frac{\partial C(\boldsymbol{\beta})}{\partial \beta_j} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}x_{ij}\left(y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}\right)\right]=0, +# $$ + +# or in a matrix-vector form as + +# $$ +# \frac{\partial C(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right). +# $$ + +# ## Interpretations and optimizing our parameters +# We can rewrite + +# $$ +# \frac{\partial C(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right), +# $$ + +# as + +# $$ +# \boldsymbol{X}^T\boldsymbol{y} = \boldsymbol{X}^T\boldsymbol{X}\boldsymbol{\beta}, +# $$ + +# and if the matrix $\boldsymbol{X}^T\boldsymbol{X}$ is invertible we have the solution + +# $$ +# \boldsymbol{\beta} =\left(\boldsymbol{X}^T\boldsymbol{X}\right)^{-1}\boldsymbol{X}^T\boldsymbol{y}. +# $$ + +# We note also that since our design matrix is defined as $\boldsymbol{X}\in +# {\mathbb{R}}^{n\times p}$, the product $\boldsymbol{X}^T\boldsymbol{X} \in +# {\mathbb{R}}^{p\times p}$. In the above case we have that $p \ll n$, +# in our case $p=5$ meaning that we end up with inverting a small +# $5\times 5$ matrix. This is a rather common situation, in many cases we end up with low-dimensional +# matrices to invert. The methods discussed here and for many other +# supervised learning algorithms like classification with logistic +# regression or support vector machines, exhibit dimensionalities which +# allow for the usage of direct linear algebra methods such as **LU** decomposition or **Singular Value Decomposition** (SVD) for finding the inverse of the matrix +# $\boldsymbol{X}^T\boldsymbol{X}$. +# +# **Small question**: Do you think the example we have at hand here (the nuclear binding energies) can lead to problems in inverting the matrix $\boldsymbol{X}^T\boldsymbol{X}$? What kind of problems can we expect? + +# ## Some useful matrix and vector expressions +# +# See the handwritten notes at +# +# These notes will be discussed during one of the lectures. + +# ## Interpretations and optimizing our parameters +# The residuals $\boldsymbol{\epsilon}$ are in turn given by + +# $$ +# \boldsymbol{\epsilon} = \boldsymbol{y}-\boldsymbol{\tilde{y}} = \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}, +# $$ + +# and with + +# $$ +# \boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)= 0, +# $$ + +# we have + +# $$ +# \boldsymbol{X}^T\boldsymbol{\epsilon}=\boldsymbol{X}^T\left( \boldsymbol{y}-\boldsymbol{X}\boldsymbol{\beta}\right)= 0, +# $$ + +# meaning that the solution for $\boldsymbol{\beta}$ is the one which minimizes the residuals. Later we will link this with the maximum likelihood approach. +# +# Let us now return to our nuclear binding energies and simply code the above equations. + +# ## Own code for Ordinary Least Squares +# +# It is rather straightforward to implement the matrix inversion and obtain the parameters $\boldsymbol{\beta}$. After having defined the matrix $\boldsymbol{X}$ we simply need to +# write + +# In[37]: + + +# matrix inversion to find beta +beta = np.linalg.inv(X.T.dot(X)).dot(X.T).dot(Energies) +# and then make the prediction +ytilde = X @ beta + + +# Alternatively, you can use the least squares functionality in **Numpy** as + +# In[38]: + + +fit = np.linalg.lstsq(X, Energies, rcond =None)[0] +ytildenp = np.dot(fit,X.T) + + +# And finally we plot our fit with and compare with data + +# In[39]: + + +Masses['Eapprox'] = ytilde +# Generate a plot comparing the experimental with the fitted values values. +fig, ax = plt.subplots() +ax.set_xlabel(r'$A = N + Z$') +ax.set_ylabel(r'$E_\mathrm{bind}\,/\mathrm{MeV}$') +ax.plot(Masses['A'], Masses['Ebinding'], alpha=0.7, lw=2, + label='Ame2016') +ax.plot(Masses['A'], Masses['Eapprox'], alpha=0.7, lw=2, c='m', + label='Fit') +ax.legend() +save_fig("Masses2016OLS") +plt.show() + + +# ## Adding error analysis and training set up +# +# We can easily test our fit by computing the $R2$ score that we discussed in connection with the functionality of **Scikit-Learn** in the introductory slides. +# Since we are not using **Scikit-Learn** here we can define our own $R2$ function as + +# In[40]: + + +def R2(y_data, y_model): + return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2) + + +# and we would be using it as + +# In[41]: + + +print(R2(Energies,ytilde)) + + +# We can easily add our **MSE** score as + +# In[42]: + + +def MSE(y_data,y_model): + n = np.size(y_model) + return np.sum((y_data-y_model)**2)/n + +print(MSE(Energies,ytilde)) + + +# and finally the relative error as + +# In[43]: + + +def RelativeError(y_data,y_model): + return abs((y_data-y_model)/y_data) +print(RelativeError(Energies, ytilde)) + + +# ## The $\chi^2$ function +# +# Normally, the response (dependent or outcome) variable $y_i$ is the +# outcome of a numerical experiment or another type of experiment and is +# thus only an approximation to the true value. It is then always +# accompanied by an error estimate, often limited to a statistical error +# estimate given by the standard deviation discussed earlier. In the +# discussion here we will treat $y_i$ as our exact value for the +# response variable. +# +# Introducing the standard deviation $\sigma_i$ for each measurement +# $y_i$, we define now the $\chi^2$ function (omitting the $1/n$ term) +# as + +# $$ +# \chi^2(\boldsymbol{\beta})=\frac{1}{n}\sum_{i=0}^{n-1}\frac{\left(y_i-\tilde{y}_i\right)^2}{\sigma_i^2}=\frac{1}{n}\left\{\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)^T\frac{1}{\boldsymbol{\Sigma^2}}\left(\boldsymbol{y}-\boldsymbol{\tilde{y}}\right)\right\}, +# $$ + +# where the matrix $\boldsymbol{\Sigma}$ is a diagonal matrix with $\sigma_i$ as matrix elements. + +# ## The $\chi^2$ function +# +# In order to find the parameters $\beta_i$ we will then minimize the spread of $\chi^2(\boldsymbol{\beta})$ by requiring + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_j} = \frac{\partial }{\partial \beta_j}\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(\frac{y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}}{\sigma_i}\right)^2\right]=0, +# $$ + +# which results in + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_j} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}\frac{x_{ij}}{\sigma_i}\left(\frac{y_i-\beta_0x_{i,0}-\beta_1x_{i,1}-\beta_2x_{i,2}-\dots-\beta_{n-1}x_{i,n-1}}{\sigma_i}\right)\right]=0, +# $$ + +# or in a matrix-vector form as + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{A}^T\left( \boldsymbol{b}-\boldsymbol{A}\boldsymbol{\beta}\right). +# $$ + +# where we have defined the matrix $\boldsymbol{A} =\boldsymbol{X}/\boldsymbol{\Sigma}$ with matrix elements $a_{ij} = x_{ij}/\sigma_i$ and the vector $\boldsymbol{b}$ with elements $b_i = y_i/\sigma_i$. + +# ## The $\chi^2$ function +# +# We can rewrite + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \boldsymbol{\beta}} = 0 = \boldsymbol{A}^T\left( \boldsymbol{b}-\boldsymbol{A}\boldsymbol{\beta}\right), +# $$ + +# as + +# $$ +# \boldsymbol{A}^T\boldsymbol{b} = \boldsymbol{A}^T\boldsymbol{A}\boldsymbol{\beta}, +# $$ + +# and if the matrix $\boldsymbol{A}^T\boldsymbol{A}$ is invertible we have the solution + +# $$ +# \boldsymbol{\beta} =\left(\boldsymbol{A}^T\boldsymbol{A}\right)^{-1}\boldsymbol{A}^T\boldsymbol{b}. +# $$ + +# ## The $\chi^2$ function +# +# If we then introduce the matrix + +# $$ +# \boldsymbol{H} = \left(\boldsymbol{A}^T\boldsymbol{A}\right)^{-1}, +# $$ + +# we have then the following expression for the parameters $\beta_j$ (the matrix elements of $\boldsymbol{H}$ are $h_{ij}$) + +# $$ +# \beta_j = \sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}\frac{y_i}{\sigma_i}\frac{x_{ik}}{\sigma_i} = \sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}b_ia_{ik} +# $$ + +# We state without proof the expression for the uncertainty in the parameters $\beta_j$ as (we leave this as an exercise) + +# $$ +# \sigma^2(\beta_j) = \sum_{i=0}^{n-1}\sigma_i^2\left( \frac{\partial \beta_j}{\partial y_i}\right)^2, +# $$ + +# resulting in + +# $$ +# \sigma^2(\beta_j) = \left(\sum_{k=0}^{p-1}h_{jk}\sum_{i=0}^{n-1}a_{ik}\right)\left(\sum_{l=0}^{p-1}h_{jl}\sum_{m=0}^{n-1}a_{ml}\right) = h_{jj}! +# $$ + +# ## The $\chi^2$ function +# The first step here is to approximate the function $y$ with a first-order polynomial, that is we write + +# $$ +# y=y(x) \rightarrow y(x_i) \approx \beta_0+\beta_1 x_i. +# $$ + +# By computing the derivatives of $\chi^2$ with respect to $\beta_0$ and $\beta_1$ show that these are given by + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_0} = -2\left[ \frac{1}{n}\sum_{i=0}^{n-1}\left(\frac{y_i-\beta_0-\beta_1x_{i}}{\sigma_i^2}\right)\right]=0, +# $$ + +# and + +# $$ +# \frac{\partial \chi^2(\boldsymbol{\beta})}{\partial \beta_1} = -\frac{2}{n}\left[ \sum_{i=0}^{n-1}x_i\left(\frac{y_i-\beta_0-\beta_1x_{i}}{\sigma_i^2}\right)\right]=0. +# $$ + +# ## The $\chi^2$ function +# +# For a linear fit (a first-order polynomial) we don't need to invert a matrix!! +# Defining + +# $$ +# \gamma = \sum_{i=0}^{n-1}\frac{1}{\sigma_i^2}, +# $$ + +# $$ +# \gamma_x = \sum_{i=0}^{n-1}\frac{x_{i}}{\sigma_i^2}, +# $$ + +# $$ +# \gamma_y = \sum_{i=0}^{n-1}\left(\frac{y_i}{\sigma_i^2}\right), +# $$ + +# $$ +# \gamma_{xx} = \sum_{i=0}^{n-1}\frac{x_ix_{i}}{\sigma_i^2}, +# $$ + +# $$ +# \gamma_{xy} = \sum_{i=0}^{n-1}\frac{y_ix_{i}}{\sigma_i^2}, +# $$ + +# we obtain + +# $$ +# \beta_0 = \frac{\gamma_{xx}\gamma_y-\gamma_x\gamma_y}{\gamma\gamma_{xx}-\gamma_x^2}, +# $$ + +# $$ +# \beta_1 = \frac{\gamma_{xy}\gamma-\gamma_x\gamma_y}{\gamma\gamma_{xx}-\gamma_x^2}. +# $$ + +# This approach (different linear and non-linear regression) suffers +# often from both being underdetermined and overdetermined in the +# unknown coefficients $\beta_i$. A better approach is to use the +# Singular Value Decomposition (SVD) method discussed next week. + +# ## Fitting an Equation of State for Dense Nuclear Matter +# +# Before we continue, let us introduce yet another example. We are going to fit the +# nuclear equation of state using results from many-body calculations. +# The equation of state we have made available here, as function of +# density, has been derived using modern nucleon-nucleon potentials with +# [the addition of three-body +# forces](https://www.sciencedirect.com/science/article/pii/S0370157399001106). This +# time the file is presented as a standard **csv** file. +# +# The beginning of the Python code here is similar to what you have seen +# before, with the same initializations and declarations. We use also +# **pandas** again, rather extensively in order to organize our data. +# +# The difference now is that we use **Scikit-Learn's** regression tools +# instead of our own matrix inversion implementation. Furthermore, we +# sneak in **Ridge** regression (to be discussed below) which includes a +# hyperparameter $\lambda$, also to be explained below. + +# ## The code + +# In[44]: + + +# Common imports +import os +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +import matplotlib.pyplot as plt +import sklearn.linear_model as skl +from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error + +# Where to save the figures and data files +PROJECT_ROOT_DIR = "Results" +FIGURE_ID = "Results/FigureFiles" +DATA_ID = "DataFiles/" + +if not os.path.exists(PROJECT_ROOT_DIR): + os.mkdir(PROJECT_ROOT_DIR) + +if not os.path.exists(FIGURE_ID): + os.makedirs(FIGURE_ID) + +if not os.path.exists(DATA_ID): + os.makedirs(DATA_ID) + +def image_path(fig_id): + return os.path.join(FIGURE_ID, fig_id) + +def data_path(dat_id): + return os.path.join(DATA_ID, dat_id) + +def save_fig(fig_id): + plt.savefig(image_path(fig_id) + ".png", format='png') + +infile = open(data_path("EoS.csv"),'r') + +# Read the EoS data as csv file and organize the data into two arrays with density and energies +EoS = pd.read_csv(infile, names=('Density', 'Energy')) +EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce') +EoS = EoS.dropna() +Energies = EoS['Energy'] +Density = EoS['Density'] +# The design matrix now as function of various polytrops +X = np.zeros((len(Density),4)) +X[:,3] = Density**(4.0/3.0) +X[:,2] = Density +X[:,1] = Density**(2.0/3.0) +X[:,0] = 1 + +# We use now Scikit-Learn's linear regressor and ridge regressor +# OLS part +clf = skl.LinearRegression().fit(X, Energies) +ytilde = clf.predict(X) +EoS['Eols'] = ytilde +# The mean squared error +print("Mean squared error: %.2f" % mean_squared_error(Energies, ytilde)) +# Explained variance score: 1 is perfect prediction +print('Variance score: %.2f' % r2_score(Energies, ytilde)) +# Mean absolute error +print('Mean absolute error: %.2f' % mean_absolute_error(Energies, ytilde)) +print(clf.coef_, clf.intercept_) + +# The Ridge regression with a hyperparameter lambda = 0.1 +_lambda = 0.1 +clf_ridge = skl.Ridge(alpha=_lambda).fit(X, Energies) +yridge = clf_ridge.predict(X) +EoS['Eridge'] = yridge +# The mean squared error +print("Mean squared error: %.2f" % mean_squared_error(Energies, yridge)) +# Explained variance score: 1 is perfect prediction +print('Variance score: %.2f' % r2_score(Energies, yridge)) +# Mean absolute error +print('Mean absolute error: %.2f' % mean_absolute_error(Energies, yridge)) +print(clf_ridge.coef_, clf_ridge.intercept_) + +fig, ax = plt.subplots() +ax.set_xlabel(r'$\rho[\mathrm{fm}^{-3}]$') +ax.set_ylabel(r'Energy per particle') +ax.plot(EoS['Density'], EoS['Energy'], alpha=0.7, lw=2, + label='Theoretical data') +ax.plot(EoS['Density'], EoS['Eols'], alpha=0.7, lw=2, c='m', + label='OLS') +ax.plot(EoS['Density'], EoS['Eridge'], alpha=0.7, lw=2, c='g', + label='Ridge $\lambda = 0.1$') +ax.legend() +save_fig("EoSfitting") +plt.show() + + +# The above simple polynomial in density $\rho$ gives an excellent fit +# to the data. +# +# We note also that there is a small deviation between the +# standard OLS and the Ridge regression at higher densities. We discuss this in more detail +# below. + +# ## Splitting our Data in Training and Test data +# +# It is normal in essentially all Machine Learning studies to split the +# data in a training set and a test set (sometimes also an additional +# validation set). **Scikit-Learn** has an own function for this. There +# is no explicit recipe for how much data should be included as training +# data and say test data. An accepted rule of thumb is to use +# approximately $2/3$ to $4/5$ of the data as training data. We will +# postpone a discussion of this splitting to the end of these notes and +# our discussion of the so-called **bias-variance** tradeoff. Here we +# limit ourselves to repeat the above equation of state fitting example +# but now splitting the data into a training set and a test set. + +# In[45]: + + +import os +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from sklearn.model_selection import train_test_split +# Where to save the figures and data files +PROJECT_ROOT_DIR = "Results" +FIGURE_ID = "Results/FigureFiles" +DATA_ID = "DataFiles/" + +if not os.path.exists(PROJECT_ROOT_DIR): + os.mkdir(PROJECT_ROOT_DIR) + +if not os.path.exists(FIGURE_ID): + os.makedirs(FIGURE_ID) + +if not os.path.exists(DATA_ID): + os.makedirs(DATA_ID) + +def image_path(fig_id): + return os.path.join(FIGURE_ID, fig_id) + +def data_path(dat_id): + return os.path.join(DATA_ID, dat_id) + +def save_fig(fig_id): + plt.savefig(image_path(fig_id) + ".png", format='png') + +def R2(y_data, y_model): + return 1 - np.sum((y_data - y_model) ** 2) / np.sum((y_data - np.mean(y_data)) ** 2) +def MSE(y_data,y_model): + n = np.size(y_model) + return np.sum((y_data-y_model)**2)/n + +infile = open(data_path("EoS.csv"),'r') + +# Read the EoS data as csv file and organized into two arrays with density and energies +EoS = pd.read_csv(infile, names=('Density', 'Energy')) +EoS['Energy'] = pd.to_numeric(EoS['Energy'], errors='coerce') +EoS = EoS.dropna() +Energies = EoS['Energy'] +Density = EoS['Density'] +# The design matrix now as function of various polytrops +X = np.zeros((len(Density),5)) +X[:,0] = 1 +X[:,1] = Density**(2.0/3.0) +X[:,2] = Density +X[:,3] = Density**(4.0/3.0) +X[:,4] = Density**(5.0/3.0) +# We split the data in test and training data +X_train, X_test, y_train, y_test = train_test_split(X, Energies, test_size=0.2) +# matrix inversion to find beta +beta = np.linalg.inv(X_train.T.dot(X_train)).dot(X_train.T).dot(y_train) +# and then make the prediction +ytilde = X_train @ beta +print("Training R2") +print(R2(y_train,ytilde)) +print("Training MSE") +print(MSE(y_train,ytilde)) +ypredict = X_test @ beta +print("Test R2") +print(R2(y_test,ypredict)) +print("Test MSE") +print(MSE(y_test,ypredict)) + + +# ## Exercises +# +# Here are three possible exercises for week 34 + +# ## Exercise 1: Setting up various Python environments +# +# The first exercise here is of a mere technical art. We want you to have +# * git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo [GitHub facilities](https://www.uio.no/tjenester/it/maskin/filer/versjonskontroll/github.html). +# +# * Install various Python packages +# +# We will make extensive use of Python as programming language and its +# myriad of available libraries. You will find +# IPython/Jupyter notebooks invaluable in your work. You can run **R** +# codes in the Jupyter/IPython notebooks, with the immediate benefit of +# visualizing your data. You can also use compiled languages like C++, +# Rust, Fortran etc if you prefer. The focus in these lectures will be +# on Python. +# +# If you have Python installed (we recommend Python3) and you feel +# pretty familiar with installing different packages, we recommend that +# you install the following Python packages via **pip** as +# +# 1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow +# +# For **Tensorflow**, we recommend following the instructions in the text of +# [Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O'Reilly](http://shop.oreilly.com/product/0636920052289.do) +# +# We will come back to **tensorflow** later. +# +# For Python3, replace **pip** with **pip3**. +# +# For OSX users we recommend, after having installed Xcode, to +# install **brew**. Brew allows for a seamless installation of additional +# software via for example +# +# 1. brew install python3 +# +# For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +# you can use **pip** as well and simply install Python as +# +# 1. sudo apt-get install python3 (or python for Python2.7) +# +# If you don't want to perform these operations separately and venture +# into the hassle of exploring how to set up dependencies and paths, we +# recommend two widely used distrubutions which set up all relevant +# dependencies for Python, namely +# +# * [Anaconda](https://docs.anaconda.com/), +# +# which is an open source +# distribution of the Python and R programming languages for large-scale +# data processing, predictive analytics, and scientific computing, that +# aims to simplify package management and deployment. Package versions +# are managed by the package management system **conda**. +# +# * [Enthought canopy](https://www.enthought.com/product/canopy/) +# +# is a Python +# distribution for scientific and analytic computing distribution and +# analysis environment, available for free and under a commercial +# license. +# +# We recommend using **Anaconda** if you are not too familiar with setting paths in a terminal environment. + +# ## Exercise 2: making your own data and exploring scikit-learn +# +# We will generate our own dataset for a function $y(x)$ where $x \in [0,1]$ and defined by random numbers computed with the uniform distribution. The function $y$ is a quadratic polynomial in $x$ with added stochastic noise according to the normal distribution $\cal {N}(0,1)$. +# The following simple Python instructions define our $x$ and $y$ values (with 100 data points). + +# In[46]: + + +x = np.random.rand(100,1) +y = 2.0+5*x*x+0.1*np.random.randn(100,1) + + +# 1. Write your own code (following the examples under the [regression notes](https://compphysics.github.io/MachineLearning/doc/LectureNotes/_build/html/chapter1.html)) for computing the parametrization of the data set fitting a second-order polynomial. +# +# 2. Use thereafter **scikit-learn** (see again the examples in the regression slides) and compare with your own code. +# +# 3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as + +# $$ +# MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +# \sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +# $$ + +# and the $R^2$ score function. +# If $\tilde{\boldsymbol{y}}_i$ is the predicted value of the $i-th$ sample and $y_i$ is the corresponding true value, then the score $R^2$ is defined as + +# $$ +# R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +# $$ + +# where we have defined the mean value of $\boldsymbol{y}$ as + +# $$ +# \bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +# $$ + +# You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. +# Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits. + +# ## Exercise 3: Split data in test and training data +# +# In this exercise we want you to to compute the MSE for the training +# data and the test data as function of the complexity of a polynomial, +# that is the degree of a given polynomial. +# +# The aim is to reproduce Figure 2.11 of [Hastie et al](https://github.com/CompPhysics/MLErasmus/blob/master/doc/Textbooks/elementsstat.pdf). +# +# Our data is defined by $x\in [-3,3]$ with a total of for example $n=100$ data points. You should try to vary the number of data points $n$ in your analysis. + +# In[47]: + + +np.random.seed() +n = 100 +# Make data set. +x = np.linspace(-3, 3, n).reshape(-1, 1) +y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape) + + +# where $y$ is the function we want to fit with a given polynomial. + +# **a)** +# Write a first code which sets up a design matrix $X$ defined by a fifth-order polynomial and split your data set in training and test data. + +# **b)** +# Write thereafter (using either **scikit-learn** or your matrix inversion code using for example **numpy**) +# and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial. + +# **c)** +# Add now a model which allows you to make polynomials up to degree $15$. Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)? diff --git a/doc/LectureNotes/_build/jupyter_execute/week34_101_0.png b/doc/LectureNotes/_build/jupyter_execute/week34_101_0.png new file mode 100644 index 0000000000000000000000000000000000000000..d64d64e617137d401b8ecd860f91bcb5c7ef7adb GIT binary patch literal 16238 zcmeHu2{hGx*Y>IIN(pIF#*_vX>NXP+5pI-uR>?dZnH?NTp+Sm}S>~C{^9hxZ`51C= z%=64K90%Y2ci+$bKHoFF!}`{{-nG8Bby}wX@Ei7T?`vQC+RyJO$NBAP_a7)WU6>Z|+eBaI(rFh@L)(Y)lW&Yr_ld+w>IogJo zo1gpImD6Sp4z~6pJUrHaUcimEGv(>Nzf=JaIbeHR+a86YyN~?el_HUBjzWnOWt83*RBH5+<+1c!UGmzOG;U=w zsilYB!an(U;*R9rGZ#rOLgLOzzSN|m9Om)=AhVa_r2OeOawbjlbl={y2R#hoRao*6 zW20jgF>P9kS`G_ajt-jNVD&ZD4uN5z_6QC4Azx2xP{G7eSN9Dvz?ZM5DZ5dqPbZ%m z!Vf3C7pUQj(P8W^6zblgY99FEmE`(f6w2#bAab`^)@x?rxCon4Ik&Yrbse4XZE_dx zYndF%>%35L#Wk;xx%I{2(m@|4H4TlV+}vLs=lk_JIXRaw6(a8IIU)`du@UG>?Jj+1 zeXVKSc(mI}t1R2I<5;4{n!d-zXlUtZh%P0GP1NZ(O+$}@tm2`Ajqk6W&2ZXc(U1ZJ5M7+i@hONODOtd}C!w#*xna=O z3>Hzb@&qcSb~j4udz5HbeO=v0c?AWJb}*a=eJ~jfC8gJ-Z)XG68+Se*Rn{WzY_DZn z6AHCMjJ5aEGUsHKFUe1L>Y_;z^F_F=OwushPB*&e>4Bpe=M+NLSv^;On0B5E)q7B} z*6*gst!p{c{jszAqmf6z4lIz4u&{6^nKZpMrd#He$q&;e4m%S)x~6{QB^Oxz__ZPT zN_uH&53+BWY}*UI8m_LcpR7ABUAmMDt7jy(`1vSyb1?2Dckb%UkKqUn9UY6&kr6G4 z%^%I~o2JWSjh#97Yo1^#3;lU;lw3Ghqa;k`k33UdkImJB7cX8=dfo*E8R=st6FUue z$eQrF>6bi}Ir|QsY?+uwdHJ-=?(%wMXK(K~ThKL?+rZuag~2cbwotdr(6eH$a?^%* zJuxv614iN4=h&5sCRPR}eEX5)D$Uf(N|hPtSp|+Dto9M5)B6kvrNax_pQhiMjh+AJjTCKZmf-ggNhg>6T19v z;E1h|yD5JBdzdO(6c(Jk7~pwD4ooi{PDks%&GQ78PC{O@xNb*B2XZ@UG}Le^t87jM z49;WOswZF0AX^h`{v$D0472iv!@9p*duwwIZOfFYThwV3AeN*oz99`a+lfT=lsRYr z`s=UNHqI(4f$$6wE`xHd&9!;W+Se@Lh0-eA)}{g#xZiRFzk4^)z=Oln2=)#XS&cVE zb!6#_B2UJVJF<6Y?SQ3_*M|f6T^8;YSaj6S7dT9JDnfNjK|DxdM3B0+Vqw3 zZ&1c#JhQSZ$!P{m`rM%y(o|w~^-UHDtX^tr>aMP{Q4=qAp$5TqUe$oJZqgK)8nZQH zS_&TRU1;dITUL3wUf_!14#}_*n;$qr!UTtfrPkNi@9Wal)=ouS;Hgum5dZZnFYh@I zK@J@b8Gpt^xLYY*BeRXgx%UGYg#KCC*BM=NVCTxPWXrmRj?g?#E%gVgx4uOXvP{4wsIx)!L z)e7f7@xamZ6N|NMT_Ie$?xyyZ%`MxjMYs-SPjdU4&W*YYF=oxtV&T6Uzu;N#Nwn@iQC{9zhHQrxVf!=8pEN4!-(CE)tf?RXDtpG{H1 zQO~jL=}(XMkj501ltLVr45YvT_c;^Q;a|`!7Z+DYkg|Acw3yq}E8eCBvC!b)L~i}k zU#!+%9KXD>)W9QH+~1k0o9?-@4IVMH+#s81KVP|-HP#$MG%I_ePWT!ZcLq0UQ=}{E zJcoxvIOE*OfSQqB-+QmT(3&vyl?jbcQjExVoYs)L(3I0}*TAJyXcgu()6G69E#xs8 zteJnmW*;%f@Ufml>n#gnWd-8E<7ETn0}oPDt7~a#q5Bcj9~~XFE9<+&$=UjjQ+4Xu znCcc~TtXA}g_JC@B^NK_zvR|5XglGI+l(r-`Vkj=S(BWE?kiQdA8#V*yLY~sY)?j3 zK_}-X>X`Kax+44 z8N?EpoG079Wb>f4wZph^6uSb;E~T)rFrAMNo~fv+7C20*R4mnBUUIpPxBv)X*SE$5 zajn{>QHfxA6W}n~Agri^!Or!SO>b|`Ywm1s6<8Al8l%N@4c!-?HHoe1MS;U0k42!7 zZ~>QDf^i z&m;(Il-h6EpKl%ndz20Cl=JLasmaO7@awL2@eGq4J(V71(XL~sX8S92VGi<&iX9L_ zNkK6b>A-xuQF&PDR0w8KqRyS>$x%rR1};_3f~85(+3I)h3|Ph+Yu7h5b;2g#%bd}2 z7rtJ?F4bLt_24!(F}cLctEzl`Q954YIx3d)4;*Ae2his)5b5^&4^5rfhS|@k87=_= zQA$?LGigi{OmWRrPuDPSjk|@_6jEp#`tq{m=1I7XM&3^xck9+IMhR>=a%yH~tVYMj zrvp4G@nDq6`T6;n(n=BcAg{BEMk$e|Iu!Qq-TTN2xm_o`cx{7=hes3MTA}M@Vq#(e zMn=QE1|d{&V`C#NE2}fiG&&U)yRUFpH$1Q9*DqV&=NE@-&&hFzHlRxzMwi5#)7q6K z^v5IZUq^dxyCSwhM&inJHqHQ!$su#XftJ2g@V0Bm)_8FR@w~--Hpm$&t9r|X4EuwW zFPSxfSy%;$jY~^QD|l7qy%pBJH6Lnx1L}qs1;lv63lcbz+uC(i(NcHNdB1<~odl&v zbU1K`7F}6U;3O$YaGX=Y%}8b^>lWEu;^Lxl-DC~WJrr z-uuiU3{=*J7@LE`XZ-cy?N1vqm?=&cQK#?dp6#v8j#Sm;l)|IPGV9)aO6={`v+gav zd+yx1zyyBF&To?r7V8UxfeB%}rWSw#DmRL3`qIGv_N}?JCn=Igf|MPG{RO|~BjDN0 zV5omqGLXDFn+{>$atfbXDOwJ;w_#&@S6w|ZJe-k8DeA|h8(=2SRcx&i;Be($Q2=vq zAE@#kCLMuCv>f92%+f!UT28s>;dAIsO#D|I_SJbCNW-n@bv4eQ}G{-YAx30`XsrEAhwVuTBQIBys)E2 z4r}kZ+Ol1NYHDf;r+7`wM(cyYn`vONmFJZrQ^22ca&cu=jb6HZne?5v`E%9>#85kr z41J#mm&ETdaR>09LZA-~(|O1gC=`lJXoRS9mQDHM*OU2&;aV<9I28bINR#cgYu%F$ zX`0yvmo8p3vke1xng+>)V6Uyzm2fJ^0#BTt>tn5NXi&d<_X7Y^+$8v7aLsO?j3LId zV%P7Btq*wH2{F@KOK@_sFG<8@6gJ@rhwy|ezkga&?0ake&g~dCgrpkcuz=D{KNFOd z(JYH1E$)9hh&$+~m*)lr!SM!;IF)TBmG#*W0YMXShS?RX;ea{sH>}dJiNoY98%p04 z2L?vUE29-FX#R8+D!ORVFJcsvhYwTq%bZ}DmLm)xJx|d|bX3MMxk^B3Hga6j+g!7rJRJh8%W=?&3R+g+d z4IvOMAN{Ywttvvym zcL(o_v21qsewmhbsRNJ4>)JrjAwvS^zJtS%EV%Sb z`I4;nqfj@N2Xn48yl_7IzwYIK>Y@K{@JX*mTm)m0G`PTa@aFI4V*xq42;}~Inl=H3 zNGXJJcR;uXfT0ElnaQSDQn1fK^_4CS9Pu#@_G4)((8zgdtzkE8en&YzgfQCgx?Z=NFfeH6( zd;kMnl#q}Zpa>G(^K42+PY{Oz!Cw*+(|22)d5$$gJe}1!>2&cTRUdu5+HB|CeL(Yg~WhyEv4quLIeom-Bd|fmIcph-k>RAO#XE zodFP@4&DU;lRV`MPedH2jx{DBr!4HH+qxK*Qp`~|YPX^2T%uPw&S*Ctmc?h<>>}BC2TKKOK{Fd=mf{NC*QtiU_lUeX~xR(Uz}j z23sHt&+bUqw5)ovH(b=&2JQ!+)2!TxG#L^$GjcLiw%;2B8Hu<(#N$-54T3Vt}LvFtPxRPQxALjrC zU_2i3pRFVoA=?|K(XJ-n2&*pTX{m!XMhE`_H%?6X+n$s54f}UdQ2?u0TBrf0Zf!k2 zK)emIBR;#4Z|@74U;`QzZboh(Ij2c;bS8i;^4vlQq`^#08EI+q{QUft8_B>O0i~x7 zs4lo>@C{+M6CRRgnN@P$fCsi6)#z1|d*yP!zhCmwQat6yXH;FClJUd*#D1UHev=PB zKdG5TQAgqkd*JHB)c-D9`JX%f^pDoLxVbwgTH_sog;WR>-&#>^j1m7@{* zDV$$HHCag=gDK@mHEoW@fUTUdLD2raBEOn>h~0emKB0(4fObblMs8~&G-@~Z&_YCb zc-o5>2Zfnw<=TLtc?SsbQ4RbJRd&n$g;V`I&hV@EpHz~kF85)kGhwe7JP_*j&6_i{ zI3Pp1mYXHAfexvkCQSh=r2?Sd+Kdr-DUYtmc(hZGIirX}3QIazG%D`=kT|Z!m#zyK zrxc)GMSGB?l2-gYxCSI^2LRek`m-_+ARtm9l8E;Fq`vF#1NT<@;yP=<_VyZX z8jM#{Qu49*_PUX%29x9HsI0o`u)j0!VdHvPvG(qc?iPpDSCy)o}OEZ+aRv5TmXVnf#>=}H57&e@g?lvL# z>nElAzQm2?am#JW1`LGUzg9UO5gihe1gR3?k1ObjgR@BL>ao^ms_FLNwdm*fk41Nf zJ;Ox&{yE3!?DEnjk_cFB0#>69fQFj7`e=l>X;b8FP4Ka(?n^_|{j4qnyMV7UgCt|B zJvl&phhjgs06XIK^w^{Iz7qT7xw3h2$hS3yzxuZao>NfM)3cH#z}+a_Tg+Hsr8qdx zox25?<8Fk2HE?>{eoR02lMC#iuhg*}9K1t^x~BQ}_t$`cZFe1yGX2Z#*L$S&bqlTT z-n#V+yI#X~NmzJDkL9Qym&Lv;S^;X(A0LIA{;LC|PR|1ws| z5eDn=(uM;9o)^&Rg8zCBcs_`8GQC&fZJ%uDW8$KKyFwUvh!yfkJ_cU)Rt!?%;7IVR z&IaxbVZ4FjGXUcY?wann#h{-6bQ%GF9eIK{{WBs|4dCD@L0k-igMh$e@Hfhw=&iuIwF z2;Q?YKL&q0OvC<+h*Vc*y6-~P3t4Cn&J#CZ^6uR+F0GtTtAFeTCk4)>1~8~d`mc+w z7-hib<>ir}7SaB1ml(J&enFg>`&r?4@7}FP+n4o0(z;qm3-8IfdwbA%X`~JTt4M|d zh~Ju6sSjj;nz3;raED0{Uqy9he}QYyoM^jaD(1F|hkA$_@Dz}Dr-Sz&t_@hz|78t^ zE)(w$mw9pBWF0Rus(sBU?p{2Y-(*c%EV zSMI|6($doKh1R-?5dw+~dbW~(qN66tiQ~u3VLdRwA(4P=w8W>}Meq?+IF8tEYv-9v z!Y{kv#O6Qva?oREli1}+sr8%y2H={FFw7x5=qVNMB?C`q%Ll38s3$ruU~WNEq(+Avo{A19(>TM2e&E3*-efHj3;4F5@!7%bUb zIEu($0bEpI!qn8%gaBeAjXJ%X6o|sW9@U9s9{!Y0GkB2oLz|?wjG?!e92d>kVIg>rl1a000TR`wA$XT z>sx*(oae8x1SDEXNlEg7y*5Qe2Y^LxS=WI6Pz8pD|92W&y}RP}O}jad;7q7F_`6UqhH?oe=tr)H7(;4kmD5D^iP4g_RcLIN;} zCXI~3QA#vy;(CF6_G7;(#)$U`?m*^@a9NW;k>Y{lgjIC1hVliQC*{%gaBy&5CTAck z=0M6d;bLxuNlH;uWi`462GMP8Z5(pISOaUn*eC(d3l0snq^vJur^A5O{EhGvZhwLB zdwgEzv#RUHUGkw*r-gr{zIs)NCmjAW=PV!M~ez&~YAau7m0(n(OR%4CAayO1!(LRAaFe6(DE5pLMOJ8X(&;4#PmjDRF~fqYioR{;NK+;R(M4LDG9w{mm2E6 z1oO)01E*_mZ(s0eSO&n_ZKU{p7JTnujjuu=Qul)rP3^~LaHZLSzomwMk1cB82R{oov%{Yl^q{0|^-&I;EbRi=`GUi zSBg>>GxHm<{amIF6Q%N6VA|ECYZ<)S;vwIoIutP`ET|xBB6eo21LtSd09PEQ>SL`Q z!B}RtU3k4rT^aZ;KUsS6Xl)&-vqz}l5L`gRPVuS9eFqP)bE()BXY>STMcS)gm`nUCh?@l@1&fuX~5FHBe>ymLdC4c+b3e!z8|SKepKg2X`It zSztO3<8B!&PfV+;UyU#b3eK!?RX8;c+L+V39tFLmD$d@1)~l&+KSYb(T#G8MA$+Qe zTm(#V`_5D1ZDHm#R*JVv5duvlV}B;6f6>eQmt0qI#Unng5ETK2Zzbq#m|Hrg<#63# zjwe!fdHQt0lJ}pidye(bD_+~s3D{K>Xux7`Jkyec-HBUyK8)-H7qTy>uRk>`(+X*0 zEmfw?7941&&DOMLt2E5b_!a`u@J2TkPC)L5-ktfG+0;MCb-$gxgZfiJ zgY@jo<)5$E(>hyaGMDbjzNX$4InkNpNqjDzf1!-a;iSv`n)|AO$;;Xz5)u8*l=JQ<_o}dp^pa1#vqx4uK zj**;Gqx2__m>@&*I(dfj_4dX4yeH0Heaa^S)^UTnYA^bl3X=enYEUp&ZB=gft51hV zCXT&Y17<+yNHzL&NaDL{MT-KmW8yKJT+`3q_o~OX=q}HG*fbK@Eu|W-dbJ7ES7zoD ztF|x9oP<+oB9N`BI~TK=)b-GKdSVYQ6ezs!viHnv9bS~0x|i%>hRfNhaE|!?l|ejj zc7ZzB(VvCu{0*wE5i=h}w8*rpJ#( zKkY(!h4?JEsazEF@C&)e^WtS4)uG;9kPGDBcuEX9GL%GjfN!_d845ytW?CepV;#P&)G{mHm1tza%UFDo zTjNOD&ZB&3qwA)SJ+;8Fr4t(@?JI1rJSMZrdQWE-OkG{LS%6!WPb7*<&SqDrfNgxE zf&@O_mbO7YAAcN*>%dId%U{i`NcE=Cr&T#9n7UPg+44IdOO>DM6)C<=$S z3*mmJF4O^H2@_q;4!AV6J-^f3Hye&CC1tH`$h?&LSyO~Ufo_-ibdSyp85{E9G(R5Z zm&<(}M?UO)%zSeF=zQf=_}C`nF4PTgD%@MDGw)^W2DPet^&7vk?O6=bE9M+4dIP89 zxwkm({=vd#vtowVkDLo8Y-fzA1V0Jr@g0mkb%x@@o^X$`>6uX@=}d_aN6W6ZDdti3 z-XjwQ0!&=4F$Du%UvI^u!Y^~XP#4sc8>iNMqK;suD|TPLfPF0Rj=J^%_O0Hr&Eps3 zyJ(1ej}Te!lD~LJWf@+z^2mHNH_dXn730ZKz4zMlrTze)o};MecCBXLr>(z@^WHsI zJ8N{avGy}?)Fi3_iM7tD3;b;ZDZaF z`Lb>~yoWyX@qJl5&Dw}MKJRT5_e@Innd|P;;SaRuLn!e{C4Ou0;9KkdPwq%ur{CXx z`WL30%dDH;ua&=NI{OHB^Ngji#-0_rA>&|ce)@Y9+1Pygu#JGvV=bQ~ug#4}d&ilt zAD2vwWd#orSYnk@%N!>Y>nGlvi19Ys?IR_6ZfL>(<3TAdvaI5nlr84@EuGhBy80>G z((oG=GM|yf?#5qcAWUqcR%0b$x7qZLA@I z$@6W(M5JTD(5WqrbOTj#!gIwY=k*c>HOIb`Btjxpf5U~YP)ehC;^@Qla%YJSmpF&p zFG&H6HKiKAkeRzgNN(SX7nmm9K7`>yPmT-j!oTis!86>`D%#3i-X|tNK3@>EmBOq@ z5lF50HfP&^BvH)4vt_ue(3MQ*zt4~}mmOElNECW<&4uZpeZl6qS=+DXz5dA=Id9o} z8Oga+SH*@tMk>qy&^wI3YLyw|s*-7_ePTOCKcc(Y(PM2pckNY({$@s1X{b9l$cx@F zgr<(Yu&lN*6?)=fJsB@s@0!uJ)*Zse7?Rkjyd!BH?Wxq1yJt0PlcTn1zjx%kF0OW$ zXF}woq!HVTlM!3&Mf@$~w!B=&k;g<&L4gwg{hRkBVs_zpy=mAkjN5OGauw8j>dS4e zZbVzu>0b^fUt`Uj|KRue;vJcu#`6=S5w16Aj}Ho}U64ha&CEO$D%zr1-zSzA&#}RB za<@R;w0chk@4Ni9X_;{a?hQTb?S5gR~pRo7TVHXVFtKP$b%r`}}>z zI_dg({8?DkB48ziu#E-s9?SmZc3I*vJSL*&Le<=Kq4IT=dD*wbst?cicwXHs$$Dc$ zp&wx|;}|yV>eGm4-+tH^uoaVTL_Rl{(H@-Cx-s>OA?K%E;-9_Wg8Px*oJH@(8g`#z z&7e)*&u^%*XakN~g%ryUfr@t_;N2eL(~i!s*p&q3lb7dDnfe@0VLG}*S9sXl%E(ka zF=*4cT8pJ@OViHK)v@8g1X;YLPZ=rV^A?V@$guES0VhyBCK> z4E?7oAXKXJyFcyKG%(u1Dtg66UF~E2JLO3mNY#h~6d>?gBZcip>`I_4%5U`b5NVz$ zF*573X?j$2^OXQ8d3#v!wfb~u$kSdK(WiKi_&KIhqYDLPo0{xf&#&UQ@ro>0rd)Tn zj(VQ3PU!dVyr~i!kUZ*8?}jn?w8yp2m~O{BtXGqcdST@GR>;ye5x<)HLlO5{#-(ZI zVuAV5b%l+G7o9R2H6zAjs9pOMD;REyoc?9;!o%h;7Ls+VDWJBNwdlZ+>7+9aDw~6v zw|Z@K_qIl*aPn*a5W(z7UBI?<9?* z`JRfE%;Jo;&r67xW!iH8#mZ7jpX%nsiLoNB6c0P)4RHmsC!u4`8E0+xQOdGEWO$(J zJTBsjZw8z{~ba-pp3weq)M#*+qlRb3|=4%_{jpX8o% zw~0rg1M{Upk_ThfNVq^=*_mZ=O;QqF)Qk?gxkardlr5H#5jgZw#yX={#ZTIbC^1A8F}W*w^-B52F}*(3}uYi6lAITZq{L#mb7zB}ES zsj8!+Q?Y>qwx5Pc;J0S+&5V7t%qaolvY1unUMUQvPdP`dd#4_3Ld z*%LEv#hm_IEontOH8kFJb#{3CI;)@~>Y%_q0%yyuinlYvcet<9jN6~GdM^|_`deTP z!S=Sfi*Rz`QP}I#hfkSj%%!;OKGH-GK#To^t7es5d&uK{j#jz+G40o zxGpL&cKBdWSXteQ5HdrH0JHO_|K_j2SQVQtZ5sb6srgW(HY!IzuLg zwm|fTzMSGOhLdI7Oa+WRXqt~_gw77PwujY9RkuBcs1X%4X-r{*>bZL|Wzi`^blZGE1_&-G7|Y$ECF#KlRxCvT_3H0Xr1 zr^c^I^iC8o&>Xn19<9}suHosLXTZh6&u_uIbE?2~{jpcWmucF#0tuJ18{8I~BS+;u zoUq=4ugs^;-%+es6`+%7831+DWH`h3LPkRKpaH`m#$K_dN>k-qm8M1cmVd>Vz;cY- z`v~WMR$yf~c5Jme{n-dh|&!E@5u&n$jQ!^G>5!QW)3-* zuE@kvZ6fmw$&5e$!sS1`y8qC*-|!EO z)0c{HFCR37_hY3P9{FHyEa+xE-=2B9^w61ZeyIKGmosKDr94rJD>j@!`NnW@J*bjxWwBp584lz{l5+E= zBK2CWMFQlA(s%}OwZ|6p-!0KI+1y!_U`cBNA=Uo#rtk#rY3IhHnB0=n3ES)Z^sLKD z);Jc4KqJzr7n^2uwE=9Q93^_(xJi)yt?9)Y-weZDlapI)-04gVjXoKa|Mv1{u^)Zv zV%9#dvGZ;gTLWA24;|++!h0grZ`(PTmv7$GuiUbqwUzVl5`C}3R~!{| zzamvcM`HbPAj|96wTO;*-wXvFlE<9@F-l5L|F4~^5)y396;hq2Ouv!p6=*7P2=FOn zea&skS<>Wo`kjct7dI1G;c0`um_B_utgZ$~TV}pq>?bDsEC#5k}UIeNod+pHSSHBV0jL*y@t0ulXmY$@<6~7q8GDCha ziJxZ@?Ris<?YB}Ic?8k?*N+9ZtawADii!Bzbu6Iv|w(|OFaEHPUn&wrzrG)W;~ zMJXYeaf%&K4p44#6=o^MW2@-@`ijyY!CQhEemBw;-ZkVm_1_bs6*>?&5n+ z0e`O@(r&JOp~-T6$6iGFx?#g~L54pQm*s|C0|w%BfVaM;>Q0krv!`xVck#7iN8Ao} zWHWV6e4@W%an^ClAP+?i{Uk?Zhfem3v(u z-CftS*YM5vP7h|3%~j?ui&=N!T$iie{GG6-zGdk7(uQ1<#)BRuH;96Rg@brc+=ikp zaoK5h{08wH4NJOrpNUGKai-ozAGxOSP;#L}+ZN$YU)*VS8Mg@{4mVdkOsg2V-ly*k zs;OXT&4Kb-z7Y_bGwdQ$=ekI1!!^IMKM zi153w+sd_l598(cAUhC$DpQc&7(^t~_MFM!7y|>+VR8*E3Qq_RC^lR;FZWm4EFw!k z2wmL;i&N6lR1~8g9V&B}j6-Dn_G91gLSq0FL7)rb*b)sl+Y-4ug|uI=9&$E^(Y`ubpLfAekY@x`&&Ve-Zw#^A1`>{ zE6Ys_)khP3m%(Qmj~;!5UYP<=s!jV!P(FZokluD1RKjgjUD+`OuFED!qf8B*L_W;g z6zK!IX31O4U{L|>2r)XBH8K{&uw!AS84%S!gL?0!oEv@EooW4PkZU0QI2na<2sfCk=CR^Y~@L+b$ieBHu4B87wDSK7jHv#9=7p z8?=UjENa)dVW~rEZeAY0?Z6GWHfT(WgFY%Q-NGvxiZZ{zaOV3f_$}}kiNl~LnuK~E z=qmkg+0%c3@mtRImXt1jWio2H+0BW^;Gj1QdeMUCW#j@K2-Za#9%FBdVTV9H@GdGU z6AWBFP{esI1L@P+n)eJ)52Akz+uaz(dtn0i`3Ew{zg13k+$-k1cyS@ZFc5UjQYS6W z!{b2W&6T`dR^$olQVuyN4!2rgw4%4DIERR*p%H3G`9mYJR8SH|6fHP&77>R#AWMz! zQR+U6$Lx$;khlx`DZV+wD|exK0qPUb+j1&s^~|rZU?LzMe-cf6UEX=72JPxv0?(VO zScJ;sUj%+4gRUy(UrUJ7*=nIFP?%Sa4=eV#OIU&+E zac9dAij|0djZqAf5Aw-mO#|0kC)OO~T?X-PP~F2&7Hj#j6c>HR4)vcxv?8=2R8P*I zP?ULvtjfq<^gjrAY|qa|V!Di@%C(06QqK$)s1}Du{>U{(%o4LYbBeTDWiW`}To|mu z%$1BIy)9xZEmA#srU5pkljqgd)Gi7N3Qk$etXHS5Aw@zM`b8cdW#~o%%Sqi{3dLOn z_0!bst(l(#2_yChmJ?=IBj%jSm?&2hl9GJ*B#y`mk+LlN%U!3yQ-NgRa4@{|L#G?~ zJq&b-@wqOIyw9gwJvhERYb%6@xgm6dslVduYv8%kK zxjx-Me1WCr?2kNI+|Jf~JVER8GZxp(TRy~BuMQzyAjZbV*!G6{dbp~~tHEKmhd*KC ztx$EUbLMD6s3n{ZWKrXPNOxbvW4a7h>>!L;LOGgI&?b@t>32ZBfrgbMTQX;@2Hh*6 zVv?a0C9l+)5BZ($w@XgLh?0~K_vf|~-msMqC_IWm1?t?Q44|W{*v?5Ng|pIUAnEC z5x$5Q`Ugs4*$G%B=m6vQ+;$hEEYv^?57KW2$}nZzHmod0Ia25@t*;c)R0oPSL_K8& z`YL4R(Bp7;Hw2OAM=wmlxd}`lQpN;9pb|9J6m^+X#I^DMwN)er9%_A1_Orzy`OJT? z|M+j~DK~f_z6c;Xd88vv@iGrj7bsio5gpww^ztFp`Qg>CsFv@bgp>nY(Sd$Y3kwT} z2CgCy$EiSgnXjQUxp+A>$r3c|{?9Adfda;K!EWC}r3CF88b716n3 zDKPMsrO}4rZV}XNy;rNS?kFQbWy=?e7$_?K3smqOqN zkE5u%qmqq@ql>=1F+^71(bmew(aPL_+}YUP!Q950jgf9EP$&DlyG5e-19B4e1QhJJ;>HW9bn)=968$#EtS%b3GK;VJpKe@|2)I}D^!7T9X_%V##j|MEyV}(TCMf2s77piTf6vR5j?4Oc zetWelw7H(0I+l8SWvZg8x=}GAH&W}g$?m+RetncJv~9aRoGHy=Ggn^`@>JLY+XSgS z(_wYM{pNhtY0@}OGIFIaA-!&=mTv2qhlgilx~jO=W%qj?MI!C!>E+eoV18MdB6#M! z2Uw&9e7p4{DxYA|ag2A@Ic>(quHLQ+-R2sG@rVVIc~w)b7uD?yi{j~IGchF zRf3D*Pv7%@3)8FIsk_ovFOu2!#_bxvwL@3(>Jl^b+C?oSocBjU1-*2h0^Zy@BdPG)SnvUh8MlD(bv zJe_sFJ|3^Mp5+xdTkvx~hN~St(WA`Q5n&r$67sH$dRWNnB>q zysj_SlN|$-yK3p9f|0x>$ z=hz)SHI)|`iKAoJ@tpViXymwuu6QjikaW0<%)5Zdre1+-)=}xw&}JGY+jPBcqoSiz zI5VSdY-+lB(91M_w!cjMI;}?NcOr)+yZJCxp;nb#B)j{OnIx3rBrD0*V)MiGa&wAV;Z(P6_mn6=A0I3b*2c@(UBdVL`|ko*3v+W>9$*rig%GfH z<`fokptmu9_t@cC4j{4)y1ay1S+VcbZZaA!G=J*cnk?tj_6BQ@@MVf)D5z{s&HC?w%y8)h6e6WPr!B}Gj8ivU(b8dHUmYA3*;uY2wfWd~#sBMNH zpuaU$*?V?U)dao!0|;j5$a|sLcy{y1q2XZzZ~^~*`_s<-E;5_?>yzp2@co8e&%v|V zU9a(*UGLk=`OY72{HB_UsikA_{T^a*!mMX45Tj044OW9CfR)hn3Hc!*7XO~`Ws}SH zlyS26&3UJanvO9$2M6NO>C}qYHkv^0)C#I9P=_Oifpco8!ecO)HI;J4$hkI*z>zRTcBjhy7f= z%;*9K-IUu^I^NgEh}Af)rtfjOQ;5b`E&N{KSh;((alsdDw8`6Ba=7~&rJ-(N-N}?C z2Us&Ho+pzRAR>ivy6%!fXDwK~AvV_+$BA4v-#bGH(m^^=P}HjFs2tbLzrAd_9RYcX zQZlcoNL5xgltwN&N6_;^!a~cb@lT|5tDv{8?)ReVrN@$ygR3K8rH@uxsv$P+RD(y= zPW|D)N%(A6I9W0HOojJVqhS|jp-!D@M@PqYxXZz>mzKjBk{eT%`BdhAdtwgGKW-oA z)YR0}g8a^5HBk~I6o=T_!Ay<6Q(We9+c5cLB9%2Y^Q0MT-W{&=Z!%{qS~*}rO3C}~ zNhFa9xaa)%@#A{_{QNv}O0Xz4qru~JL+|z??g0kjXqs3VBCn1Yr&?H8OxbLLd)}ON z32j`?-!_6GfCvkNZhdtF`{IW#Vh5Q^Z9{fUt zIwme|+^fILxc3?A{iAIg@4B{eQje|D$Wg%^#FB~302?K(%?bG-Ci>;o!};-IJt4OS z-G<{aEnYBXd{#aOuKc_!@rFmBnn^P-F!T%#j&n7g%OdKT%e-gRoadd5qy+E zud%*uduW9eVj$X2?^D6U43V9WHcJ4T=#>8?)m| zg6YNp->#M3FU-@chu&n7Qz8dpBz!Nt$Q}*~K-#~8+U593Sh%pTu&2LYcEj$YU~0W? zu^itv`}NIf6PCv;p5DhakS)ZcUcR@+D$D(ytMn2P=+ha8@Hw~Dl#{IM7L`uEoobQ! z>m8RiFHhU&PN3M4SkE_jxXc*-8XQyuZ+Sf4bi=XaG+XO5k|7y|*p+Kwi4LWQ3+91B zBJSqqb^-3jBuU8Ym=l0Wl1IYlZ1yXX`U2$VACI2Wn5_?IZl|%09)iZLBI(8qIn+{qRCkXtFR-%&JZMyXev;=u5s=CXrlfZNh zY$|yrC8faW+v{C#OR(=FGq$IzEJ59foRY%63)do+cf{bTrT2i_;mUf$>0Hz(t}y6T zEM{%nFs4fM{RjyOUkKiUR^$j|PnYWJS(o`3aP!}3BFi`5`yx_O7}}VrDhn5wbZV6^ zPqw9d)gkJBV0y-^o35SVaQOOZ)2*k+D~)oK(4EFhYtK!s`ADjm=;$2p+Z%T<%@v-P z&XU8~sa|<`d9=M96DK7&Jv9oj`M6A!2R4F(L%D^eC5QQNMoE4z zB>9)&{aap$F#5^PZ2mV&@w~LOH+3$%s>QkuQ!axK!JNFOrlzjlhD}dh_K`YI=x>Z< zhky6FFjrAlMoi-G?*bm|ADSU|&C+rxSH*{xdx-7;B=iwPe04mwDQb9(Nz zJVbPrmky4bV+CBAum1bS;jG`rD}7_b!U@y>4GoP8un{KaOjRwkeubItU^&{RPJ+hKeC>A{ zVp<%3QOB8$WXW!a^EquOr?_wAPBnVg(a0ud^e1uAfoAb3jf~l?8>AZ@0GyJ9I6W^7 z(CS4Y8Xycp_E)oQND!^%a$JZ+y7@>JN!iFP2$)K?T#B-yVq}&Al8*zx6y%}49ChTG3S0)h8BV;D-4*ITGGChp8j%w2+cHwn6@M zM7^kU-j3AAXGIq=>M;h6~CUO6BZ3CK*ro10eh3;^r-JnIeflC)sZtdIbcV>eUhDnJ@e zNO;n-i}kJvKP)!pZ_MDh|GD#{m1&)L8yEe?S!(WilX8{#b^rTVp8@{HMR_W1(~o>=ESr3c4`8Tn z;u`PUq*^s?i*>}jfLPG(W`*)zWh^TmYQJ_vpjsorCcW7%-Dj-bJ6Eoi)bAw`AXj*p zcQ7fxRq(vW^Dq>%YWJhXw+$F)_p1%#Ux$h&;8P1?0cpvJe-EtZGK6;pEeQCP4_ueL z;tRrs6k8IE17{jn#mDp4@lF$1hJ#Q6ZvbJ8(o)GSV#|7E2?hQ4%}s|zvY=K!Wv`(} zM1zOQUJ9#mq1N=`im?7SSh1_xhK1~@uO9+v`&`CX(5}C$QHeQ_Er@vedmnYNpt2JD zS-x!sMFAPA|BdO-_Hl7z?$}c~R%k)#4-@cG)l$|fnU8y<;D$%G=S4|BNYexJe-tQ# zLDoqVkI_YROl5eex*xdRL;BmFOiLiveTj7m7ohO7TJEb;sj8&-`$pBk7%@%|L4KrM ztWtTsi20jWla-dy!u_4Rm7)G`|0QlBxT{B{jbIP{*z2vBxR*pT=Uw z#&=FyM5C-nArq++%+U>34C}es2Dgd$pvnSDgknrZa-KuY{M7T*)Tri7??T$aDN`8! z#&yGGPM0WNKz*@wZ&5W@56LjkVp;Ff#f2GQ7$A@$?9IH4D(n1@Pw7}4MB=;BJJ-vN zUE9Q)3$W9?FTc7>OFhOxEo=PpTtTfp>J4J@i?Fq0LPTlC2RXy()j*Uj7#P2}Lzlf} z)n>;Ypcqu2U@KTe_rFD4x`W-8_ve#u-36P%VYfYeykNBPZ~zPJy|D+|*Xx5gUSaCs zQsKK%pJ>M%;o){TdKNgrV8-}8xw=_$!Ko^8$7LrmE6E$iv|DhU6kX515loXOsc>Y<0{;^oJ=c;$KuA{B?3i08+1x-wHT}?e|P{W2e72x<)(7f>vG5@JwM&?)Cw|O zR++`qqJLBtxm2Ve_v~&U#cG}F!Vc;&VoRbdFSu2yb^TP$AL?9bJn2sCXYW@(zxy{S z`EEuviQ*xOa{cyhRaUBCN@2tIqSTDq;KiJBsZe*bu*+`EU=;7pl=&jns?L-wA;G)h zle*GlDca~g&LocWLF}`N?&0B{8#HoeCQb8`R$>Z>^bE3Hr@dwa6JH_~eW6m0Y1&W8+l2KkA&l%T(;%46Lr&BUQY>`vS-m5dzUFW~hFBaRayO=xcO zCy6nSqtPRtwXMLejZtQ;o2V@a-W>)hz@SQEi0Lempnv$V{Uf*Ed@)E9`n#BuOq_iy zFnME?MJlZa`W%QrLkbLr7t#Gi8%y56tSv>fACi5hqYI)^f!5>PEwmDO)0l%*ovx39!L`cVpwszwU0Jf+^azjfzEHT6QrcN z@>@mBo^Rod_!b;Fx%f+5iHc60U(GA)l@Fi1QGL;O53q)rYiDrzPIcSa z6ubE#@Ud)zoTrZr*NT3h^6Ej*!~{XG$M-}G1c0bf6)|i3Ri3R zIX2Rx$p7M$0uiAJ6g62c8^y8g)b=B4IcWBC1@a%Qa{{e%{T5amh_WawW1mk8cj}DJ z$uE8AV8B->@#=L69^$=&Xotm$C6dSF=putCPJ$T};}~BPWB8i&gZX>GhVGQYLcK%2 znc=MY&8YB(-YNp)thTDK6US&IAO^%2TY;Xa0docMYwog)1)tg#P%NQIyTmsEFYdw` z^lA)4TefQU^=-2u$%q^sj(C8&f056UC*_b!WuTNnW#uOow>hGvKk&yW z0|n{D1C6?Z&z%lFqFIP&Lew@U^F1c8>pVN^zF8U?YCn=y`Sqs2sXxKh2L-Vm-{GQL z>-#2jF)I?zME*6ODQtbAH?f;-ZRvNfgZc?sn>Afp=@^IU6cR4Z-R1gK3$Y=sFH+rO ziJu%aN>HObET5J#D=FfMl*zHg^hd1S7iC(u((l@9Cm8CMtgDRcVt#+nDd%#m=#56E ze|tB@K~{J(OA(hn()uD+hb6le{uB4_eJ4?Sn1B(W_)g3S=AgS?C#rRtqW zb=JmF(SnrOTwiWak$CpaaQzP$h9lvb9La$RBtyQQviL`g*k?h9cE+yfqts)v#z6&O*}y zH;gOzOhXI0NXoueyEDVpe?24S2LoF)zWIPO-?HHxV~k@Cy6NxJ%*qpBJ}mpSLq(Us zo`@y@6>cJPfpW`L^V4zlT5HKTw!!QtL>R58$8hx)eYaqPtZ#jNAC~^84>dYH37c>j|`lby7kIM8a?7+h2>D0S*1# zdA=K3UwW!ZBjXbU00LL%P>4*)VeY%~BZj}HFc%62?~Bg!!&>+VLoElqVi}E0{O?5= zfbUu6>|jE4P)mch+c>0_WE{uK$!~YEZ`0#cy=cex;!S%SrU< z@ICDNScm+t`zGd|KS$AN(;eSC9seV=eboLoSpQX>djfwF=dZO9@j{URJ6^R!_2XJO zkN88-fZ+whdb zR-C^&0yn>{QS;S$@wwz=4{`I1M@1hAaN<1u=$XW@tq%jx6L1I{02T#K19~HWV`HP5 zw09TFHnuAGE=z+I>X|c>Ck2vPW*bZK|J$Rd&)K|uvT48tm)6iwWIeROKajmFE`c`i zfAZmdpZc=_FQme0LfmRtH##~cr4SH$fGlyU_5?V4W8~%jQ(=HZp8ad!*RYkTUt2;8 zK{M(&(P#c-cd#%|=C5tr#?jWu5SOd5G4(Y+4@WQ7xdt(PopY$YYI~L00}m;W_?Qf` z>osxBTF2b6Vm+qZZHPb6yYhbie9=#d3?c8oYw5pK_qa}wk$rXb2HP`IU!A8X=W+1I zgX=SKP>1J-%q+6+?XJjX=|1$3_DJ|dmEyLNkjU$lXWe*i;tM>pM`jF+3euC< z!D2qjWwMBlAduDanAtc#A|S0*ceZUVCDIS;Usj%P{~GZZFJL@B%e;KzDu%*Jfeq~E zw`fJIVg^Ryx=B)STBH9qdJ)4<>p{PKm-cYu9IG2Zn}gwiH&N7vjMIgHDJHDwe@E%# zTn=ea+)vH(m1Ip&m;+Qg1;*!d+HUrQr&K8@;T_iAw@N7P$Cw^XF^Iw3+H+v%6J<3} z!JdGU7fJevoEYDpr&|oT$NB`N7@fMT>yxo19MDr|B6j)p37Fifw_~QMc^@Wn^>^R@ zm9kPN4c81n55C>(^HRFk>xxd<-@q%QtT+EI3X0b~Z~~$l{trVZ^Ut^FmMp4@%buI@ z;g_@)vWnHszf-vE=xi=J+|EFLl}z=Y%(!*Yi#7J4akiuM5QlVg$9 zKR?Li^8`8HfQ!^zbak|fek=k>w6`)xD>cZZ;0omf#6f~dA+h>J2`o+UQI+taw!d}O zzV0EUMZRTv+Sv_J z*6*jkC@rc!@#~sMto&-WVo{LNahOF#@=%_f9npygk>lfIVe-sDoi^Net)3?{%WFKe zs{PtNs$2Y8U6lT=(&k!M#ylqW+mu1*%B^2;@WDpe#3gZ_pFP0C75(_C+@nELV+#9q z`_4xTg--to@~{>s8bp_kN*UYysd~*d#jyo106*87?(W~frzB#9 z-L=5`6w&E_e)1l>Z>m;Q)o zBJ<}grz>|L!FoDXgSG{HZPU9;<(Qe}@X5+>PlVTVGG!c}4>1(NA|vy-s3izN zE8v5QSZRx;{)oW$Ox05V;gB3xq_Oi=d4sz%2_oy;@QEm_LjUN4Lz8_ZM>TD}~qw7}xQeRTPCg;;+oAM_ni`-#nh7U(T zmR!m^=R|)V9V50aCa+h{npR@9icJ)~^)Y*?AJS*25}(-zsxN!V#{FW6Y!|4vM-0^0 zYFN=WnJF1jfrSaU`pwLnHM6<%Rp*#J{k^^o%ZnRl+Nb_UKExACxcsJj;r&5pm zOn%ul+ERfQG(H65HDhFZ5-ywEp1olZ2J0pG03PTHKm1$>i1tI48k^EQr%`Pmw|)eR zb(Ai=e#nYH_uKm`<#AE}W$`b4zk+WmDf@3!FE09psn)r>@8ybvR8w9;(ZSK@K7zyF zl3!V!(YF1~eH^R=$BEP%{DXbK^_Wd}lCXH;VaBy5AsOLnI<(kn!)qUgCGLWK1CWvH z=%zEHTkiI<8FA~AL#U{9A-?;#iGk!N4NsG0fdRaAeHBYa3lpSfDk}-QKv(CXuB$BL z+ef--?;e+8G^M6{&NtD+R=ef*iC2UB-hl^-0uMx@c#@TwDe5myOnPFSUDT=hq7~d+ z5D|p-klA&~G!k{Gj3rrVn%HL$Rj{CZG)`%FHvHYIWyqk^s;{Jve@I`d%UX(pXEbP8^rY6 z+s$CmX1PrHMLL|fR<1>lU06s1XZa13JBy@Vezkr8Suo<{PZh*3p_FuLmW#1Xp?*8L zzoB)1Qo;b;r6;0LXg_-O2PKFmw>Iz8CsCs|pA;O0*6~jWX3x9z%WM2-hZ=?%h3`?^ zAgz!iPi%j@_|VLm?i8ccj?)jgnM*B+gulq5(meOm0eSvnEys$I`Vam_Cr^>fwX`%d}l zE7Jvrek9Z_QGam+#CTP>2RI)%-~N5*!2LIzqz`o6!LzG;ib2q1PH7LDt6aV|HD818 zyHsHq&Y@Ic-{hW$m4h4WTzNe|`Gw_i`svzW7PikoZ3U|t3e$E9IeLSrbE}UXtO^ov zczt8{{GT~!(Ul`3h{>>L8x!i@w*~+-L3_H+zrEdpgCPsN)4@4t ztS!V??0xLx=%7zmP7WrmSz&2%4JRJ~rr_#t>62SS1S{p<^&@`KH?yW_+bV1fog_XG zyj8faT;5i(3&oJ&;&L8A+=95HAe8Zu(`!g_v+*oD;+4c7Sv!6iBNm(y!+T2~p^*cK zLT+haBAplTjveizdv9=UUmAZRkVI>0# zb3t35v-TZ&X_AYSzSJ`aH&(F;M7WcCtFbP<7y#iF3GB#chq73^DTx3fZyiGzs##D9D zt?vNsHZc#pcL}qsrFLTL`cuHUGu{f(uM*L3J-`-&RRi4|E8Lh(5}SZ5u+Q};88pJI z+nhOFktDt{v~B^-D@#q!v+{a=kpD0QW;0)8x7?X&rws9Vci;W=%X)2!v~HxmJv}Lg zkBn(BtDj^Pjrq+L0vde>YFKnsR60UuM}B}Varzr#RE9q-Mj^=H7bw<{1=?K(V-ZDW z!s^Gx=4xf4@eE8%YCL>uNT%HpLb#yW+u$|)S~fNYn_t=0D*$XboIi42{L9x-CeH3X zcQ`vVmh!ggljnY-Gm{no)3yNs5`l`J*R{5Ipu_bBu!CxEG;#r9G$K$@{YO=$hpY5w z8<~J)&)qJ6+~q0cEU*42_;oi7kkEj8Xho6fu$-E;noOiq01Vkyfa9kvtKx&h!}#52 zXg8&MpVzP~TpMCho{g*mEHAI>-c8nkTnRqgXhKaFSjvq4{A7dlWT%#)Zc6o_%F0^n z7ZVZA{%X(Ww2n-IjqXU;lVqC_vhgGF`w?z;IOq|95%sU9sIP2RlsUrjjW3080i>e5 z8H&o-n&RluP#{s-&Q9uZzYhgmc`iPt_SbQCu&c3E_`T`37{_1=`xSBaNq~ppv$&DX zXUKR~Yz{W#yBJAtdIKoY^Gzq~$c6(w1vSHKYVmB_caUOpc@qfvgM_25pT>Lg=a+cm zg~jPmW${u_rE@zi3BlG5qCM++fzx$Wy5~B|?FmD-S<~~b2;F$4DES5`1*`Wpz!p1Z zb<(^lu+NK=OejAksL#>~XGSBhMZYv0;xE$%_$BAAO{oaI(7c%xVKd?2RMPCB;MX15 z5wm3E#w_$0ep!{L2$CpmowMvE6j1jTiJZM*_Z`u&H^+>rDl{N4^i)u^(tP}H|O!+a)}7mPt2rP+go3LYE6XcQQEFW*>Eng_B7 zYDM+?k{uul!D{DQM0Awc2w2*SX6Z?s)xk&%cOzLcLW-8gIAv4Ft__di6X=i2B3X+v ziG6;9T&+zxk+r@MQ~OW#>YfnRQvah=RYwz8KdA7KjLPRGU3ev(1CKaRm2t;l>SJAr zY}C_D0|pPz5%1C1GtV8yij{jlRMok%N*UPGitc63jaT8%!a8ZpUOzOV2T5MdRyv91 zWl`rnVW27im1$&}-PkI~$!_%#q!jd6w*Kl<>;#eg(^Z+)@-#uu3^w&Vl?q=bN4E~e z(}9QsDQ%tE^gj)As1{@jR6QR)SAGjnalbj&yHf#$jfUrgo)v(CVmhIf?A?R`V3Ko3 zmP})m8v2e>H3rb6wI|)~310!ZYJsqGF_3QR;i%li{n(wzd%bEh7$P8TetU`S?o#&> zTOJ^L3sRP^k;o9RtZcEeZSK+iLr)ucNs+n@HjTLjQyfp29yP855Je9ex80y&p8sn& z(nv@r$qMM&C!QVb>ydt;M259}MK-*U3}Ef`vAoaqNpx6Kdw5KKo(50`8%G;-6uXC~ zOk+h|x5;It?IZbIuRSk29TR5f93c`|VBpjLcO`rNgXLBVfdsi3pe@JK@YOpER)^OCy$Nd>#2VL#n}}(ub_u>DTxf1u|{RLvsWJ zt|Lg~*MLg&A!TP-vNCm9pto@;wQBV1KLeZJvQv+f3TA7h^7MAaG^65mqT6|)&%+(pg#e7Fnp7Rys~gwQ+I!L`lyTJ z`Uw(uT#}sD!5xXIt!JPjohLz8iV@CFGw<~Dr=XY=r^q(b$ewHgYR#j*t>+^Ma8JrY zTgpWO(e`6ShD1*JYJzkKP%|U=z$g%0!D9?3dkN8LDvHv+o;W#o2l8i$&clG{IP)#< zk<#6z(uC`k$HJ|z*nF1vN~6zx-!RNrXz^M~^7Xj{3klG5(RkO-nGdlY=m@cFU6W`+ zGZ)gYKT~ODJwYkH6W9I0ZO4L2UCe6VYxTJ4fB86alP{8k_6*m2x81VRba*>9jNyB% z^3aCr@(y0oQuv8stZ>2ui25g1FBb%zAlD#28BU^|KFR%R?9D%HgouWU919*BTom((Y!j&Pt8qL!vR%kIamwk88_U z9_|`yC!$gCQkB4SN$UwLZ{C95DIYYR1Jq1U3>IZ*nzE1pSVfAjZRy^^Fa|89B=Upy zt1HzLEhwo*(M4hmLa`cUUbWlSK8W0~RLwii7|x+aiA^0qQEz{e^ls5&wC`So(cA;d z5w2)A{^4RTYQvf8prq!Evt9}ud->VCb*ECFe)U?(XUs_YxXt4-6Ca{S1Ipz2GO5k` z@HzUmau};*701RHD9D)L*Q%5OfUuuf3=Pq*Qd;xzLzu;szgWLQ10Ef)>8D6tMn$Lg zZ?Y}w9ll=qiw5%R=3Nvz*Ll6Y!)B~O&y(IEvDqav5C<+}Z8Oa++okL5x}9jiQ?DC} zSEqlbE2pPrG`rU~ZvV=oEKUHMWP@7^ur2Y7MHt2`ThU{Gev!CH?J@;Uxa?0vvqQ7t zxHDl&YadJt0HyBwa+jQ?;KvG3z5wr&E;zcA*R+{GqIk(-k`Xf@Fh2JqYqCjZ(QM|^ zcvXMGRW2%^3{(sI%8<}~oD3%9aW^s7TruBEB>qSJG@-^V>Uix~LeUDmi(j1{I@Sli zq@rD{RUasL=u;6y*S72Ql5b8lhy*SNh#B$S<(Rs{WP6`=j+_8QD zyvUMTFzW&HsyeXx$06K7I~MxZ_=?028`h z-9(&HyYC_0g@*bW!VXjM@!TF^Cw2!GG9&x|PMf)cUr{eNhoza|z*1*$!i`j|4b~WF zr}BXLqt|8j{xQCQHi~hk0++fdZ-cbOCU3+V4Fh6&j)L<|PGHm4Bj5l)F>rzXVn#pS znpDcn%;a6qh;jrTFghV2J>Z!3(e=J|y1p218ar(~1$oK~)xvt_6Cs_;FF;=2hQ=9( z5aEs2?0vguRS^CJEq|cEh3-F*QsF4K^Pem=?AibZF;#Q(Tq$Yk<4~KAlz*H~=e?VN zU$DY)U2)_AA(}4wPBW7b81dlM?UADS@kYsb#a%0y`Deeo+$rB3lb3gAm1mD-DF*my z{=>YsEd$zBG$Qorzx>#N3-o`4NRb(*n-?y4(Y+NI7KG==x7W+mulR9TW!0Netx5q; ztigIUGH!JF=i0caN660YXwNu_^EfcWnSaVChmCQicMrTh!Ye>V52DRBIj5j2d)Ql~ zlb-w>z9Ll9MR~oU##%q*t>eFqx4EBVcd=CxuJ76a8oXdaHHa-h$38j3Uc>Ns7JW5; z>h=8Iq9FIRH-IPP=+pTzwiJn31^ght1u*ze zU4Ag>ySLj9$51KV{1AIwIz4|a@u^X~oy9)Bzh|L+lg~>|&_4{|qWp`#sUj0+U@GK2 znJ}z0==kc0h0=`l6P_$3RqLd7BBb_CCzQH7ORZM3f+B?78F3F);|E$Wt0BX10R&z} zrL^kL%|Hl$-ttM}A&)N=95b-#+T=YwquP69-8=A;U-Yx}J%6A5?eqI8WUB#Hx^iHs z9RsfTq#gZb(Q?g_jMUaDYwxtd~lOz1h(f>ONg6#1)S=YO|5=aSan+Jok>!R4XWjV89?E#_ zdq{lxktm=nKkiq#c~pp?=agLRQo6j7YuFmmp%X$u#dE~)khP`eJnMbEcpk96l(PDc zd48W|fLr8d3C9>-^vEN$$;c-}zbzWtHtzytICYLe*qVv$pulMk!sEL1mG0JsWsX=3zz4>=f)BRX-Z-&m#nF29{ZdzYU3yUJyHYoOkPy``lpw0n&iJeP zZPWGDVRtx^$MR{E0D~@lDA7y*JmVAhOw-iAr^Kw9Q)(tcS9>DBbL>&O9L5*vRd5Hh zQySh-d7Nu);I3o&Q+eLhINxr^z+y!4dhU=n&f&*ZKnl32g$p^3^lZRcJ3}3=rK&XS zvj!huJ77M3YDBckpxi5pCw(kSu^aBvs9z{?IO9yeCsVmK6ScR@HvfqG(xWG#N?zy2 zBd=`ij&tw*fpz>V(2xLPgC{0bmHi)uG$C5M5CN+pRf z?gO*A$>8@wz+BK~bsbzHhwUoGRq7kcj1v%X^4@sJs>%Re1)=r@)fMyollqLPUrQF^ zg++Sr+ykwB-WDpE8);4hBWJ*E9NXe?%A?kP2TQvziCO)MhyLj@>%1)=yRwg46{b8| zx_@U+^DL}(BYcmq*tq3cjUOcZcH{x;5_FHG#|Xr44^4`3O|&vS8XPHl$-GUbmXYeg zpHyd-XRXw7dn+S|8;{@^Kzib3r`V<|Uwc(#DDp%lYo`(BP!)WqnWM&@yd07D%ELiK)vav-))=X7v1+D!{CSkcgSmw=72QdzkJ6}P`EkD zmtNHQAZ&wd@HY~KnZ)>k4>GkDGO@JU=K`kj!1|OJb%;+20R7ogKOb;^vwrv){G9}} zZ`JbrvzYe;+_8<2Ah3=e`Wd=`Smgw&*do+OmoI75DtBsb zhUAr zpJ!(e#~1yD1R;M7x;_;HsAP=9CN1ArWkq>li8R)EyyP|ltRMjS+Qz7BDRzw1_JrB% zxwBkYSgj7k3N%g60jt4qAS-gU6g@+(X0Y^>+=zWnblkJg3$a{4BR(=T#@$&$lq4&H zx!<>&M7qUb<=Y#Pn6^jy7Gk^>I3qvHqQJ>+0%uw$ko9qkM*cUMVFJsL%DiIhZ4H=c zP?M7wT-PDE%0hyCV2UPa{vXxV;B~nK>n&mKqmsr$_9ZOpJJW!2wwdskFE87rFu8f?dLD3QH~a$O2W0g?g`J^H z+$0QEQORKtG<$rOO5!$SO;0o-;}NC*zZ|Ko_gEp-&xo2AWbRNuiKVomKtu1Bm!_E( z$XZ#@D`W^2@YI=J47mJZV*H|R@u6yJIEVt~92FfmzAyIhd-C(mq(GxNqj$W@tz4z8 zC;8(Lhy&O^rU$MSE}eHv1#LPP#(tUNEv6SLz}-#L`hcvi3UJHbw@X3eRL3-J{Twx% z@p_OFpw|%R11ATd3!29&GeJB2H~h@H`x1yY@y@<+P)&U|z_nPitvLzE1%NMenQBnI-b-44 z;InNM09*`cy~VE+LwI(^2M#Tt@HI8+3xeeBdkDQ)PP_^7k?xqvT=?*7t1o=j@H9G* z4~??~#nL|ApN|iEB^^+>3SPyR6`5S}7m(jCpOkG1I#*TGj9Y7SH9;Z(yA$yP;HIuJ zm!IPui6v}8&kIYoI(rhe>+F`V?g`feN^|K??tI$!t`jT|?}DVdzBANGC|TV@(DAwa zNyFK<-vpQOqm~O3;}iyCbVP?ooO*AOZeSz5J`{A`ZT=J})!yHKhz4N+@I2_vo%syM zqSAOwyc>2qkz#TyS$YiyNE}K;3ybPeWlXts`BXc55EXHHet85BI~A0>YU#JGGg73Z zXQLj;r3e}mxSX{0Dr`nL!L|{soMK;Jq;Evu4eF!`cq~2+;;RM}K8}(+{Fh1nOppl_ z;|I-e6y4m_I=wzyz0C$TK!cVZ`VzgCU-OD%0Zbxu<9mDQQ(PwxWBF}pcKWg>g&^cA z;C24vZG^4>#|6R36Z^gtLQZIfD;2~?Mm*j>UqcsfQ?b<9qC4qEkJX-!rz-g#B*J+V zByAx6SZiHVzu}Z@F}WQwg3dh~DFp}gKa0fsA9f6Nv2T{Y2W5kBwsk%ne|vZIgKPTj zSA8~o2sr`B4x0a$#_b6?080Xw3F0Y%%QwXm-FT@%+s|nB{KCU2psPiz< zf3q>@pr~8%^OW)X#2Hfv`;9P@bI7JXAE_+0CPaoT2!GZf?s?P|P8d)t6F$>bG0hd~9JqeMk^im? z-Sm&jyLLrbSsYY3{QlzD5l<(jY^!*Akrvy;0ly14Y=V$yLSrUnaJwul`?|uXjJs;O zp0ack>J~c-+?bB|TJfriJ6ObuqnAY@YiVu^!CdA$3`tZ@(I6#EtyT|$U!r%^M$KUI z4IHA?ilM*B@^Vb+JV5JeXaU|j4AgQ0`tQuodiT@U=j^^8UgYa*K`Ti;LkL&+i$I&E z=-DqdB6#tBbt5Ng4%liUz;f>SVCSE9%{M>^AJ1~=a;u;;D~ud~D`;v0&jtw{7;g@) z%D2NE?&z*|9VXVR1nVWjjYlw$42+%Q)bBZG90U{Qej1h3zUg#(A-AU8~yce4V}txCL%9L8~VwFTn{dfGvuy=fMfX+IydTz>(F}YG z%ZZG9xT=gVT?ts=h0XSg&mj<)aWFHV+3z1Gx7Oer24-p;oWLnr%QgAcLhY=Am?#nC zxI&SFtx7vA5Ps8OW+HaOPISb#MXaTVkAbUW;_&ozL{(KU_(BOxlPSFIzLa53ymDmr z{skdu3wX`fvX=l1hfWECnQabd!P1DDRvA+JDy^$Rb7wi6cGtA(bG^-OVbyy6>syk; z+mIGFPLD>1JM>3%3Zyq6nY{%L#i^6kq+bPzHWN?(6Dt7QCLqq_K(s_~NZ!|*ddEj2 z@4*?dSF<8t=l?04LDmlr5Wp*@Go*ChP>Wj8tg83cD}aK;A*90H>C7@rQ1jwQNxZ-Z zdnduU3c%M2Rz z_5N8v5twW8XWtKS>=igJw{K6}8+bQz(kpm}X*E#2yqh@-1vHt6Gl6dseYWRy3kzT& zgYf~bvh25Ti)E0u%>L{f)s$I)p$7~jT|yRrZda;sZU~6G`_2cZN?5WsLokS-hqv+5 z6YIwqn*ih8og_mokJGmQq6@e*%qqlT)o3L-$&nx#1ADrc(=JjIRqy@9aRVBrF^TG1 zOC6WLi*RisK^E@Aa;k5@w@Vy%64_MTUV`JxYEUTDV_u^|QX~8G{bzUHFu>iwFv}uG z|KnTK9RW%Std=zJ|4SE*B47mV)*voNIT|j#yoyQ;IM00Vl8X~@@E>7zU3O0n5qi@e zkR2>)Rau#<+tb5(%~QzB(z#_A1a&JYMYekz@gFE9EkDdhbbn4EQdNeA<69M~;N}`?4-E_ZU9JXOz5;_g(kGcwH4R z-pc{IJoU5cCF7NKoki7nzlYB+js5< zR5L9?qq>`|Y~UjVxMg{M#uy*S&r;ork0cN7+FH2aLt3ilfuv(#lBFu>o6-qsDRygr zC4H|jA;0vE^%<(*1PO&^zg7M06981i!4=y4ErT+5hiW({Rke!FQ0M-OW%UgH>kvhO z2uGrpbH;5fAdC~2Ib+dYrPk-DjiBQrTw#>Aw37@(UWJxb0s`SKuXo&{}{TPjba8aHtFp41gnnFIDl9gzE}{hy>tU zZpz@TEQSN=IM6S0f!O^993E&7VMPvP&OaIcs;Z@kvkf~#1)-(-#Jou*d0#01tyKiw z6cn6nnojE@Z~L3dK?ROBd7on_`4;^7?yLnerLKBHzj9$A?UokyiFG-+KI!iI%MZq! z$j|Rqv(fwA$H7mFK`;riP1q#bRkif2%W2w4eGs1~hreR$hG>x8h@+SW-kdc8FTpY0 zob04J_dYn8nSrA}wpg&E7=eu#)aMyRA8#~+2t63;s{-35bINKz29Xib&E8Aj0*(L8 zV~;KVVb!wh;YNa#pwg9Nwpv(O074;eGvqbNR0t5v^8tL+f*Ls7T>%bcVPcm`e^M!g zY14uV2M%-u9`*jS5IqKfDa+i5gvU4;0#OfTwltjevWJ4uZl`O$EttGh|AADB>Q?nn zDqX*QHUE0~K7a~AT9Qzq(1J+H zsEA4tQX$D6F=S98>)2Y5B@&Xg#S&x9Rv9BXD8?>>!4#4;Bu2*Y{_fHHo^$>G*Z*AC z`|r9+JM`M)GJo(F~=&lWoj&=`sbs)+K0?xH}IuAnDG2;{Iz=>XilA_~GnNrazu^WfO z)x%nC>Lvqx3J^C1*H%Xk`T^&qL1`BcRmQh2dWPBF`~#ExAp0#Wy6A%l;UwC}6Wn^&cN1{J;iZb-^tV&5LMz@Ad~^K@+| zg68Y2?-FeX1JqDy#oX}^AGSH)`MX0>?*0L|kzk{|^{T0DE4t*|ZBH~lXo$RZ;n6wh zpB?`Ai@0p1-wJEqPq%j2NkxC*g`=*0akBU-o+rIG&0oL+5=soV-Jsv6#-ygnU{5Vy zF#uQ0@;pJZJD-s?f!*IGsODh!84z(9^1c|9PJm)=Z+*3k5_fZ@ZTK{qH$UZajUq8p z*iI>HZ!$b0PmyuQaHrc!+Xmw$>P6kjMX$hagAzx4myv^^jxu!9Pn&XvaolZ?^MJ9l z%4el5p178|`3?zBy|V73(un9f^{u9P1X10L<4WujorQXnk{A>_RvFE^l2gTmrV?zD z9w*1F+YN?+6OO-y6rnP`y=S72<0i^DFLHx%uug`#vXd1lfeLh8&Mk)$@cGxO&x3y7 zx9$B>{k2rrV@^S(?AQ678^>5zvXS|5XMZ6xJUrF?;F{ok_8Glb$`n^-^T&~eg=Z0r zR3h?OV=Vi-WaDR&xh`C9Xo_GsFs-sf7n+1X--x*^@{NYU2^=8}1VtQoWgIQn1trTZ z7z_MVU)fn>8CCpktPzm_Qg|ek81vQXRP2o7oNi$9wl}=Td(u25q@=3E;EbQ3oFZkO zHKJT>^16D+yq9v-#X znHQ*ndQnBQqy4rlR>;Rg-dJ@q6zS0U>VVF-4_JlF&+4?LL}Psk`k!k(faXE9&d!d$ zW_cHS-}WYS;hE|fL6^&DLaW_z9Q%G*a-_)#&)incXMs(}c?7oHI{y++-D7j=X0X~y zIB~$r{;@xdjq-J0dPo|E$vzufAGs@Bd+B~2PRi=31$3U@T4u>J)Kl|}2R^{twSxS( zN{1i8zf@g&w?J@<$%-(PFVMky-et+<8i0dSLpb~t?Ac@d_~axhID>s5rG?<_yi;rf z*K$VdSg%Ub&Ug4^_h$U;PYz)svf!ouk%cW{fp{*(_zELR(J4JyC@hKeo0*l%foDEuqC8;K^v{YH4Zpag9v9 zuzsIs6~Jc}5D7!tRC+vlB}sPVfP{B%6z@%z*&H^qy>#@-n=INl zPYIRAzKT}{wS!xWe}{2P#ab*+XV3SK(gCBc zZgmED?}uOdo#M0h<`T#<67knk_?MRxAlGL{;*yj_*tER}vQMB~L^V;FtQAGYMoX5q zM;(GVN6ok8w;?C6dE${vnicNdlKRsdyq1^PY&E0ygW80{&V?40tLo$AwS>dfMF^Ar z%0P^A;gTw3vFtZo8xl$0$+CymdVRI=olZn_0a{&iw1HU$qVahyu^2P+qq(EN$VJ${od!Z<)-pgmmR zZk$etWy9?Ed#Rx!#h33T)HR;T~u2$Bbp z(Rg|TTTQx|1z*WeiW|;1N)kI)6#Vb_Wra7o1b^N3#I0K*40S2Ch6@%kX&jVgxmD)4 zUx0Kxjnj6|)_1`$$6V)oXa8*0w|gS~63-3Z*w0fhh8rZhyy_g<|2Pw_F0C41INbro z=7e}`qo_~!Qpbt$L>EP=R5-TT(i*El=!S)VgtBnur*1tsc8C|5j4{`d91zCt5Ku5; zFg3{2j(MAp89`_l8NRp7Z zvdaRbk?i`)OCFYXH2b;4^t!L;y%(Si~>PGA~!|o56jNhd#G3e9pH80V#p$sDYTA;*e!QjL03i z=6RR>H8nLs7r#pbp&s>3OtJtzlxFw-^a;}RQ2-BxZB>B06CWR6*K6nWn0@9+!JkDn zrW)Lue&<~ZXa5>&yVMvxgH$`P$P)dl#+6?%yTAE zd}g>+9f*1WGb+)Qw-UDi!V|(!SuUnfvPgqLvc`bAK|ned-7MDtRNjP+yqAQZ-#MvT zEo*is9DC&y1huiqoOncK&Mei!^CHEI3JWhOo{tsn{TY+OrLu4y)GvL{)j5LQ$mTvj ztrK0lt_nFy?b*`=*r&ZhyD2;qgk(Uf8I^e!0b-0-!{PJnFVvKtoSC={fE`i3#**Bd zMZz&-Tnia5W?I(Rd4Gs=+mMADF1hSG# z;iQu(7xT0gnd=AGF)|Hu&EjM&Ztd*9{TiBwv~ls!W1FPrFooXONK_UQ89+*G*E3LU zBU2VxO#E0!ew>Ogx#;$(mW_`m5R>(5vM(8sh~jL1S`ic$=Ha@o3Q=iFExwqm0Pu+; z?b-3g4z%{qv(U+gAsk!(Q(j%z!^(tOsSG!H0!TVd0LqbJdKCr{CU2D@y+V6C)zmFf zQd07Qj?NB^xC1u#;qN~H)OSh6R($URsv97_+5l|ESuz6}H?T7Iw_@Y9tjOhaed~mr z{^SvIWWtc~v>b4BF=4#JSD-q)!BmWpy$~QAXO?fMYyt-E3+SB612d`t>ziw%r4+ko zD>y`Kfy5PapLY;Ov7N&pm}>$o+Ka=^gL`0%gg#82pBa7Z%mKqA80v{28gqR$-f<=T zBPe;ak_`ic7&G1@8R;X??Z&P{xoA5-;4*o3%|bD`;G8x?6n|BDnK z?|!Ym&Or3J*ex(zM;;FK^zK|`-1ozq+&S}GF>b872ufBV` zM;?_gr25PjBo`%bmxIyy=6IKWWZ^Zsa#3+{vAHim+lD}q)(qU&bih2yD=6R_5)KCv zTvm8NZ$~IGB47#(OmFH3);!H=5rk&<1nPFEUV(1EG*K18kHBscLW+I-69f_Pdn5DC zLFd6UK(4I=%p`}(+=`o1oG?I~1K=D)du&4wbZ&WIoxZG-X$=Q}u#5M)AXUKPA-HHj zz$%TF?2_J002+?^a|lORnVL-t{7`_%%un?&Y#qX{)oh5xe=(!}H4kH_U zGp57L%*=c*gPaav*HiQrtdVuo=trgggnM7i%-npaE)y*sX%8=;} rP~9L-fgk^MDDzjv|IYv@Yl-V>g$G^C;0+20F&d}Np3FIM`PRPx|3)cY literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/jupyter_execute/week34_74_1.png b/doc/LectureNotes/_build/jupyter_execute/week34_74_1.png new file mode 100644 index 0000000000000000000000000000000000000000..2e80ecf7790938ac7b84cbb5b09dd1d761e2ece3 GIT binary patch literal 38196 zcmbrm1yogG`#pHz(k%^_ZjhAj?oR0tP?}4JNOy@S(k%@VqNLI(DIlmc2olmr!@PXo ztXaRA|EyWFW?Zw0<5Id44A-uv0ljnmOm#>Jw4ZK6E zfhh$35%pFy_SScQ>g{LaX%A`Icq3fgydySYv7;r_j7>1!S98;J1PE5 z<{i5wIyil`Y#$o|ryCM+6E}yc*>stTbB$7gRE_sVOHc>bOqn2gdgTk9pH6+BVt=}` zQ!*3`7RkuSz75-DqQ4Rws@^;s$h@*+xq9@~M5=ZtD`99Pb4YR|_bPZPKW!}_4_%f3 zMV25kd;;Y;1YUrjFVQG?|9KG~@&D-sBhwrV1C2W!+sDTzHZc)pz@v-izXRHaZU%V! z_|OXql2DEg3Q-zj<+Qeny(=j}eNO(LYiopNWl?J6iuv?98?Yl77#Wc;?qqd!b!BOA z|8qySg&#f;IXXJ7N=Xq`5{ry*k&uumBFE5Sgwxse8+WU-4QAZG#^{s%@0U#vV1 zJ04=qU;b|9uIS{%A(GtleE+lXw)4Neg@pEY3GyT*jcg&@+V-R20_RVzzoJFGHae|& zt;cdi=f1pXqgB!_&n-&*z{V|ze_`%XL7~X}3?{(z(szR-kUTYO{qP3`v#E=7Lo=TF3M%I1tB(`5} z(;i#O1_p`=x&B0fG9;f1;n7MZF5TUpHQ)UEg#={_xpp_$f336WL4!J@ap84-`<5<0 z8!_EGZa$4m!wN02ustDYyoD^iJDf^)zS>GiJ}8e)!ls*?U_;|o4Ea;97+rIx=&3&2 z+yUFj%L36gTBaG7ueRqIe9%8$fk--U$MB(u4Y9lyCIZ^5s( zbAk&Ovc9=IT^-NO&1Le8mF4|Vx4G)^t`hg`+7S^nyOVFYDieThY376aDczzVDu!Ss zfTm;qmc;32Ai88zfEB9cT{W@7xT z5^HC+cA!GP?BH^z*7kT~)PJ{rpx2q^#ffgv$)vLXM&>ib<%y%j@d*FiM`yC{?=5!5 zU&@t7g3T{+ki^-36L3oBu@-=v@D(X?bQ_;Gm*T`S^5~GbPkQMe^X(O_Rs1;DSaQ$O zcTUQatPI|jx_?J`$33NJM8oz4Y`0e{lrG{w-4ENIK3y%C4RQ|VfR+)4@Vx6~1#7fQ z=z6YGr z=%D9O=qFhiU9TaM&Z)e7@N@XJ$X-^T^BM@4)!B!TJn261_hv6iei?0rI`(na@*x@~ ze&1UzGc-xxQNh{eKa<6bU-qf}=EKDInr$0;7-MCPskN?u%aA~H3XJDn3F@QrrH-Uv z!>DPn1^*l8c~Cj^?PH!q&yB zB6zbS1oH$|vwo}5Ekmnb25vHv=E7A=BqdXxmZX_sUT9G8iybP?EK+Rfymvm4rYb^j zKv`DL!~o@iM!fpyoQK@tgxLo>N;wI0tr0Q)1A`kd{k-C$ScY7rIpsT@v>hQWvsaU< z<`)tdU8>VQCt+)$F!=j{?>tUQA2(dD$0VpKocewwESw)^E7zSpo1J`G@+i;Ws=TCR zg?CU-g39dH8Hx1TmrU$Ab;j9x3_`ah>;XMCo@}$@_BM>JcX{5&J|vfkR7LEKk5bi2 z-LW0)Nw>E{)7;_aO&auv={@QFU9R4aVLxSN;2Zu!;xC~`aCo=)oLPWMXU3`P#A=t$ zqnIIHZLFxelp3nL`0{W-zzXUtpIQysYbJ)4S5d#>kEXBu=aQI{Sb-hZMhvN%O0}EO zW(sJg{+3tos6P9$TK;^IC-jiS+OL(Mbhi76*J`qayDpBkLb`pRIr*MhPMok*Rb)?} zyteFiWx;)dTaaUsl6%Q4dRWxv#OV=KwsU2Mn>2G&Y^1VGfD$QZ_G`ZH$DT=s)caYm zfjaw+=Ni8zj!B`%!AU$!Oa(ZFSXaAGn}gr_1@0Sv2^nX?$(~ZoPQya9`j4o@cY^6` zKK*8JXe~&pYi{nOG|A1+o$1t_{*ihxmr!vCmr;hBLoN-6^V8|wmumTE`7nK3Igj>I zYQ;T8+v;iyF?nxgkAglWuIq)|)vLc3`UUtDOP(CfO_cuIc4Ek|l2&={;D`i!p z^HZD7u-eS*YN2~bI%y-;_;0QbVLD&O7w!~eZV<_F$o*+1reWLHLBx8g<*i3sx`c35 z2z6~lOFf)tUlhMsOZ_bsDqr3H^mBW%WyE2#02d`~o=azP?e;(p-)4&m^?R2-=lnOe zXIo0ww`1yko0#@TZkfwicmJ;08HM6#VmIf%ISLvJiDTZ0=q2Q`ZXZqyd%cj-VwORU z(cyx@B3V;CMR`9q4QzyhOMXaY91?!dEAbFt+VjS{ea)?vlD3f6uTT&?UT`1|SUrcW z`jz8&qbwe*9({5vhyA{;BO|D@X+t~HfIW`nI2AR_s{u)SUW|-#c#aCCr5wZX_`qkX z?~0ztAxpIGz^aS0u?R0L$Ti~oopTV)DH2(EaVI6hzUbH7 zpYJJpu4JAL!sH0GATRYO;EA{ZK=i~VwJ6K+`&seiL zS!Cm34uAbhASCmS&+ePgOB{MrbLocc0_8h@dWKrZCaq)!<3Y{}a<>sP zeeME5S1y{T{$`+{}`>wd4*&1F_M^Fp15KMrl*%ZO7}(+uxTF(KXz zy5GG}Yag!;e$8x9+m}iuOa3}rEt7W#7s@C0;p7lasNE5_nRe8f`nkiL2!i(-w@;x0 zIF6OQTdRi-vUrqj)llg5t_Z}~WKaGOfQL|{o|Tb4?&(I;phuQ+l*H*-hd;}HL!kls z8d91&mQu>GOAKMf8jF$~y-N;)ImXi-j~oQ4EYHuWN|ms4ccmrRPbQ0n_-=G?@Cwzw zb7hD1q}BL;nNG~$rQUdVGFzOnJH#QT@v=RUJjik~e&HFVhkNh_ld`%B(kGAgQl%9I z3)Z6#YiNDz8s2-vX*p4&E8)>ZVu`b#On3!ZG_F5SHySQVoJ~@gJM5!EZ>rlU@Aki5 zs*Z!GzAw#6336n4Sthz&7rnvPXPc%F#&@_?nK5bef&p<%F_p252Z#%yDVVVH{_V0E z#QECdu?in%*`bBoL%iJn-`ao&^wH+SASMOM6cXA!em{R3S-fCR3n6IP+*UX)8z}`9 z9E<#1cQlA|EctBfiZm@pQMr~t!+$vlJFprr5*{fnD3nv@B zI1)+eIjylE*0~yEb#AS`lIFrgA2zGweP}e20$1O!s6SW|9_a=sFJg1WFgBGA8?R6= z&{gY=VG|O;w(2=yO?1!O8$;fxZELUAgCC-@DwB@pw^R~5#y6g283c~g7bl;w0a%Er zdW3}WU6D|D*m|LXVxrmaXOzjJJ)_dLq@WQ<8P=EvTC}L2AEFY9L~>jdHIM)(Y4G|% z=AH(JRoBQk^=DtGbD#{4ZYaUBJ{0CDyUcs`ruO2_E%)K#nIV?g19KNVvSb6yn2M;W z)u78o7mJo1wZeRp3`_OTBNB9B^5uk+l7fqC%PM>)MmF#%pPm&k?G zVmWyp{B^C~MBMVxOcdV94~tmbSw>dA?q~Zu0Sh_eE>rpG4BL{+Klf9}0h(Ckxo2T5FyWn*mgu`?cBY&Jvq>NU}ALV?VOJv7`Yg zB(`x*=Rg8cUAj=)q0W7OP*7pm6^i8&rQOtLe;x_xMxXsM!q8 z{XFX!r$msn3_(y{j8*Tb4B-AjXHON&I09B2PmjH|QH!kCt_8Zn^?y8(b$&VO-Vt*^ z%T4x@hhqV$KbYeoZ;FJUFD<4m(tXShM}CRqIW^N~@bR?*GLK;~!T zI^XP5AMqS+%TvRtp>X-~_|#fBgi+>|Jw=jIUHWyt?9f)wyD}9H*7Ur#>d~aF53-C* zFKz^C(ldP*&xc@%!e zhw6?ena{`LqC@QjEmEH%4OuCgaw ze>B(+I$vI1A|HBSP^@|%J_$X{nP!_b7*R;>5GHubt>qzdR0I-j&tB4#HNOp11p>PV_ey{130G5NiijS=8l8Fq0U-9T+v2!x=p*Bion%Vm%8;O$ULMQNzX4b-*JId4lG!^o4j5CEi_!VXP! zIyR7q}T)Z6Pn+)Y>k^!n--b1msLHzM!(o@lk@XG(mD1L)~^UXo_E5Iq+cZpif- zZA|;3-6MKYQImFRsUUJ!SJ(5}yGx$h9g*bYR&G(zA4{}RDWa_07&m2}=omRNokEps zf`1XO+GI&ZEF_FV7_p6FumK~ZP<;gTI8O>bc8eJB4~Wx1y3# z^z7Qg!ouL-VCd9VuYI~j^xbjDE&j%MzV!NY;~FFIbGSJ+Hs%pbD7V*aE}_*5cfo61 zMhVYUeT6aOIBD;*9HcMoIiJl;Z*f1b0C&~{N#!Y%q8#nF5E>kB2}KGMgdGwQ;nZr6 z?&5qzONA~tEDaDH9VRXeJB#MMOAnv}NZ6ilDT6;Q3kVDMY**0vQc^0t z0DmNJkMr-A!_a9C0)xI+o6I%XV@sVc!`gwI#!H#hyj@vkEref_BJ4Jg@p_19@!Pj= zEVqRiCVJp{G}%N-xoAHySCH;*=uMN@s)q+Pnc-UI1bFf-6=7t3KrHN&^Y}d~>2S<; zUZ&(1L1pb~dbulS5+=$nN2*U(6P$;tDZF2$Q`3xuOl&vevG|lPZxdZoh}i6ti>$Y| zz?ah+_W;Q+4wDQ)cUMaTCFz#&%<6BK_e1W)pN?dNky-lN44(a{vw?|;iS?RkYK}%q zW7BJAoIcf3BbICIQl)tp-+#2GK<(v>CbeJ$77?fyv6w_zEDZt0M&Ndgn_SS0xQ1mc zVhFBQGfSysjivk79zI1Pz{)}s8#GZnpHU>CrNyr?5+5|3nwt|DtmG%i+dRD2T4pK@ zCw5L5EvqIS5WmwN<2iTnk>l`^m;Z~+x~LTLKNJo{P}Vx{qwD#(?|}*SwhA?4S>%>%JYIY+?8Q?kMb6+1cWMB=ZBe**Q%cF+@d0|LpE6czEz7Q;SEZTLum-(Z;0E z%Z1h>DzTJ8`IcW$`W=QVZs{mKbTuZ7ZYgUZVj!rTMRTQkNxEKeVqhzCRrY`=rbdwS zmL;VQHg##wKQV3JXN7M>FUhSjvW|>QM#qB+0>obH4tHZ?l5y{^xbvJ{*T4QWZK(<7zYQ3&>_3->x<(#_t_w{dd?B2~81(`ry$&(3r|38H>(LruN1B;;H&(x?h z=Ir@mrK|fOK0ZE3P(p5n_iTC>EFu4w*TSgddlE3#@hM01Rg_;d5C-cJnUquvU%uMLzjn) zZ8{0p!uLt50vW)A;x`}0VdQ`vviOjQBgaXQiiT^jFiEYb@pKN+m3vBBojj7a7j3df zw>gDtfmBl=mx#}ItAuRP-h)d&7>f&|8-WWF5|EL|qF|>d45<1N;>&Rnp@}Eenq@`Y zEj$i;FNJP2q8@G3^XjW?cwG_)BR5Xz#PYk#z*hKE7VNZIZQ=6S1v+%&cz@o7th^Dm zWI6i;)Us`HkC~HmYkbz_dmJS&Q$y-)LOl`V&}%8z6`I>mYas^XUsw~7FY_4cUUd)G z?eb?c&T0h=9k1-+U@?8Pe04SxN!V+KYQeH?073}j3kEG7m)6mhNq35f5{lKZI4GV< zge)blt5Cgn>e-6s4k7}mnt^pY-2L>}&Pju<`_%MgE~ngOtLsEz+nz;`N-Ldw0$$x* z6ILg)rhmG<+o1v=f(LICpiDoiBAN~2w@20XnwAN^t*@iLm1Ry!QY&8JWi(kR^Qh2Y zoQGBQL?J#KCBDEE8aeWNUk~>sF})oOmuJHpFA8+zD}*8|8{MfZ-zqFnXp@o9;{s4< z@xu3w@EJs4j;rTb!uL=V%YNoBSk5>z9!=bBa*)z!ZTFw z-bRN$=&O*cQ?O5KQpPIv;9}`KCISSc&+{zY`|!&fvF)sGZx!F!L2Dn6*yZl<&-o8U z6VlIeU{K_CW2#e07iY+NQ7Q{-gmITY)6*)Ei3}9HLSp~{l-kF|w$-9;_0uj%sVlk_ z#`5?zp~wuZidcB(ucKMjw;cug>$*Mhn*p_yA|N*tw|tdGfeX1$lRIbt4TcH4Ih0l7X@v{+8yTm2e?t?$)4f8>>Gsp%>4XkRSQ zQ8ckEPd3O#adTijVj)KV`Zi0rFA_v&YbCvRkC*ufuBTm$UdM2Xe0wtW{)07192K{v zXA;*lX62$U#rS2Etq8@ZE_E@*P7K{7plA=I&`#gTNYrr??IU zDkx;1N`nyQ4qe~oUDMt)BZUHx`)GBZ3vM9fet#@&0e`(Oi14;)8;a5>PJ(C^p0|e$ zO%X8`%pd^bF0NP4O*(jo^5OEu&LjF8F>!ii=kzhUSr$GHkOr(5*=l;{vv7pmzp#YK z6C9OQe2q`G6c?c4<)W9E>!FC$EmM4&(+Xcv#e8~dOPabB_wBRV|?9djqN zY#Ocf##9RO8xN6fQ)f0buo5GIuOE=%6>GuQCvirr*V%I1x4SZF0WM1xG5+ zraEZRq4j1|Jc4U~F6~n0Ncpng6Rti5pK8b)2`e(Hj52;}_aJf8_%FjvQsa(~6>2gK z0Gua|`(a`Av~@s_!c9 zi} zH)1VAdQ&lXl-HNZQXnK?OdWvbG+c()QOsxJAFiXE>aoO z=W5>p;O?*k`Tf<(NvP)mus%WaVLJO&i%{wop+mF;bQg zd2p(zT#D6R;=Xo4Bf-9fb4(cHc7^CENW`fJ6`M90%;lI%=v>z~R&G?{f|!26vNhn! zN5%pjNH9Hss4?y*x`#6>u>>B{GmlFq9>oyyY#yn2-3^0gwOuuJuN^5=Iem=a%3k$j zU5;t_{w`o9ky>Br<%`&_Tr>l$RY@V49@$4GmbajoUnnxBHa`QMw88)7k%&Fccl<+0 z#vt|zohEPd%EEv!sl*zvu)bguJuc2N(z%?WMC^3C)X>eWqXd!5o9r%+E{G~4)?gCO zxCat1_j!gp&0;FI6lhRP16>?G%yp?sM8vPOu#3*`p=I%e4R9X576HJi)+3#y+~w0U z>#QbOJ}YuUhAi_+z*-y|Vt&m?j0q8>07N-Tu`3-=Jt(s1E>&D*ImlRGJ#5SXd#WUW z@L3(v#r!l|PW4W2kz;-8u5B^eNYXd=zbgShwUY3n=oRj>(N7`l(R8}b<95!-dRCX5 zk)wCsgNnc9zjOb!vhz0e6h?u4(p8nb?Vnme{3@w{LL*feTj1f57e$R0xUfkl5k8~x z_6pX)VqI1)Nx9ElEyf8*g}EpYhzYui!hwV$QtN-^XI5Xdy>5VqBmF9Wm&1VP-v%#j zAsg^)vNFHlZ}Pjc%!lByo=KoS@Sn?3O?c(@xhBl6KPqA^6oe`;LtQkaXH9Vc$E>Sizd|yGZuC8ce<65kr8R4664X)(E-4gc+dalNj~rN zm@oLXHC6WcJ@7WbH%2mH5nZ9cIAMMD@9*s`dC%&Wo3ViAA5L*`!PI$Nq;7uAkY!bK zJjctedK(ZK>gIAB{_JA^Ll~fE^WQ5yrRn3DwM*gh+?Cb{%V=bDsDTL3P_=>oM@Ig@ z%5P(YVtRHqp(!WlR{w7p3>`6R;b#MgiAeCZh*6DcRNBMGs}qIt0M*JrYHTZv0rwJK zvURs7T{iK}7B+yHkZk_##K#o(+k&AEdmh6aQw*k_`9N=V-bx$sd(vOJrUw zvZ34Motbx)kLknH&)^cGv5 z9P}3e`d6u`;o;%PyNjQH?X&=w22S8O+^yMW?yS+Y1`v{y){BJO5wzdhH0-)SBing~ zIb-4cE6|?VE%bI&4fm-nJMSB_*SWpFm%pRjr`*({Nb>mO(Z)rT)G?{HKU#0|6XH>l zwH@5tqNlbL^z>3^*Hl$i0iPg05@cs*_y7BGw$AkzFbV9nWr%vQU}0e$0EQ7eDZ!k~Dkdq9rNqV=`@@IwYHIQ4Ucte)1!vd5 z+tGiTcWY89ZAMfu zS0l5!^t|L?#&tpqwc{(`yF>uzJGs4y}0^!akZ9+$@e;*G&gkF`>;901MSmptz5HZ zLZk})L8_~fr^?i(2;LMS0}zy{e`^j4!PVnJsNN$%4kCkNHJB&v28LXc2oVd6Nj;Ci zkd>|P0#R5V!UN9Y_j+x^$7}Y5@HM|MFAz5hfrm|7`bGd~d4R$vP>cK0F*3p+9V@GS zL!$eGRzE<~1K2(RC1pH#idA5_5_W6dVmZwEf3j=*@3;R5j@;d9&#!ZIs_62=DyljX zQESX8$|W{M*u8)pv!O``{AFv$!$MFek6RXQ!YF&&_F6@$4!I8{D1MYX?vTvVbV~y~ z-?`#lc)sQZ(W!gDQaTUW@g+|PXafHf(&ztbuuqsBm%jQje()jx#%21feJxoWF+lBi z{={D@Ml)}i6dQ=mMy2X4dMl)_O%9K+FKuHFiPwvc#7AlfXl-P?nn=Lic-(N(M{8>_ zC>O7Q%2t^EX)c@lHcJ>0o3dojH3b5>+ zsMddr*7(XKbqB$+y7NU5Z|9#yb0vhW&w;tgZ8A~RGnCl{cau=gpb_(i$9!bU^N$eQ!? zcN=jkDl0E<(^I#~6exv)wZlbIdzv5yNZ-S9upc+OJnA z?WVqX%rZ|9PW;;XZ!s(_1l<3`9J9MKpI6mi33hLo$kH@dJY2N2{UxZpU!xo+%=(;k zJ@KN`(7v0wEz(1xFu<7%m#;SqR>mZynp{XB2m_cOAR1XqB~Ivf3;Z3F&Ch2eW@45A z;G(sU$RSXU9d;Wy^yVSEu0!~wO<-^H9mF;<*TF`wn8TNZtON_6>MQ=a(8j4>%IZex z;u_-@g57$$gN+%jtbCpGQ?O+c;xj2XO;Hq5%HyUdqMi53WX5g{)qV86Iby#ryz`_s zA2O1RgfrS!(N4IpWUpxjSmjW&el<9oqzhY9ZYY0Lv*fr_Us(K@cl_%QAYJW67g;<7 zC0~E8xBt}l(3Xmcbw>^oKfH3jfR!7{3j*@+RqxmI6z3ZK(} z5cBhsC3NF>;o0|VVOc6k_~Fz7feQ#EN3MUwQ&fua4j%yjMVhi+gq}_`P8yv?FR%XL z682guas@I#h0R!X?7y5)C za1-qc+yaAA-RE88RmKyr1bo%690Og*y-hEiPs8@2Z3X{=2ZY@SBHj>z%~dg}6zL!` zN5w7R{cosM?gO!|n)RhjhN4>BTj`5^EJC~TpUjd*tlzmt=^sbnSLo>gAQ#()cNpIZfez+>Gh z1EsoO2k>c!E$@UO8-}TEW^@~3rfE{M24L8cbI-GBqJbVuEB1L)a`J=VL zqfPuboCwI~rPEe@VP#(tTbMkb2vVqF!n>(vcQ($!uAkvljV37h&EJ5M*c-V;;7B}1 zJ5wXz&LaOxc^+OBluJU({Ygt#Fa2Mlf5Y`fL_Dee_fDHS8QT&}TS73>jV_z(5z1?K zD=EeH+FoBYs)5sSIO_K@45pD96LTk07`x;`mYP%UYRmTniN4E=c(o{>eA-yOSxQe` zROsywnKF*0!}cqMvBHkrux|=lAgz6AoscEXQ3U{QL^BB)67c=Ea%1Zg|KVCQ2{X#5 zr*M2v4yfI#y8YJps{aloYO%;j3fd*(QM8}~fxG(mtG&JK0G})rlR*&b2O^K5X$Qsc z+p=%LG)df=B(+NU4#8sc4vhnl>s4*tZO*-CG!;i@d+9s}yJ$R8-U;{--=S zvnkW27(i_X!}NN%UIIK1-)wvO)H>9v_})$H8`d2V^V9F?!PJ-{A`jmkCjTP5rPbFg zp5`Q5>`sG?990uRQ;>`_^7=YG}oBN{Nsopv` zP@^t}(aboh5FRguwCwuG=x1Dx34*)vw0i>~ODnheDo)SbsUI77f9NWkI~7GUEPqT$;bG(O_d#p?xW_={5Va{EOo4^O?U+JhytwjX2V{AoW+SBxS`SIx13$^BoB2gEF{-Y z^Mj5sfkQYNK<8z^N}EAR3b9ku1U@(NpE^09>dzMvmA0PZu<25xV}JUg9fxVy{wsQ8 zblqfA%ID1NrX6hTKNcIRrY>v%iP?G;wfrL8T^Y>U^vgwk9jlfM_P`p&L*mBo=aOif zEt`2g+cQ;%oiKa|x~JIq($@s{1@U^1wSL)I_n(bJJDW?l|C;=Pap15vZ-P;s2}#x<}+rL$P#`Xb3@DVVv$^~5nx(HS2LbU)4&u0M~r_d zcB=QiA8nVYnk%PMB=9CwHh$l)wF6e!qtT4Y$<`M~zyD_iyxE-lzDY*tds~dYpP%rf zM~`@YwhiVzzj8%zO5}mEx3{lv-~uu(kqDP0$bY((=XTan``^~yn7r;LYFIwU6CBIN ziZU$4CFXS7X~X^oVOaFe^5j>{HFH6Op5vLovuNxxtU_{D)#>=ZQt+wg*DLrTGTPes zpu>b+O33*;r^}B9RN9~eq{ZMXSJ3fd?lWV^n;nzq4aAGcAv;2M(2zrPr$fZ_ZK0l| zVCvkif>?WXc&Qu&JNT{^8QA~G2fn)OBw&Uj`dqoPNYM+Hw#%lQYd~kr`Ei(ggQ*Y+JK&_3xMa8X*0>_s;d+UWLO$#RSrotED?B?$0NBoF)w@A)utq zXWNg{e$b6A|79FBt4vHz(%;)FSkA&&Fu26N6t*f}0JP~#SnK)$P#@F)fuoZ5pULpZ z@L$BDslNJ_G;>U1cacr?WZvkjbyYs$+9%@$TH9DsnT~9=J)e+k$_S$WOZ7U%Z!rM9 z>Tzwq^z9)&r$wuGud^T}?t9SvW5Xqa7J-B&e{eIA>9GO|)ZA#|gha)|f1PxCR3$`* zF~B8-Vz5DOP|**Y8KyM|FuoCZWKlf;qb?1SPgASzgPCOl^SpuwzL`6EBJViilR2U~!ev~6xruHuD_#H53W zheyfOd9Q8)+fy#I2DFn7*!o{53{pXk_+K0#;S@Q_Rqxyd`J4V*MZ*Y9a02}*=Sz1t zR_(W!J0d>Y#NTVoV*W?P`oG;1_p7t`e7_IWfzj^^)u7)7$Tu>|$}#?<`r49F_lN2C z@82T6yW~c-mPxh)$#MJahJ!8=0U4(crMg2A5n@;#XJ@1gHWIoSX8aZ_6b2(R2eoTH z%OZuBZrOd@&mYUzS??rK;^{+mM_BYyGky1}&UbK&GFj*MoltMzf^MCYDbNj}?B>Sp zyE_+_n1}<(ym;~Xv1ZO=s~04o?MM1zNc-S*{-sGq`;p*{?T4gwUb{FgSo3!A4ocSMO&7!Ty}B_m^K0v!f7GaPJo ze|&g0eFj>*qTdO2w zyMK751>Dj4^;4QN$z8@2DCG#&#<=w!-B;^wx|fW&x0h`Sm-sr`LXPZOW3Y?l;v+sz;0#A(vu8C}B; zNH$0)sN=AuR+sqQo!gX7OgaD@23AtqY~&0odqM4Bc7D$H1R0{&`FCOQjO4*JnMzh` z4Ul3d5S`QHzsKP+b9KfQq^z~~Bg;u>R9M&8_=M0{h!uC-SyU7 zxO*A1%HFS)G#q|Qo|m2aO`P&7f`i2H%hUJKpx!$<{A1p=52oH=2aXY%Q^Kz&vD_S} z{uG0XBy?y`g$zEOuodaq(N!4ePS8e-zEUsCLbQK^0SDvY0+syqAd^rlcu|Kn5_J4Gl zXxQa0w}6ic7*JJ~yAnDE9V)$7zN|b~YOlf+TvtQw&?} z!Ek=Q=xv7gq$l*0@04PSKw;9bhLlN(B0};VraWm5)NPJ8HJ|2Ko1dogKH8(x+`!4) zdroP_t-iQ;mS%l&5dD*GU)0Wn!qi3kyx6}sU_5T)w^gQi`Y>HJGkSn>yZQ50vobNE z(A`Kxd>J_-UK#dB*U!*&03rTV44qUr6grQ)COf*v4i!1S-WWOmQ`=PXzz_k}Sm9O7LDl}B2+&_bTO_OYMmOza#2e2_EG8!}XS_&iG^byjjs1zRkK^#C zeDB>}`LU#3x_pxDuF_i%O^RP-kB*+fLk|5qWeY*TU*PX?COS!3$^t+s;$5ZYzp_f1 zZas^cKtk>0M{>0@ogg@TDSj+i!@&~qc>htF>Ft}a;E{khR)Z?m}9`abyT zbX6-rp$RcgQ#*b7X~^69C|JI+g9cklB{K-14aEbc-=N7@WjQen%r|gsLO0w}A1B?r zUNv>nlNl0#k6ulbB3STuu9`t#Vpx zVc|E@%*qtNyB{HuwG8x%gqXS4j7*5|P*A#V5I;>8s+zoSrsDT>G@MU}dxGdkOIT=7L%6%=uur*PGa}J>1yEf z_cN%(%F-lx+ti!E{c_O8VK#wWfO=N`y(%U4lP}PD9?B<=$pr@oF%Cz}FbVS$`_{*9 zgU=xz(N0s_L=3o~p#gCiK1d%O-6GhuTMN|n1D>FutE;INi&XF$mFpT`&rHp#zXfM6 zEudxUld^xn&4SvD+?B^Kw~=eq)3xDU@oC?z{9c}hWwlm=f*PHe2JJ4rD&BZ1aU#C9 z(DHA|O+)7k{qXcKP+IODrv)COP_ad;8cM@`auIoaaCE@%4*?C==2{QNXKv2h^i$1WmFTMArd4&rO;Gr zI_V(snsR-Mu#7y1E|mrz4Z2CT3SugKJTVT&#ozWhCMM;hI zLgh|WtA^FH#oZw6zAxwUf1r{~0&IdpPqj>=`@dkXSp z)-;CCSGXYFNQYO1_6iotv>-+@A)>?Pvli`QO|lK79oYaJ;pfIl<05@G5WND>Gm^{< zgYuptgX%KMt43or651pnA{|H&yq{$d%4v`WPF$VY$(U`@ggN`%gXln=fSzG&(5%iEJ?BPVU{+y!gWfZl zI~wMuRvEi4Oolw9Vs0}0Bn=i#I-8ltTTF}EA@-((B0ai_+#Ec{diN$QgO8@bp29FE zpGrk0eg!%`2#Quxo;(6thOE7{MM0JkcHWW?sbbqmgIX>twibaZ(C5M4+(s1Ee+9y} zO6oSfmI3G+%P!XEGAhQ!KT>%I6FjG;@U^~=P6S-8N#Jy`wFiqqz$=xK+l&OxGYK%6 zOgpT+li77@$RIL6Ys}3 zh?vB|>12@q*Gh7-|IOm(=daS!(Sga^ajJ9$^tDrXkAkWY@3S!~avZdgOn&rVP3~v2 z2RB&aHhPThg@rB|U3GT}fc;d4OQN`pfgH>mwG|HHUe7A~aCRHgi5bT_srHIH2q%fp zsOe%7(k(3LIB zpb&Gf^K@^5mcQtYMB(brm4UKs2*WEn4&Ks2b&W|vd>TFa3fPmOP|y_JP=&JR7j``F zHEwgdJ>w6&$Gp$~x!iZ@10Ht1Mtdi+ehR*j0PEY-RK4RA%SfIi{NDE+d=NpqeAaRI z*?+WDGrJBjewVpAqz_L=y8(65(a{M7wP#Q-JEt0cN10d3{Sah+&<_~jMp^Td9T>}j ze^S_v#Klt>Bc+o<19&5DcPcfYPSBXZOL&e-?zqhIBIe)`|90#H#^rTCY%vgozepX9$? zQO0tpdL1+WQ1Oz8Y-q?N_rb6m4<)3ZOnxeSPkVmuzvu<(_J98TA#fKI6kMb1kha-wE*Xe+ zt7051enr}-iwkT*cyGtFP@wAz8D`M!6X1&yOC7r^3Pw`0KgpUlMsz7D~>-8kv(Mp9^5R4rh~pp~j?WL;%M zHA&AWpIJM@Cmc^dA`UcAfo4ayD%eL?QqXt=%2NG_R$L;1pJ@UD0@&Eu4+GoZy+cVL z^hVbF- z)xPcQdyBay_k3^`wJ9p8g!=SPVTiR`hO$sEugAu-*(?ccXu}kB7)K)AV?`KAV^6F3Q7nP(%m4Sh@^y+C?L`; zEiEnG5)uN^u#g zrOVs(zWsC6GCkwXm(DH!y~%rPf-#UN&~SUY5o6pj_pyhQme{Yk*!9c>P>n@K8)+<@BLdR@qnJaPC{3AGVp#Yz zf|1pa;(37o5Slqux=p5u>AEZxpP?PKD23=6eGp|i+aq8kZ);{7b;I{*nveZq zE|;O@zP#oD$qKZOAHJyU{}mne8UrFD+=yKL8fWaB>W|iXWqD#=r5*A^x1%~YBo>Z?#-bV;3b{QoC88K4ba9 z&-Yv=GqO^!sGBX5r97;{VB^tvELMBOwUlzqh@@rr4lHNe~Q5xngUPz ztTWD+*h)IODm~hLsx-SEKOV>Dw!}oGr^8SGaFS}zNDbU zhJ2y5sj(X`fB#woc+wA@(+ffBMcNc~bR~Ad`|@vx{Xhp)GPhPZZ3xz<&QUpz&aEq| zO6J+I4@e_X5e%A6Qmu_?cAR#KJXc1pI6UCjPhr{^z$7U*;JHnYxF#Tq4z3bSu6g#W zM3Ph>K-L?%Oz0A1U&27S^IDpIo5a5!$Bx2PIM@GP`*R(g2&eQkUqYM-#avzP^LFHk z1i4t&d#ll$`x5xeX_=Luw8Ef&#KdiBGxD|5V)|8MENP*Q{hLVtN>2#fEOE)ywKrPc zk5tQA_`Qo-|9Jug1SA+K*^$YGLx)Gh+>P?~XbwJBwDn|ywXaa?v#99FJA}$JZQ1NU z%$v8}AAPjBp098G{^Cq0v}uR%IE&(_AG(kaaR(Y9So>TzYHTNK{9n5n>fuXsOuHK4 zHf9XoEXY3;FHgSzLlh3v-%WPDY@3mYBJah1I(X4Zl)*1%Jt}r#aQ~pY1igE zxlB}*#~H{txa~=F(yS6H3<`i_5B@y&X;4GBWRg}W*!7e&i57qgQIh{JYYexzTlhOt zBKVkCRi8vS9XK(mPSz_-9=+PP>Mm}Jdv4(D*(rIDU(;|>TVd$LRCYic=OB+e&oPub zv~AQ$w^QG&;Am)Z%PvR(9!}hHaWUt=(3t{Rmp{hGIF%mqcJ=D9Jl)i$RbM61$nQqZ z@qRrFRZsQmvmtkw|KWu46|AF%I+$k8GVMbj)3R&%{%Yt-axE`~@e1Fl(5DrxYD{ED z*%THlOOH#+v*EuGn29!tJN|L1D(PCr=xrnrt4JJU29E=aG_U>k6!CiJ;i_i zij=V(xwT-ofGxCXuchN53Gb|qyE7S2n_`ku46TS>lig-HNF@9he8FS@`n{(QcNGQb zgzEiUpDEDGZ<`n}U7M1NX`St8x<*DAKFPZW6`D&dnpicJLxi~zZ&U6j@{5zp+r7Z{ zR3l1lX;*^=h&b`D5((IsKqiS!ypEk}gHL69i7YE?q%23q&UV`@e7-z5WSHc66$4wT zz83E&3A-PM&?5DR&AU*NgcdK_d)9>=c$C<=zT?WKa$)Bg%(pxY!7amMRHSM>i6TgN zr@xwZ6)YoKlgc8kX_(P1%J)ke^PiRauK3}~huU~Js!@tq&+5iCz9Zcrg*|{ES>8*x z_~g1M@3eg(l8YwM?Vx<`|4n=Q6-g14qtE5;tie8XtPuxKSwRHHJARw;)3A50U$vs} zMk1he2n{<78WN+&H0X-MKO#C;=e5>}zw?+$N!M2hVt{4W@xh^2v^qfPrTDY#d9EJfhgS4QnDUO_LZjT@u< z(-jiNalMjR{Cr`t^v?p3tlVJ^+Uwe?Yex%NIx}JMo4uG?hKhv!TM1bs%ofe1YE56= z%%)G>eXyjZr#PRu23*azp)T~kXYoyk?U$zlbbfp!G`Y@C7i_<9O_2w|t->9!5#*Aj zguACxrmw^0aQtx~?xR^@@lcPg6(?%@^Aq;rAvMzjUdGE$mI}nM)Iw=BGOoE|RMEGQIEigY4WLk!V^0*)r}F7t zF(_IAEde0{lwWG0F-Nw7)2?Fg^oOVWO|pt@Pomj=0O%2F&CjzZ52D~%;xx88llJtp z?q^9YXp3G^F=EN|>zVu3#+19gf}vX=9t%;%ydVJ!YP)v3^v4Q8z_9wAT(}u+gXAf6 zG|}=@pYonkpPy$-B3lXWvK=Xg;!VuxA5%n>dt#2*)eJ-IcBn7MYLr%UCr!HV+|Ekx z43Oh0yO#dpdTNpL(vTvnAvo@9saK~07uy)S)4Uvx@A=++kxGU!7JRHTwMZz!7GH>d z3clCMjpg=NAPbhITbuUje9kz?g8}4HXb|@y+(b34u1lF#gK%UzQ+pmA94O@Gh>Z9R zyBAg!J+~5kFWp$1&N~H;VxfMki#D{^(@H3P`x`!fY4u;X=3@Ji7jMZQboI&al8lKp zRQoau5NH8RGO_tZ-7or$`Y;Vu<_hrpgbo#~VN}`xO^S9>a*rcRz zV8zh@1kYsD*yDHEta$Ib3p86#cC#>%k09ka<&2U|ZvBi85A8>@Yl398p6cYNrlgGE zcyZaXcg;~3e-}n>*7W~@vz-2&-mB}wNkR6Q%57G~U%OmorYc+2g$Wq!_H@%=&ieYZ8*{Y8yweFiQpRi+a{+%RZ3-ZLP z(HL|d74zK*CSf^Nof(?wESKwwRbW*A^v#ZD{gj8bl-1T+GW%m-Knv zmKTYEfq}BRdRrI;55nAIl=1d_wI5cR_dyHMKrCYHFw8hp4h46!2qK%89 z0)y&rZj;n0Jn5d#HGQ`lC+3<%&>nWkye>IM8~^Px2|56WDkC?r&ex?GLIRZ_U>rU< zX@va#;fl;fm~oBE9B{kj>qpz77+Phm<2wVvF}mZu;XA9#`=d@mSM)%+T6^&}@1Np3 zC0IPX-(kQ0m=xW1E~`58G+{IU`~6qLePY?I!Enrh^SNiL{2@vEmgZgB8P*V;9Y;e4 zX79r}t?LQbu3z5)`jHr|W8n zdTzk@(>D0)(!Y-WyC^%3QX*|Lkl*o-at&P44nKdJXkO%)!_~{wLf+TrhX-|23I>j` zFCeLfoW5}--6tB0 zfEtA&W|9m9S1i9tt=rhBkfViYa*4-*T}x|g)Nze^niL-F1<;aZOBX|2&5Qd=$pSfZ z1_Q&^_l?q-`jTWKN`tm4CzdtU9@I9xYKD8Gqa7=g9!tj(Oz^ud?`g?wc=s6G9(xFt zR1=AFG#`D+_k-AU{l+nrd+%IZnzp~y1}nKON^Ck*#CBwMwkbf|ZJQMg=mCayRjro; zY!EMcz1Iy?Qwx9fNx5g1d%IqeSER z#p4yPfXi%umkFesx9PDM!gM}~dw&Ypg|{@Du5|~Gfdo{7qVEb+Vfvh{o7Q~>zXzeD zWZ(#9Sy>sFp-D6i4I{b)Y6wC9KOgnTWOe`L>x6{)UxV9Ny+tziewz=u ztUU5L_@WL7vy2X__!{{M{REq#$?x8+?(b>>i_~mux=xxdeF$t&?E`Br z3N;J})MlnK=>g$kM7A;qAsf9ulfY2f0k1%>Z>!z#S3H8X*NK{ek3}eJL{x^%n=%@m zXX2P?5R1C9%e3ZIKD>CkqK!z}a$X^z{FG-7&SCf|a%6|6*NsJmCos3-rTCrwgk(== zu&5`!INnX6^vk{0kNNNOG*&B4Rh(in?dHk3<>%wAuNAAUdJqJy4S}Th^nj>TXTZF1 z=F*w42}m-r41`d=eZ3;_l{;vK37YrQF@968{-sXBHR0mL)_3UmBq4EaXd{!I8wYap zOw*L%)6Q^`b}&jQ*iubozt-91=?~52!=sr^;>RPCn~aLU*?T|~maIZvFv>3RJrp7( zXl$~!rOv*`Of(#X;A>Q1PoRtT;SmQncElrgKlw-oz0d+MKr(;Cn8RN3G%j53BiM|8 zW6e$YOfO!3m6{PHt}}oy`&iM|srr&j7mj61P)0^63Bc~UHDd7oI(W9pR#@e!M^wlZ zA=EX4b8fsFi6i9(fFXQO5}R;O`9T(2Efj1?DY%DhG&7U&QGAo>moJ+bl(0A5UM+J# zs@6SIa_etmMvT?$)L1eW-S5GDvF}SqM}ZzsW%4kzMHwSfGHUEX-3Ee0aaFdv!^89}D8|0^(H!FCn4|5|xbmH5N`rh3bQ{+%S8DiNzWjs@hL1=IO?>E#+}O=QMev#q@)W!qK6C?Fm<`sqXCu4l}qD)l&BdmVlx`WQxMcu!rgmKyRsm|*daR; z@kYG5^>|%WdZv#L-YKHaREWCGbA~Y+=tT`2IHP%0Lw_{QY_(7nF3 zB=g#)$+6l^r$&LH_1RGx)M0?37@u8Zb}cj{KhUAIJOfuaZ9hnV_Ae6aHk|Nr#`|la zD`A6SZXxjl6KE5|$;GMW6+fh~>gUgnhfOcdXT&x>`A!4d!MA1STSddHB9g*j)Ftem z64cQS(3vEPzeOp|8SeGHkv&|STntr}XL-v<{`yi%fm@oA!aXq*2(|*?U`LE?l1BeA zws~St;~^!KF>d+ zz9D%Mn8-d7=fBMaG?7H+&)II_->D_TPB8S89=C^KqrP2~##Mj4_ptf`{2~Fa?N3T+ z8S`b~(_Jb-(HJ^tt11Srl|hE^<>6F($eo$j9tVqmxXH!;y&Z{hGLfvZjb?Zynv~(U z{@NWbRm`iYskpM+qRRQ1;&_a7tV@y@6d&TQnVn)gi8%${e}_dyq+W!O-}c+9KL*}0QM=B)@ntqfwC16VeU-2lCu;27YE=)gu=DmAR`^} zkSYeb7yiS+sfvr6cBEItq0+|iG*;`*!6J0iU?D&okDP-D68p^UuiI14RrpY>Y(FmX zx;Ec7v*pHPyl8Svar-4c!6hoRr>L|B{4SzRNys7wi!&f&W+Wfz=Z(|qZp{f#0eGiN&yv2``KRQI{GHk=^99k=tlH!y^ew({tPDeWRL z3E|)*`WJtG@fNqLuBo$;)~*>~58VSJj2UjMPvh4i_)Ny{c!j?ui?wokxCYQZRAqQ$ zsuDvK0@C32<(5`;+v1SxyzsPFU2*{ByTlHE)8N-?(xo(L=+ii?ZKa=n;xXx8Zig^MjofIHS$s6U_PA4 zu-RD&`P`FGR^w+gdq`vN8o-LJ?rrsOA z4F`$VC>!R_c&zpd)%e*bp6svV%fufBuFAhMC;xjXzY>fk%HWulcggf9wRxRhlhGj` zsVnT^O7_;{1jQTwn;HX{2B1%eN?%xWftwPYJSl}@>(A4kuV@J}RdSo#jqD;~RFKX0 z%DOfKTd+kE&IB1Q#dyPY12x|^ejbJb9 z_gZYfLw{CtDq9wGK%iWOBG7nSQ7AQIfhQBGenF+qKkx+Eb82y4*5{ku$@sDkb+MIr zgyYwuns;1m@hz+!4HrlDuHunlgG#M&_s_FG>(6&|Ztyl*s&xNRLRV5OrcGI+CfHEF z|4Qk)kbqhQ^;TD1T8t5Gq6!yHqG4&}(1z~OgUx4m7E$evi3uIcQNp4hceH25all4R zN)!S+B&;94heAGYu28GAgdFS6el06Qa&N(Rs_+YG5&fK)Q1a|KJD4GfUZ}N$Z0}EX zn@!n#tP=ArPfsaG$sH%2Rb-TxkHnyV>CK`@z$hdM3jPyViQIlm4lx zD2Umy)N=<-Sro_}7A_@!k4s@ZYfWU=?sy$jxR=#_;`Ll5iG;GBV&zqG)L2>2zB14`$-dakqLN1UEb4m~A z3k65-Ef&c(X~QeuF(!3T8sy*=S6yPb@+0>pUuRx<(F0lHtL%5rgj#!>Immv&QG zRkbCO?s~CR552IcXm}r!#B)rS8Yb{B)3DrCQ&X_B+sxbgh+Rfi_BNkIT|il z{*K7&Q>&&YAZ;}_pMT#@K|wKqP~_JUtQ#nkDd_0v^78T;K^|^vMRd-~Dwafv1kD$g$zORJd=K+p2MU;-hdDMt!7QvNh8 zjG|_`h)*T;`{!QdJ7b;TckrQI`cRZ0!r=MyYfnWPe*XMf``PyC zQ^vnuVo%3>A=Tc|yLaz;v4uH=?i1rR3MA6aHGl$hv z3AHY$XvEn2oJ^olP{RQd+Cr|(8SkCDceV8Ng2Ci|0tL*pMBNY0i*EG5D>9t5jBmsf z?^YNdd!%qb{bolN-3|ZC)Cf~mK#neiQX=+BsgC3Fh*2UH0jh(o!b;Zc$#M6mjq!Ls z+9${VJ2e1ylLw!uO}+N1>W_Nt&D$dBD!1xS1VP%*DDHrYjg6hR1;W?%Cwnh~%lALG z=-xLDDlQg0g`%dv-sL`T-UWr|4+ke3>O)`gTk$&mDcmw6C54oA1d^#{$tvE+I&sr> z^eD7WCJk&x?=FrI^7E@T^a<$vw6;|%E{HiP8@GD;lz-ZDw}-4~hwwENMp{Eyyv5f} z=%W-7Yk}Kb5DoWDMTsAvux!u7z`ZlwijkwqME;ADnG#r?1AtRI%CRd-pwn)f9K%8- zrrk#8p%v^hh4wa7AdX3&+{ZLo`RwF$wsn9)J&)`7M5l~xRs2Tmg~u=}{+_GsRLi5# zM84@uECkWx9OqWq{F2{%RPTm10@B*Rz@eLlSMa7jk~sxqfXNW3wK^3KZm-+IgQ8Az z6(671?1i}2>SuE)4&LXlKjN7}#*4ArDu8&HYkdldrAWUDdp+rLG*^Z&=W`ZKov8cQ zT-N*Y+p!iKdOpic&vg8N{Ei?_90DAOt^C@w?Oh&BbLvNMM6oHqR{72RVP|{aKXHz1 za%};E2xX9@Kz(Ps)|K?_>kPtzS&oWnIlYj3Px+DozTw+Eua{IE_?!}mg|}XaMN(N^ zNsFaa8JTC@NvGIHVLDQDuPp$+igz53Ox$QVcvo#IF`k}kV=glMaZ^SV-z;U7#>7mT z43=+yVE^iT2BbGGB08dm;M{j6Ct-(kZ_KBj4yVg-XQOw}VW-In-eOmQ%`RdxsR>l~ zk!m&F^HOG+w78y(lz5Eq=6M~mqnlUByIkTU__m9?v1$66N!f*`YWqhb4x)`;Ib+BP zH73rN`+qM`cqgtujZbqm#wyo*lDc&UGbMxH69Xd#L&8USwu8`xzE9mUe}4J}cD5UR zkzHBi<|5ge)^RE>V;X@fsPsnc7{XmrTmOC`h?9=G=UXCtnE6Gawr`S1JN3G3gJ6z3 zg`0lTdE-%&H?eb)G+bJf`oc^#Kb5S{(}-MvOLpHBxx&UsABN4%m+F^$DId{iPHXx) zZbO=8(RI!TwoXVQWGeaY{m}IbBIJXQd@T>XPdUW)dnUVL5Y~B}Yay|3f5|>5^SPJ~ zm2T!`-vW)%SG)T$!ST!3OL(o%uxR?^2cf{BFMdTGUZl113PR<$WOw&BR6^L`*6OcM zr-0&%(jC)TCDw_CH{cFFw2tSTX-1rPUM*s`ZbweAn=vv%WL%2i)pIiL!pro1K`n5G zPq0HMZa5;4?6-2aXBG`yqMe|r1s)4Jk6ZEJA<5_OEimXL{Ij}jQbG!!>!emUG%9yC3+MO7lf3Vta znx^m?+{Yjg>{SnaUwH+O6`!#p=ypLOaKJEcp(_O!6Rj7lI-&wwl?oViy_8r$2@O{S z#U=>+3t9$;5dhKl64g7QxECTRS*o-uLYEEl%5(>S9w%QtJg7iGj73YqO7mu88~DV@ z=HDO&pM7rsP}@%F2 zy6af9cf&z6kQvGLYr_d)dN1cBfzag^1dBS=muun$xl!ceUz&G?O`tC-dnx0*`C~4U zwe}qnkF;C8lnczhH$IF>KznnF)!0?uhlo)!>@HgL0+5yQ(?I~MD*u)fOa0MZ5=H@L zvS8Be#7LO0#0E{&4#K`cx96|P1_1GIxFJ3|#^l^Z)Lgp!y_WZz+?Z8~isTz(weCA) zlSH`;k^xe28Bhu-krPX8`OplKFRfW+vV>a-Z;J4+q18+~r^A)vpfP;cw$wa7+tm0M z3dLCPn9GraY%C@J+4_G@0%^mHYmT&RE_no_)0x!|KQ31eV{_*`rw;+Utm4y|xG)_38%K zogYh6!yV9;e$3Bg)_uD-ci9T%|G~E<${8%cEml?9j`xJROtm@;fhUM$}pU20^Jg8p=sKvsV2V=W~CO(o>b5DIiH>lyk)z-g{Ym zm~fq(tnI0>d;Jh$NQ4rwNPL(-vvskc!N=QxFe)*Dm0!~IE@yw@hcBg+jtSa)OU$A+ z0abbGqpkqOJD60mw$#cn{M1cpJpY5N4VG@@QXC}+LziEB+Do%LXeL#n1DP_+DQZgG zcqVAwiW`y>bH05&7X4i=@l>HJCV6qa7$Bz-HpK-BgBE6*qqug)hly~g);JjXOi{^6{*o6q zw-@BC%!g#|VL5C6O19$Mz=!J{pLm6=UQ&j=j$7I?ZC^*>3Q36OMl2Jc`SOd0KP<<@ zP}CT~vw%{&QlMYXJ4ysZ4XrHj_5}HltK4EDaIzh7-9yOgdo?2CySKN=uK0Kn54rBK ziAWx%PNzf7m+A^Ns&U5Wbxa==zn5ZAP|zY1Bd81k>d z0@}(_Cr^-B9ywi(I6H!BiIe8OB zLe#4cwho04m~mN=AYX$MNqogloA#-GBMTkFBxoI zNPhbPofQ8xXf3FjiA$IiDs>8tjw!!#;=Vm$M5&M9U7^n&dmTgi5Fw&qb(t1RjE)fj zqd?gL_lHFQ%bu!RbNJdm!NPeNx4yzYLNJaLXKPHX5}|`H-K6dWcOUjUw(hw8{DSs1 z(g?er43tGDyGYjzp6kpP30|!sel7pv%$NKDYmqh29f~7>etbQq(7Oe72^;=dtS0yH zYOwmY6i`NF<+1@T$QJy*qmHddHM(VmN-pHF3%bBR8ybf9@gqk=42}6f;fE#s{N+o) z_=1@)^G82sM^eeKdUcY-0>$8&z%7+B(rHqkIjxSKx6dZN5jDZ+ZyWw-7He0ycQT36 zhr<{tOxfkLT0T|3-`}ccVa)<#A^IK8I;k<6SpBj{0nB_S=TWnY)@*;+oq}T4ZVM6r zJjCS{p8W8sz5vaPeN%=z#}n|B$v@4_AbJk?)5($KjX>PM)lRlsZVUJQ{-=HU9TCQM zBKMQ@izAkuB*5>w^zXXPcvZ;Agl8kV=22;0rG%QK+}vb7+9sdZ>;fcknGOg3!l=gF z<>mK(xAGsa`T<){E0@eLhV%!hh4{+{9k8oX+wPm65`~$Y3wr;*X8!ZGfCH+tD{45G z1b93sQ8!;qLD#+Zv&a7M{j(!2P9i5U!HRGBFWGPfX*aQJ2Dx15xoO_)9qsoxwVX&k z+g4vtrR-A=%Ovh62bW$&wTBGXZCd{aPm)?^G$*NrD7{NpVOAdg4cE ze2)ha%0k-=;91MBitehb|2;l}vPe5<(L+Eh?|ze+np*VNgR*Oahg;6_*8P)2E6J1j z^>CD^yl)^Ax|V#P8+!0teNLE}FxDyVna*Nb9Dkba_tg8Utqt{sN*`N;VH0Qmn)%XH z%U4JU-Me%aiPT}L8l@PFNEu)UW97z7gOBPKqRU97E^MO&>L?No zla`XCa#n-0KnWz9cUW^(B5VSVriS#1ikQ!6Hm@kE#W5J!iLFy*G@54=>?W7Ig^tP*C4>G6BL)5#>6tyiI zy{`&M2;8N3Dg8#b0%z&q9(<4V3kh}L+uhEiyZi3lyF+NN|I5<*K*r9-`(&@THQoD| z7m;y*AHsUb_G-iFbI67hd#sGSkCp<`X5cQ46q#Ndu%#Yasu711PeeQ zxpj2*?-|2ipcdcR*#VUX%YOCgsi(8E^XzV)OuIi80SXmD%p|9zgoDt^*4D-Brz0Ux zH+s4HA2_W!A+lYsWk87CyW5$fAi)I zq95O^-Bm~Y$@h(4ylIe*&fakrGm2$PjEWg`o^2n+?nR;sj-2S9T$*#QX<6Rp)=Uwh zNk(1&XbQ08ECqR;{u6pK3mzB?Sxt%YJI7|T9MMZN;&gKn5DO7{M|Hmhx`O6^8tWJ+ zyPDskwTJT&CZj15PS{WmF^PFLRZRuc)oNzWHs2o%t?ysDa-|jSc?=BJgD&~pa#@**j-IDDJ0Y9pK1Xm%5ju~kD7Ah4 z(K0wihQ(GxWKP+<51Uzt-Mt<>pocM>>+w6hWKqG*OLRhGPC0l=1wS1+hkp51}-!hx9>Uz7M+4y!B+@ECb8Peon+vI(I!5i9SWK9k#i{BsnmBZvHQZtLvtK3fNV z6g8vOJ_c)fc%oR4`B~^(MyNE_w@Do(bZ2@T-;1N|uVWbNrRk*3fw>bI; ztP4R~p?skU$(bB!5syoNu>j~y3d~k?y>o0$$leZOH5_2WI#7- zHCSOYr&f>HefhbVH3a>_61{ZR(_jKExp`@Hrgfeh<-6b;WKd$cAJ2QX?|{w`nSMIU ziDDz>d4;8hcLd={20XF)+URT!{GvGZg;eVE5FIO3CB5QjG)<{IbpP-H;!B6wk`77>F4yhiA#yJ5b;vZX~6x zrU=-W{gW+@(*VWhuv|x<+=wD@%0UK%YRr5xotaqZ{U-z1TwokTM|DR{{aP!H=sT)b zBPcx%VhH?#9#Oi`xuz6}i^mugw?Zur)BJ$?f4J}JWio&+>!1-;)c>MWZvwvBXK>h|5wGWepB|5DR1=8qJ|q*0Y^vBNR6#5PR@nmM;a z*nv|FnP~K=Gpo#mPjW{S{v4*#Ca?QFo#)ixr$Nb%`?}$q(rDYloB?2q-vDH9v=QA{ z&XT^`jcA>+Bd!-7z^Oj$^1FkQ<0Z`c+)#WOi=IOM7BwT*;8u>!{=^ovr4(jP&>^Wr z2KoKY(=^S_DWg%_tX4c1kGj6zlk)mf;^b#OxwQWXZ{>Wx&Gx>i+T739{Z-o2q#xylF6dr^(@9e9bx~biOaZJ$j>%UUhB>J)CF8Ex-EtwJxbHich}7@{!hDa&okAt-gIQ z+KBEG=($tqw-ECe59fQ3I4G{UZ`1LfEB3mHR`b`I=o=<^6zW$9x392UU$|5I$lDSL zi9%4_gxpwVOqKVo4+>)Wnumlmb{>< z0}PIyeRq-4)_1=oLp+Q5w zVPT9KT7+&bGpZ3BTKrB@4HclD|C8Hlwl&WEI?ZM;zwGd_?hPmc!6XAg4qh0Sb@vj( zb1hy%QS$JXW&87<6P>ayZNLw&dTxif{e8qb&A z1`{jt>1cj>XieF`$fr36-H&|%(_ypPf;Y%enlGyda<@Dhf2Tc2QER;&Rst+UtobFH zs~u=K8JUiHMu{4eM3CVW-!OKYL-Ns^*2&6~mrxLOed!Fno0GKt=o!u(4dawOPqe_j z-S`)Bo*Ht2XwYNQ*i|$$61z^;r~;x6sF(-;5wthdnb-luHw_p{{lQv=l2g=YD z6-E@yOTb1Al^@LWj6OPPxhj3LWJYx(MwNm6Q1G%k%>7fUdhVRKwACC>gz|+fms|{= z6GZ$l+W}m=V;<#CzHPy9qxEu0he>1AJ;)os9FtkVx7?vrD9A*@RAdZP z4-FQwgMm4X1DuA@CN=l(AJ1!-K0ac)3~Vc|X`YWPLMF&{gK-tH220lvc=V~B3WMev z`dug#Ojr#+6rcF&se9m$E3Lp~J@BXugLIN8Eq7<SN5&q4r#QD7^!Y8lBcOWXCne1cfrXaUHE> z4p@0yi%pH2^IWE;kVt=8m@b52A5Fs-r3^r_3IEFGgOkAsDJJMR=fTZyX19@6Txqte zUNk%#T`TNZxAWrBGr20*IJw}AfgmZsA%GgNGtxJBD0}Jqo=h&y@0<&<&2CNX<{rPO z4HQP0ai8AeqOC#Fruw*4Qh(YX{OK@$q0NUV9;&aXu-oGBLB1S?uG!_mIi14 z?pBB`3RPj8grB9pK3GDAPh|Z)h_HqjCW%zaw zjQVB=1^-na8KWCHgO%1zpMF0oL-{7Je3&iM_?|zTaOZGF<1)fpi&3jlq%H&J6m0Dm zDWvB0dM2@onfh480fW^aBr4ef1G)W^XK5ik)xBr~9e9VZr_RCW*x^dEtTkAeuqQrL9y1qi@%3{eAz+rU$_wG9@`r z1Vf0e!$0d9)GB+soKFwSuPS*V#$LOWWWwYdYKL&IR3`IKeLZU{m#yaJ)^jjW5Ycj2 zmAd8lHXN2XsD0^V4QEIa`~K)*aw1H|SM7ok|E`K2(&k z-WN&=3Yaj;_syxf03hq>N;W0?5%dr=f;aJSP)qG27L0jpNzXrcElFExFjXN=0gjU( zP!b=u?NyRt+ys!SLq;9VmM2@w%*M?L)p8rB(8&&Qn1z1HB&RWe&k1& z()T8%)kpm{IzZIfxBoEPblg#8`~r*EFJ}SBo_lQHMjM{I9bXsTq~n1FuoK_F1fwZ6 zx7Bi&@(y0#p&c9S&KqQR4SzNMyfz2pVA%u5=Vfm)$e>)@b!51;3QeGUi-v-UkZ)bm zfv%O9C<89k6rH z|FkV??TqKC`&{ziliOmf5-Bt^x3?W8T_&F8efqTD%%e$+p|KFF$n=!BX~`0I02$f> z#oNi<+yIk^Ui)(|eA5vZ6iPrNCiqB9OsnY%)pZ^PQgK*jCWnV9w3-`L1~2qCDlek? z1_1Ofo2maBIoUs74mwua^3)38krC(pd_B$fM@f=Pt;eGOp(J^0i{*bT@0EbK+W`;0 zTX&KB-*M|5aOuyr(wMf;n0opxE?V*%mJ=Eq8_$a%JOOCYmb_Tk+nA^zfktiRQUbKP zB+qt)D4QAwk&&Py)js8tM!z=779zguZNQ{%F#jXV9jASL`z?msf;V_B*r5N-dN#m#>*CQ&PAs|Gf)N0c;u)U`Vf)^`aWz zmi^X0T0;}+jkzr}M$IjA?>!F4!CBeZ&=5JdBM`9BP%!5X>Jt9|exTc?eIW&&eh_hhZeCVLXDC}HJ>}@A z{A#N>9;zdfKprL#%^1Im1{R8MD|cvHhHC3VzOU+~2-^#5V4$b%z{K8x{_zkWX0pQk zx2hs@0U-@+(lP^aXMxwm^#-Hu>-OG_T6`?}{;lF_y8pyQC5v8#!VxUAg%SpCE+%%rHkkHC-=sI5 z6b0p2&OT^uGb;^hKxKv%jBl6!R_5X%8uoHoZRJVIB|EF!+b_cyd<*&uQZ;(<)x2-6 z^ep%BIdl=6V8d_ox3(QL*-2WXP=^YV7C?__6l>gei*jsDty+0xQNSRGvF%?iW= zy^p%5Ri&kaZ<`ZDATg_S%aHrj+-d=80GQe=@bv(lRJE1kkv`elA{54E*}6$;;+32e ziGNeGks5=DD_yCG`Ogby^7}j(pM+xUPsj~#^0)g|RGJkZ$z>)c;{KNeWLDg+1+jYY zh+K-^m1g)OCG)GqFVeXOOdMv{-nAG$c<}SpXg*P&!pfw5ka!r-g)Cd}rqafj6>@pL zQS`Yt&)X(e;vu%G$GxOygiAKZQ@`InbS&>TOS>|`|Ivxnl9!wNI~2E|bR@V5m^HGZ z&hGA%{)ch#L5iNxP6yfo8H@eXr<9*0ft*-8{`9e}?e3R((C(On2t>Yq3-A-g-e)h7 zS$^un5h)qesSr3b^4pVvdZXMb_WpfSNiT}5Rt3ThS#_si+QcS2+}+Sm*CEz@soRhv zA%7U|NRaYR%+Y>g!4G&#IgpUtxpPO*=iCz|$2|g^ke%-9eJB*5?cWcs0^QypCOg2` z&GmK1`4&8>lZ{gO`XZz{^l-Z?ap}80HIuI>|F>kn-L`w%#Va=>%F%?ahk9k#yOi1A zEfdb&1u)<}W8@2;|3( z&dyuv>crkB6Ep1m{Dg>B4Als_CO~)`f<-F4vQoUIt!*B`#Mip-7CWzM`*<@d2)*uf zy)=OFiY4Wj^=xg1gnT4T()HW#w{4RSp`aThb8*xcz&umuA@;fM?*lUwV~E(a8^*y5 zJl|{y%lWx!-#!-AhnJU&@*7o>BIE%W?AAJZu)jaoaBC&afuB#1F zRuvQ0oh84u$ktE#?`eobYG!^&+N#1$w>ka6GUK9Zc%(5T2)2@ zL&L*tP|am*hl#{cM~j7BCnZoipQRR+D+4ASKJYtricg? zeNTs$Xu;Rn$HMfgsglD*4IF%YcSFu3YJck+M!L$5d} zU%UfAal`@72AI<1O^_$d(7ujpZf*{$o=U)!c`7;)x@BATV5fs4h84=_@F$hO%O}7% zvmT-$DE)V193y%mxYlc@m8Q9V695-1&S#kl5$NK7O5)$WlXzShBXfQjS~Ta*40UKu zU{H|3ECOhVQ1fV{5%!0+D?NI|g}@r2Fb;<-`;8k68d5tgc~08jXXfDXh=kt<_~+u% z(%hhi4AahWoyP(D)vF|mS!6KKr1^4DE&2P{==3NCvH9>#Qg>)sT) z9g*g<)HG=?N!cvs8%Q1ey(2ddPs|~`gqKfazb$^d<6)K`pBm-7&7C1JeVGRSpUB|b=$hZaRw7ww_bPvQs)&+o4EYB(!6DK+HE1q zq-dhVLdjS0rZ-ZfEh;%*yWc>+b8MaNMcsGvfYHQrN5Kr{ z(?O%h5sJ{1R8$C|sOhTr4!p6{sA~7nLu3us11l8l{7T^@``RcEk}+Xyorye@Fj~$hd~oBqjq5np#S@z z%F73}4UveOSAAR@IjRxfUt)ee`}k~sZ*TU;-K(t;SL?+WcPR`=NPwJDvsq2>QSZ~Y zorQK}dS#(e6**Md|4cQyo6FViv@z6nTGH2#$qZ{Jv z?C6+{2oG=btm|5ukB^T(g3`0=yah~Ih4-5`bzh>AiR?c>Hf&Ely^+vS<^3B*#ShUg zz&`f!CIYHG2X7$DpX2ni4eL{o#?>;P48(Z1Tz&Ah!`@;K#sx-6p}gVlHJg-zS&tbt zGcrP5>8JLJC(X^jVeiM?cF(%W2AQ;#BPABE*L}Uc%f5lde*V0L*rWqSq2A?aMu}^Sw(wtc zumk>!KtB8nHfUhd?ZHW6dovD;r!UZo5Tt9CNsVhtsRWaZ@pMh_kd(mM2rj4E%Q&_{zp63&n4m#IVb^8djKyPeHo zGy-@j*r`O&%5yI(KGClA)TL|$V? z#~>U6t#?CTQ>WL$|^k@ZYxg|fh=hwrKV)am0#2wIU|rkrL6KoU$WaIeS@On#=qPw>=K=XeAbq+m|IfuNFOLB| zrpSbZ1h%5Ly9YQZ*qNA-Kv|0N`|1fp$L-PPx%&I2pEC+JxTRkqFQajDe<*Ym*&4Fi zQUrswSEc%53Z}hwl1i?{_8EcW@gH;=K*ma57qH~+)}}+-54Ih2&=ns7t1D{DbL9pF z%ONdY-wMdZQxC_u-WvHQCeltiwlihwT8m?Qi=Cg=2b;6cm>C;0A&OAMzlkux6g{66 zH#Eo6Ze5&dj*{$G4!Ir9g&@XS^O<@t4ihn_W!3H!iF>CKpnio8-{IMoPYmo7KWCa5 zX)fy(l5X`Si;;lqwWGJU4P5Qkg{BSO9itulCm4b9a{%qGEj-4E9Xt?YgdO9oE;z2y zuQ(L(3UinshfU7G9d&hPO3vYt5$u{NY2{BZ>)WhKGd_Ll=?aHQz@dn?cJ(0oZJ6gW z60tZsIwD6|+u4b>w)U&oShUh{sm)RKz)sIx{(HZ`mo1Oan6LxVYv0$eDR*|-+S<$- zEPMo~`}FVM4_W_QLkt&+OpSsUKtE$5RsgyG*9X7f->TyJpa8$d<@Yz(qHb)C zeqTCC{y(2~;y{J#($dpBTW`|SV-TPeaENIdHe#fYap8CZCYF3XH=K=?WVK6SE z-UAqlM?Iynv`>0H-jT?%HvQd-JF=Bl9!z^=46eeZ142sk-o1r=oLLR--PD9PZ_pZx zEnx7#e~)MV)4gB=LuMUui_YQ+ruDhaeZw*WQ8&ywZiamcEz0V2dR5jZL_TA59q@P| zu-SV;%t_61m6x|ktl_AK%`_ga8OhpAV!ZYSlq%=LFLMjOJtKk=2jtSlUMDU9E5-q1 zI>pO$09U>DzHiYAHj1JAdoa+v6ZXU3C%cPaJO)WTJ1EJQ9UJrnW@g~whrTRmqumVp zWsZMw%rLKcQGYHZFDNKD1Tfw?1f%m1A0t247Zyftt*y!s%@|MMFus4S<3r z*WW)|HBu$t_`tK;dH}in62uOjERxR3o*PX#EQ`77so%GLmPN?C+`9yF`_Ww(iOjp@ zAdY@c7h(?PU@6r3dQk)OpCcd*{DG52pl?@dW_Y;uVlP)_>0oQ7@4k-Cc8%-S^6u{L z5%0a6ZOBml82{#hS|QzMvufqv)z#Lf_`oRlL)Ze(ft0XY-MhQHBh`1Y>DsWM#!Ybn zTg4wlt$Z=H<^5JeMJ0A}Wx<_|!~S@0^>lyKv~P4|EeVV zO0=;o7l-Z0DR{6kzOZj>z)!uCfzjM<836V#D0g6uM9 zs~Sf)J^pS3iu)8K;z!^!35tn~)agEGcaTi=eECumd411*r{^c<)z#H4b;Q1vR(iU+ zz_TmgcK~yZ$-C%uufKFf!#4lU;!`k_93qFZr?N7x&-U!C^r&RaUhyip)o!Q&JGisd zMfn8<|6`AU)3s@KCMHw*dwZYC1DF2qTfg4F26$9r(uM4*>gu1s1=ce_QqIoKU?;6# zw~p_Dy7bKVb>DaY0k*>0YvmaJACu1i01P6r?$&tV=JELxCLDNOVr6G*J9j_OHkGfH z67GF655@IjPR#aEyO&||$vFMollc{oIQK1Ie*Co+@T%?4PFYWY?ev4{>gu1_nj823 zdbRrJ`hCB&4lI@jW|WV|CKQ)F`03_%|Dm`3-XrNLDIco#&f5>XAYpINiub$U+x?rB zz3$_#I^df5KfoLDD)+{nPfttx1iB%B+2+wk<2&U~E&sFC9Rr>y*9K}T^tvg_3kx5v z2i7hR{sRkB_PP(r57zB|_b50hr~(*~%a<+t1X_LaW3sw`*ZI?be<#-f6FD%!0-LnW z^CsW^e*Dr6_ItpzWjF8n@ojQ(z;TL4Uwc5-{Q;@~C57{|Oiw>=>U&&SzC{ewDHCXM oT3}KL>Ypfxuye}LnsNL;|B~Bg`=1A`_hA46Pgg&ebxsLQ08C@AEC2ui literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/jupyter_execute/week34_86_2.png b/doc/LectureNotes/_build/jupyter_execute/week34_86_2.png new file mode 100644 index 0000000000000000000000000000000000000000..b12c8193f02443c48634c61a03b5e8970c2033c5 GIT binary patch literal 66200 zcmcHgWmJ@H+Xf5|q9~%I(w)+xbSgD8($do1-AeaJcT0(Yba#w&gLHRy!+UUF*Kn6Nwqg2)AdAY48}0>8;LIPeAk zaM_D|wpXw+w0F|6HGoL#*jt-h*_)f_k~$jL+L>5czGq@%VqqXPwzs#o<7Q^I`0pE- ztZa>#|LPnyfDb{o7E`r@K+tp^9te2?xh4<<2t-`?qoQ;2&a9KOMn}!X;l88EShnV= zWsS;=h51Y#(#Q+;cr8I1fF4n{lO{BIe-vuyoGaT17Gxn5i^a z2&|vOyP=T<`o)zd`~z)1PL+LA`>A~U2?=lS>fGw%gS@-B?W zAJHFv$bJPHCW8{-}iicM*sin57-BZIEe}>*akHr9}1Nt z;&DZUgm@w;&tzX;r3G@e%nSD--%&0}A8=sUEw-44Z;maui^(wOkI^a8y*90&8`Q)= zdWK==-4qm$btB%Mv|vTPhJ+ol@;`Tht}Gl$@$Y{GKVyi8|IZESh){C<_fB8`fBy#- z={#=-cV_F7a??D$TU#ykjn==%$J@``YI`-LPFpp7Q@5<~Q7v0b9lN%euJlVxG+i6e zwpK3wpNr2Bq~!8s*KL0j%j6M`_nz8zIjo8}&RxSU?A%dVnVp5D#OT&@y2{Gn`mFH3 z1CR6~V&!py_WhL?#Q9|C9>rSre`b(1h>}K3QAOntE^wEcEt$5A=5;~BZZ-EOns(%3 z*UD=koXC&&bW~un&WQuYf92(Vxk@J?5zK<0<-8iRT7Noj%YX0wS~LqIuBT(BX1&aK z2y@Qe4zEDjLrPtp9*x%;2jX>qjfMLH1L3ax-*F!FYGvx0tei9`9C} z9XQXhvW{X95lPF)Fg4xo?hmZ|&zuONmF^%Y<|$??=mZWn?yO+_pwHL-F2+Ct0fXu5>BX9JETtzhA%MKNtCT8`avC$0(>=p+fS7 z_64&-o?F9QnZZW}wMBZcTm0fFRF)PLd=(b9TRj^&pe?}2*r1}&J-U^uTHf+yA?@Bh zkS?DQa}Z;FzzE&&(8hLBL)Qm0z*u95iMnzx;RW%XF_AH_M;Y zMlJ}qujmB!zckOidQ(n4fefE4Gr}ouyn0u4rw((`@B0Y4^<;LQgMd_dOAt_dU-E<~*GIj{eGJbuKv8)m-mQJ0G;v zT~5#6wUgW(;4CdKBgFH$F}XH`yX`mlOih_m%Ovllq0#+Y;$Fd~{VO&cc>Ad#_85k} zK1I+%E)SbuU$P=$s0fc~|EWt?Uee*Dgp%#a55?uN7v>@LMN(jmLKLqIgGlk>PTu1W zSuh%dKNiLgEC_q5vW9m1R}`ZEQwMb*QB)ib0rT>hgca<;E8YY?Of8!z|E)Kz5UL)` zQz(D)oZ6$W5_+UH3!?V^?U3`Go&PPdNTjNR1lvR-B!U0) zF{2u{Y`dq^+*bK4tC}HXwfh7!PbvP->m%}Ysj*Z0G1KBTDwud2oL~8l82``9!^%}* zIX{b|r;!tmgCuZYHxpl`{f%m&l{{ zKs&IE!qZEYFz*g+cKn%Uj5z!4+n?pbgRAX(o!u#`z`(%fWt%KzGh6{ra;F&^r}R1} zK_Vieo1-`ZgZ;VLo166}!@HX&-}WN^StNs!;b+f(1*m-8K_I2GVK~IYNte%mG*fmr zR4m@FqV3TE%4XQeX1it8vNdP*aX(LjfS0FNqs5@x=~&$7vNift&nuD4tSp3w9e%UQ ze9vqgMVK}8XE=MnrjL!h=ASkFj7lcyiP@p8e%F%^`ZJ#}%W!-xyb*7cSLzQt2^c>A z{z_k5kNLH@=`QhlzbWmb4-$+0#^9a%cKX?@)BIb9EqkgKgm?ijg#D&_$L1@_e-{=> zRThTX;kFnt55#{%L#hTvF2B`nOrnXJ3ytptz0Lonf~9OKFP?bh8+v+zLGf_q0zG)R zV6qI2YlD`F*WIN`jKKa}Lr8Ag9Zn=AS*qH?)nnXJ)a?UeYeTR$kXAnOS-l1733r z2}WA_Hy1(fCqr%zm6_P{;$vqhUXLU*<8lX1oSeLTufY8!F2rsvc>v~h^A?hwO$Fu` zcHH}(xzu*`srhuJhzl%Rk|ejZG})kMZi(dHIWjgD8y_G4>eA|8QIUSyT;V{!L z3MogzTKg#Z$?OHfUV_&JRhZk!5LC6smg>oqC-oOgA;Kaer&l$gSyj}YA3v!Y-&x37 z8YrnFG8!I>l1HGcc&-}Bvn^*IG>BjAJ}4q3Ml@Y8QG3)~ASNZXQZb>mc)6bT@X%TL zBgcEOAdAOg$qQBsJ1uQ~Z!#Cw*S~*_nHrt>>6PnSEUISv#|x7tE6iw`u4hTW=(=oG zOfcz={gfRv9dkdMTpyiKS1nx;EmvbxEBv-y-1H4}cbpl!0S`(ls*EA$G^Z}A;~~?r z_&#b|yJK1={y!^SQ6^Sa>m#x%M}H`qsOY+%HD`*1xSUO9=Z(>=didShoy>1++&{_q zUkHUryG99%nrgFJC>?v0J_e8A-=QwQ+8E(|?|wnmbqPu@-0OD2tHgd&9;`iAD4Pn*j?myWzRL zTXWFWB_c;s`qqRaSBP&-OiCR1=RfM{@e~%G^*h!b;Z~_*0(SjG2OYd&9)EcWNdU8A?f%q?|nSdBnIgQS>}Nts!IV zr?^;PW=RQA)pfgeXjBEsc2$#yFO=jqDsltT4yVy!71enw9GRxw3@Qq2WK%gZzOxfa z%k$C&>{ykR)t9Zyho7{xK8lJ`L!_mp-EU9hG(1i#zza(a=W{s~6)Atj_Q!%Lpxppt z=o_pqI7)WM)w!h^OkO+fI!gY?F-#vbrKP7gI@z#;SaTAxva-H`VcdYbZJ)1NQ2C!` ztX|!j%+SNC{429ThXXre&3S z&F!R78mTxjwQOqlm0;j~e zCpa2X@{tKub84v0A^Z<8n5p5!785JVbxZ(smh3pKez2&0+7^j+UH19pmel^?OSo(* zUP9`#LyzR?q+BM@F;cJw`JY5e#IDYvMV$=&)@gG-n>0$`b7O~)cwUgfc#obHH=N;P z(a7wKnSCxSyd7)c-|Q2Uu_{zQXH4@rL$0V`1IW5sg9#MO>gDh{To&C|kn6MEMS!#5 zmQ@%S7+rr$)~(l5&fjbsZEbHm-yAl{X=(Wn8UINC@(gUWHqbVj)djB(;rF5n3KUFn zQ8vre2hE5AEnfn^wE3U%Uhl|*Zt)KEkh_!ICa}{5RO=jH4h^w9Odx}*tsM+=7!NN& zD#?+_s_B**z&$?{oD%EBrLt?u*D_skEHuv=$o{*#?0PFcvkxj>`q}c(uG3{=dI*%! z;>P!}G`MPjj`(n|Y3@hJrWN?&M@S=l=QCN9d<_`kfUKSP|6dMRWgS*ic}2t;?U6Sl z^QS%@_jmFvfj|CnIUX6bLlDrpUMAFwLZ?45^h~l{Ee_NGNia_@C-LSvN_4 zp8tX-!9eTD!p|XKLncFnJ5kAK*KH#f2>%xrp+B1GHv|Zg!lgc`o*=VKyjwj`A{*Iy zz#H~oWJUJJMPv7KaO|Q@1tdL0sw3Hp$h3l~nGZ$gzZ}k&l-gzIbFemiH}B~OGcj_P z!Iyb#7yxJa(R^~UKH|LNkalghzfozlA{|I;EEc;+)s_!$fC&MH#UXP`s@j^;L83{y_h#p*~8N-)6y?`O4Hv zxRC0H1SoDZfOH6-G&MUwx9tar;CEy%??JQX%~@5`Iv|e%t7f=ia-hJZ51rfw*e|wr z2#eyoY-X=t&A9*t1L4oG92`!SRyJ0e1$7g(#~&G?Io3I(OD3kN+d4biYJ5HOZi#!8 zu4G3E-S<63&%3agw2`6ep6T@AFa6q~E%C&E!~M`BRxTzpMJCdY?kmLCU8huLF~J?mq)!fY{~Rv9Xf&mWWT4#@#|UEh(-eqIt0z}5`U5}8#c;i+nu8-){Wj>KcHCoF z0c=@sD3j_JC@ivkr_8##y5h0}Uk(kKVA<#w8NAi)pw z=a_d-0T?HUO48DB>?f>bdUteG@p%6J_8CA&_j+eLF05>9hs}siPY$90TzyOvx3X3j zeCqDFVm75oBvJP<%Sh{tVa=eAQqXNFvM@`UqT1j>S8MW``rW&No(atSu!YF_Z8G@u{0f$EBMmY8;p%3GH@<$Fkt4MRSf?cQu$AA}Hx*;1uOlf*<}CwBH@gdh z5;(hZ_c@&Qo)}lmd1bQbb=qzWejDMZ;pb1a8%RNJri#-*YAzdKqGDkQnk-wVg(a|= zp{9A=<;z~vgI)U*>{^9Bc7CQyZ&7H}aFLuzd<093r0qIJ7L0#)TS|)WX#l+I2yy?> z!QmqFTjQ6#eTkFWnmCWir2ZzN$$PB`tvZ@_OX)`C0y>t#2-E`q0p-PC`DfSH?4FI? z>*6xZzSL%1StZr3Jhdh9eL6OMrGZ*Jo8P07p44>bCK>Om$GYv4GszYGvD*s)B?0Is(Q83p-1%|&d zVOsUcl%=&~+8yF2H+iG7qAh6Ho1nsy80SIEUX`Mht&Q;8{k{&D)s^~3C{GA6iZ^k9BDmTFu)b`yAVDIn(}N*yO}lw?G|2es@P}X| zmDXHYvPEpl*DCpG*g#Fl%-uGnoHk=TtZnIx1?b({FBQlp3Oy zbuOtr6PdK`)X=?mbqm3*h`6!8;nNq+y$I6=V$%xFwOUO zE9|VbBP!OfT+>#3XA>VpY@IL1V#MPrR)%AX%qEVGnr}w(D|-8rTl~1nShj9>Tnm-V zYV<~>c^j5RP}yPz7b_3-7ms;)@3bPr44ckGo>@3n`G7WNt6TQ=f>s>I9U43HSyK58 za=0jtkZnsCdCTV0fBJf>H$#tk)t;`wq0wh^2Z_4puSbTd-wEtL!X#fp92aO1h4blI zt_HQPa}CSO6qJS0R0<5VVeL_|kduA6#T0}(xG~D^I|m`TQ4H0-Su;q3yWow;)P5{P z*U(qZEIQF7w8)0!u|>o`o&7pa+IWBaQA%Yy`wQOED!bIKt5&O+1k?xg|A(A;lCNQ@fPwnZ8@^WE(f%$jt3m4HYWentN2;Uc7C zw1-V+7c4%Ri1jz62L5$#IweaEq=duUTRuNn31A51dsTV1R8^1g)ERkP%Z-F{ZCvg; zLZ$GrPp+j{{}Non$Bm5bSrNVyB=z|$?-9OLfCq38DeJPZ1dKlDE6{VxI^tu6Z_R4(EpcEoNI#= zTF}A&&wvEhU>ngFpcWfyj`7+Vv7-XwIw~B)AKkr%(f8iRK$MSmWd|wM{ZDq z_lBDE_N0m=>wJ;+<4yR`u2JBC681Wct_rJ+iDHU%j(MM`NH}s?Z|MsIkT-$z;?1>Y zZpl;7(`evu>BPw=lg+1w)6r3L47$|o6mC|tO~p#3V=z<0iM;vCpimg zG@bLsU-GO?5jZJmU!~ge@ZY-YD&(ABUq2x3|1(ryLb>8r3^iAg_~|vJlL|CVCd^KH zd$?gPlj|6GNF#Nq9C2V%+HrDYRU>!l)i|u2lCjZ^skGo)wWN0q^^WjpWtJA=LkcC@ z`Y5p9lNo3fr`@kcSUrHUBFm(CQziVI9NDGT9WF`{)Lv7qyHGCmW3pdV^pO4Qc4}D~ z{^c7_kAZK%jNzw7?fl1fVvN_7iKJqmFkg>q)Ba>-r8vb#_wxO_9BZ4s`riejj=xLd zbDiq*lH+A$liDM7CLZ5TWxjtcb%@PCH~rJa6)wrjm58vPQ{n`xiSfRu=WkLj%ZeFW zoj=m-K;xu}8!l4LPe|nchGMs-%T9gyYd@n|n`PaY{y~&eJ@bsqitLPvcQE2?Yh)ceP9IZM*v0$qDQ4IF1eev&#@tTJnf%_?(P{;*n~DC8%G;Jx$Bg!TT{0 zf(O(*QS$wZnaNy6x5OE(BrNMYc^Q=(0)JBO!#gi9C|9+@=t7|ew}n;WGR87rP<1Fo zLv=TY*W+3A$T{Ym&0xZa3xInhuOu{|ySPT- z{^^ERL^{B&98H$6^|cQZ89g<{2fpJp(9&nPms~+(BE(BzTip&E6}C3H^ZT}!I0*C1 zo9YF4i>}K;#vJ=Rv}TMZD~9FDxCqxCdEID07a)Vi{LcY>7EmzTM>4&LHx?;*7xKHNyVLu}wcyI0c`A))^*}q)O0q0v z0sM$(C}ODgB}0CKu%-G%N(?WgZF%MN)(Y{!WK%`hm?A#TL+h&>^+h5>kGt)Kq;iaB zc6k^}*h*!@1la!`S6Xu16mhHCe`bCqxT}|>vaQY!?d^RKHI%d<11y3I&YKuww;q8P zwKsB3{A{Kl`#*-uT!oLT9#T%dNSJCkKt zg~}ma(Y-=wx8CLD_Uz`L2LP7M&f6f}9=AYPko9tnt}vT;qtHi3*L?|GnGb+w0A1`} z7dUm%F+?r8`K#mV`}zR31L&b7YI%2ePV*MG9>s|Wl?TRm=Jg7P1ZeX<^EYefFNnAq zzIZVMzFI8mhLuKm7x{8b-PPPQ-ki0B^j^201%}#VqSnB|=Yi=6OL0Oge7w+*2?JP^ zlS5D9VSs|wZOZ^wH2*dEv%IgLH_p!o=ftc9Ig2S$@@S2PW?N%i`L21Zv_e7-wmYu* z(0I#D9>=ZpQ_!#%%wJ*1@LCd#FHIbBCP4;8DD{ZTA6LMzlrpS@Be8rRba8->x+R$# zfJnbOfA1NreY2EXf7ET8Ba;%YRlE)tXoR}lSa}}v{QOB5Go6s2+8tfZW~O|+E^tp& zsN8dVKHudqF)QZg#`ECMg0&31sE610y_%!|U6YKvi=s)pfg_X9JoC2B0mQ}KM*ro! zqC2ck{WO=;XkdM`EN(vV){4jGbgt5fBjl0DCS>3gr^A>oZcO4c-@9E zsGmJIoA7L#sqvr}aBg{mlQH2|Ov!uPE1BzY>t@~`bGo`INX8(Ni_JJ{XZrU74 zU!J?moxQB==hK0ZS72zw^94+*u3xza%jDH@IuWzOQpOC-A|fsa<~Uf4PeR$lp4m_M zz90Q!wX#XXEH|r#;(T6GW6*w@9L38`dJEpt+rtk_z#I9jT@l3G(ZpfeRZe?N|0s^SHIYR&%4x)#oZSPv+DJ^Z+kDE#3(9DL0+gC8` z#~(TQU@MCOTYc`X7xalKj1qOqg$BxN_DtOglF|LR&BRE|F%lh0K1>~Zjt3rNqLivj zezQH*R=v9~qzONp#k78OF{2vTaSmGR$A~i4dlZDUD{L%aW=RV`KY)zEWa2Ta~-tg^bD&X}=^NXzDP(c6xSt2#OjFkVO2( zXMZtVhEAat3>OJ@jFty!Bsw7=V;p|l{u1~pUr~eOgpQE-DX2;|l9s8nr_Tk5&nb?) z7gj!Rq|7>8JKGFwbMo|XCQ(FHTkQ~05=iH|ze>A6nngKE2cJUli!NdVy- zRcy^@#RN;Xn)BRUloA#1dsjG6f1|y;3r$FX`Ww4)vpodOd0cHI=meJT3;fubb&Qy~ z0~Nme`udN6cy;&mh)GK$re|j-08i-X_$k612KHBAfrX0hkI5E{xh~=Ge%lKQZCm`1 zHS|Vd87xHi#V>joB(I8>#3Wr{XDm4p8vGl&!!LBYA96vT5znqlf1Cj4Ndn1ts>Q z2%m=CeF01mlaa!XTj|qN96msZ*G%xUnfJNK6eqxHh@H?1V=MU=}i zlYU3=giT#TmYh|6NoaC1uVH`k%MZ%R*bxyCJ5v>iUm4Zehek&Sx6cD%vgs9LJc@TK zQLr+T5u|y~XFziMbu>;Fi{04VR4Fh)(Sr9X4n9CV1E1Bi#6a%H8_r{!VL1}b{!!Uy zp=Yv)GFqwrXy*N*K3Iu>H#`&{s+NZq^AtY&dK6j43BSUh)&%QfpH^VKGD1>p_w1H$ zSgpT|M|*J~>0TtmAnpjSO;vLDyj-o8-9omzCiu^176(?`e0|ke+56wkC2H!3vrn3x zD7f6t_trLZ)6#!i6HHEWQBqSM7JKpeh>j?UZ(Lvw8xp&2#Qn&2&@8+L_u{-;5hpG# zZm`r&Yd%w*3Q{vn^#_>POd9PE(SYYDRdS!+`U1JjdP4J@cEG~id?BvshE}aGd}G7- z?gHNQ5Pt~=(E?I#Zhx~P)!xC3w#$j8I~lp5kDXxz0}l}bOiaxF3p*4K8b?H&n_Y9= zqJHIp?#O{jdG)_pe0+2<^hlbme}3p7Zb?{qmDUk4FuJFAaK34n#$$cQv>3-Qqd4?}7s0^ODJA?9F`J4c7df6?NRs zmJ?oYOu1h;Ph@EFtUNa~kU_H^0PNMkactK>Y10(Q!>%)Nx~z6`{TkFcp(sCX4~($v zh{D`}ejb;#UAjEHh8f45MdC`E94=<-&XA1ODQJxe%qVqj-Tu9^Shl;L?uDh_@$2h; zYA5qNRNuj%l*Vf%fDGYPd|9s?#9$JuG|T(cYB=~K>8+=@Dn@1tpnTKaGjUghPeYk> zLS6K~x{V9ys~TFhAt)1R^)1Fy@a?@9p|$~+l=CATEm(hDO0R!hKD1)x=^CdF`{nLfR)NBYyWU@>hvHG6h<_Jy19@TWau50oj=)OWZt@M_Ns$J2GahE<+u-Fo-b^ zEHyl=rmfGw{qqfNK%#^zujS{~FAsQqvLpdvE8m*@Vajx`?f&wiLi)vt+jc499$PF1 zSaWuQ0e>ozR$A*`ILyvsK*)irO*#Ixv$hjRD%*gELRm{c90C^*i&EIoB%i8s0)JiBmTd+}@M=aiZLsD_frprF7zl|O| zBpxRiNNt-wAZ^(Cf$Bm0$YQA~UV6JvauY4My+?d*1lPCjNWXj`Nl2JZ{5;n#uf#Xk zvHq*%(Owxln*}+?i#qBzSSg)RoK&!)`_>`ZfRk(XWeaQIfS)RsQOuaHXac;?DLIZSrAGCA&=B-GTefs1WA9*E!aHkRG(y8V zOUzm@ex4>28yh<^IeDf)ju29z=BGwrBjmr6}0K5!jcXSjeE z^X()3T6#cc*XFWy57S+l(Gy64gbG%E$U2OJ&QATazWv3$J2FL zNapzy-UkYjpn{(MKqKoyE$L*Q^q^w4TS;}JD zl``mX9U+{J>p~Rq5lw_F+tCg9x6fZaf)|gCfzS~OHp9YV0LfZrh^v!`cJsnXy#t+k;^wRd2;9=wg&jAG(_V{5&d`bVHR*N# z5)}-+gMl%H46j1&YX7*ZrtF!!AzvzuPjBCU<$JD4&|P)7(y0=?FQIIX-^9Ocf8;#cAfeP_Urwb20cepu3&;ZG z)MWFy`xF;xApHO(cJWa+?sCy}Ol-MR4IUC2{Q#s7hlczm$8X9fG7dGfc8 zI&8uj#8W3LE$9b$k8f1Z6`m1l!;?t~3vy;#wK*7-iX>zu)BeaB&B&M)XtrzMUY6^# zkzf}-`rsElEvh{bT|U>Bvmu*d8yaPhFlI1+*#9l2+j#wVd&|uGGedF*-R0@KL1U|> z<6*0dB3;BiO9kT!x|i#QQfo`~HI$;%CzlSguk2pFE-?9d*MAL(NgVawe%pz5*tL#+ z1r-!jEGAyB+B5|;92dB30mQC(vI9|+c;e4f3*C8%yNpmMsl@sBDm-tu?^||bFK+km zAE(b)m3+40PNAg9U2>G{P4nWKuC&-ZjNdD!?}0Z}q9?lJ*?C(l0_BRPlE0Xevkeo; zZaL5lV1M$p*4`(oWt-Mh7o|G{58lcxwVHh>8|;Ps#McG>hf2&7pnPP}aV{|(J&MeL zSBYLIK~R2+>cBGfxrnwp8Ku9~@muEf@3rtTCU+IRT_ z#a^syY77GQRGc5=&t7k2p|m%d#UDsU9)o0*oT_Tq{QaF12#nV`x%EVb>Sdn1< zEGW2GCh-~o+~zPq;07&;!%Dw9+T_7BHUe3^RM)LE$jzQp)5qcoyUSITGWW{@$b+j4 z0iTHos=cU7-)fJx!E)!z(Q#$F9!4i?kR86fwL>-n#q6NEL)WV1e8-Y!V8=bkway7G zFm!`;kI+NmlNBLtG8Z6sK$L3PvpLXQMH4G+m#mAk0eC`$>p;QUHMzci{iWmhuH4kx zdg{PUJxZ%7zQ@~;_wA+D2alL~6WV~3;jUgooC^iAfWkxG)B z-?{7cA?UrZni@BRIx2|d{(=~!&-bb)IIX`3RL!1I#4^+*fFxTu7+#LvM7F0{B~C)_ zf(z9tedVJ?xecBK26SxVTyYq9%&7I=R?d`Y%zK_GK>CMZzBruWICCVw{f&`^&Pf*& zEFx6YylC?~-;ls^vaQLfvDIh&Ch(?u=7SUSMmeXtizA#KSQSs}sWRhU)cKp3U?I+w zeWMj6>5F-C%q3uiZPq<;2$X`qe2&0X)$3DcA2Zhz?EVb+E>umi+eEv#!G`3e0e#H0 zO@3J*i|pP2jP&sbkUg=mu(;X3zhjQ1w3&67IIRZH_wTP(`FQZphZS#MmaZIJn#C22 zi47kL#nV-MyC|Q;#S%T)JdI^^^OF@g_3ZjjWOI(@_Av9_idLqZ9hP}!5hf`e`S zUtmx%`NT|53QP8#m$R8aog^?S-Dz>X@X(Dk3zAI>k=T(VPky2rtLnL+dB#o%A(bcS z0vL{F4jl*RE_dn5`876wD>!FR9Y=RtXlp>{lDt_-uImcM7;6dtX;F1B8XlKd#Q{Oj z2wH$-}+CLSk`FUb5hsgvSb&MU`&RFhCpVcSYUpMlQH(Rn#keyXeu>oZbTx)z#}ax znZT+(21Fbnnj>TGR1dNS7hShHRenRmtWh!ZMfmiPnB6Y-)V333Hqtl-iJRlttl}5k zpz2F^1Ac|5h7C@CZ*D-7zEFh|W14dI;@%KP6mbeZ@J#Q1$eC4JjK1WpTQ)f3AZ}f& zEGs{FLX7W)mfukQh*k%A2dY8KBSqIaO~D*zL!Kn4X` zb5`G2cxbQVHdCd#6H6y+Brw$@h?4So2j-N;`&v|D=9TqYahKjRy&6GlQ~WF494?(Y zXGOuar4YITM1FyS2^mRAfyMH&_u0xMULEr=zY||AYc71_d{wAh8|p^oa+yRSO($=H zZ@G!snNalpJ^J^Os#kJ$3sk+AAwwc39vRthe|ud=kjKnW}33R6tXOb`o3OZ-VNi-lJ;9R^?S{DQ7>>66l<1-0t~(H*BDikcx@ zurb*(OzoI*fLazSW&o`gHf&6ojr*8Y~aP8Z9%(M!Q zMyGFd3BpbUx-n8%svCT!A8Yzv+C}}fbhu&a%pD<`bhT4+%4kn%?}PaRay8i&&P(Wf zgboxKx`k?DZ;elq)A&M;-{l+wovV@e=w)9&Hsow{A3kOOovLIAm)l4Z95(nKU4VY7!&ykdxLRNBh9|$yT@2Py zd>rFi8p<54z&Wfg>XqI*h{G9DDueqs_Vwn3shtSkdY6R;V&wjl@~#=4HTH-FUs7P% zCjY|Y<=$0cAhvyt?C(6xW?BRbSHw&ziZzJLN|6j4gfo;(k z+2rekbkRkH;=GmbRSJxL(NNrui$koM_0*^TIna|?UjEpJ%o}LvUKeN6Ru>0tPC#s5 z1r2IBBfztZ0>=9g{sHj->-$>)!{HzKM(1&4CqX01l9GK+_N=?RLq+4ln1Cwd9_x?d zx7iHvumh9a=Za!q&x}8hmzvIoFi7=SE@-%^6?t@W%f-A$p}e+!gaat@YiDvC>HOjd zAzuhfu&=bEvJ$+^Jw#QF(}<#4fYo0&Prvtsdv3?_Df6yN3g78P>9*-Y_7}BpED0L}fjNV)mtjrayB4p-Cqv2A@M^=S1l={6h>dl^ zGoUcv==pA%&8S+aGpo`DUT$atR06Rsl_0eAHN?-gsG%)ZrLdNnUGuG3T3u>GH&j_aip zBc_+AY^b^;YZ>DSGWV1urnaodJK=G(Rt1kfWDQ#{ADAU1bI%puqF=!ZFVVr-4p9(D z;XDzqA0PeDkhNWJ>$p~l7#*n=4lNB`47D7 zBs}iH%$}3q0xoR=StdI4FW4iJvk~XfySU_f0bhU!z#}uqdBLP!;7AMW3zS&~BA#nS zQ#d-|e@tPxHYX9(jjE5$tfY!?3oIK^tfY4-#v{J`eF*%Ty<9v$f5+DQ!B;gk^5sp$ zduzP&d?Sp`Kl*;B?!si=aSbC%sVj$X>0Mm2sMc-H3&4q&BW&GJ0$1gIJMO3W!OouN1%GGe2I6od znnAb3p{dsSQRw*;t{?vVj{?EiL>J#p+vZ@q!%&E;}X=a6xv9D@mM)q@MJqd3J;(`^t++M7*u2gwr zQ9}DZP!h|o`zg6SY>Phs8r_59HZ z`?Y(R$sMqPEAT-Ri%0-*77_rkUfkBlwAKmpO==yZ&B>hh{S$pn)LiZv`Se3uO@>aOJOo2nz z(O=^cO3Zi6@AaE|B)T>tmZAmM>E_36bWC zt1hpBQ+3nubngD%TRio!8reH5f+}PzwI*``%j8>N6e(pSAqkei9bytg59buB6E%kc z-`>a;R-kkWxI5f5O9n@379%s-91!za>AyTNm8Jw4vB1EtW7Q_#g|enabdU@ZPw+1z zy(b6ocVc}41J1ShnV_KRA<5dkOMk??2;>JE_g~YKQg)~KU}b{QBNz#i6-Wz=pCR6b zcTSh6EKw{6=ylg@g+kjcEO{Pw@NJmpHs!xzj-btY9l=!=XX-WKk+W}f0 zqW9xVoHWUQrh-p@abA40bvyh{&&6z4YpZ_r*ChWoRk=qrdx_$!D2c55l}6|xg)AJst`)?o~w8JPew~~`p z3inVG!nDa>*O9_?8CRf{Sv3bD*Iyu*qq#{=yw2bFaY(7JwG@?aFKBHHH0QtLxyH7v zh;t4|ph91U=BflPYs|Y$O`F`fzuwAwvt-owk!Yxiz!*v5{8S6Vfu`NN5^YmHW|2{2 z$QTLFnaLd&Mo)4jX2}W4CG_J`t)EOplLO~TZH*QYHT9oA$f`r9d+j-y6~RW1jlbNP zGs{Ky@a<5Z9O?@5X#6P^zKwuw81LY-N%hu-8js{-Ma&T|hnGhm52vnz&L2}LPP?qi zd(!PWPpQJm_mIfBI3&RZyMNsFd+b&Pr^Ogpq$2S_(kVeKp2AJk9MBXC5Frs(7@H_p z8>hGM&oc+vr%a3(JwdI2o=$;EzCuN$?x4OUn)R8mMFw`OSMZv~!}1EbLs-+|8=gmB zshPrm&=-;K%_QWpE;_L+f_&#UpJH$^RIf24^1Wr$BMeVftf1_q`dQ$eRzt}l z`?|3?BL@>A6(N4(-g?`XQZb_&y+#u+z5NAa?aD^!0}jbiO)n?=|l`QQ}jqHAFw!!i{it@oItqz5L)U(+^1LY9;0K6Th;vY zgsW^I?^mt$**{%hfVZnO6!@y3f-RO#oK?%>By34rF8L#^2zSJR5K{4vBK;@CP6P|H zS>&|q5+LR6F{IQ21G^-K&+7Ns5+fdQVaSpZ)*j8IG57WaHoRtz%knV|du-s%$8sFY z{X@>>gNXPdO<`e28PeKK1tvE>X{CZXO}=wFC&?d%+MVgqjY>y8a~(<3W5M8{ZuW}` ziNdyMC&q{p%~cNr)wSg|?FUgg+v1nYtE%v7)LYe1^XDxhN1OnT1ZDMIHz?$n#~lg+=VMD17ONq^De?D;-slRHk;P-?oc)wZ`oPOo~w zArOEKvkNWh2VY}ZY4p3vCM7iSOlaaI{*Mww*2^_AFC3kOZ2F3VIAD ze$|TMWTVqxjXG(Uo8(?R4Ze$aHZgpmVwME4ORL@ZsxfLXm%&qE;q%=`t&@byypaKb zj{?oGID!X7PuL`E=IsrBXQHftlFR|oKU*3Nt(0MCs5)RRao?I}%XKD~BL^#vJ0$S0 zcRJ_hJ}f?p5(nU{k)gj8hg$kGZzFAsgc*yRks3gd%6WYd`qhafpbkFMYCQ5Rt)Y># zKP54@Ff6#~{x!7bTwfgv0E+=HKYm6{mr#bDa*H|N(}pUW2?ruQMOQv#94+Ccj$=7! zvS))tx_D%Wau1T4X%jeHBLU#DG$3#%$XURrEZ8yQeD)!_|Buk}J{2Qf_Qd{BZbNgp zdH{l4ewnIr1I$NyE?<&Pk%6%OK@hk)lJg@4zKtHOmVeJ$ibck_@E2nHHt}4>)L`at zn*jU}7lZvI2_2NLcz6n`pRE&cxlXYD9SaLE7IVKt01&n2!3Sdj>!xHn&d2W=D_%V?`vAbL1@5{xxF#`v2nD?}DG(Of_ZjANz#&3WwxzvC zQdVq-PEJ)=-Agg^vxH10+S?=TSC!iPIGNvsU^&BLcub14eBfiJ*9G$1fzqimQb(+- z-hm^_`X-AZuQ=Q|ZbL`lM+d%cZVkC&+RM_hRQUAD4&ihz7Uevboom9ms6w$+ zM%#j?C%v8&ILZ?PNJzCQ;Mx5^@tF>#Nv~?&gK!W!mvT zEo~7;YP61^y)`=yDFgw;M6>cy@oe|yZZDtRDU4Vy?yIL|w(T-z^E@E@m)fJu0_osx zd9A#UaR^seKNxu*hd>8&&vs1+YqOzlE58@xaWruTmmc?r<0t-XY@JgZ z&f)`d%^}_~=u@>fwn=3v@G1dGX(O&v$M9|D&lZdyjwqDz8V2oCG# zq~wO^0P=N|Prn64S%=nD8c%w33rmctdh?eN!3mtDD5oJd?u#^}+%Iz4jsu3fd(nsa{Dk_?vhNv(>@0{VwYh}oVq0i*0{NnGRt%bw`+ zp|ux^d4d4nmQSvHD{HdBCu88ro|(#GnrVWD!3O(aQazHQ^Ru&pAjFp2Sw%JPs#|lI zqjh&k(XK}|DJ`?MKn84adv+1U^;qg_@tRpjlrW6@j{XS*IJb)_TnJWw-P!OJ!Zwx=!fErMWMxr2hE z{H`65n$-t{ShWC>^2Y!afC8YdS{^r<^`mNtb_nK4xvH!Kv*eUfVeRCiA21sPRnJgp z|K<;C#}vDta~euYk;FU{4?!@G+FTcD=<3#L;K>YVvkSi*Ewl@m3KKbx1nanK``J_~ zX_0Fb-?{F6mJ`9V`#Fp}{lL?8n72G^>bjgUIhkEK2pn&~-+xpSDEB=4KdS(qRK#Xf zu_iiZ?p*k&8t+oDSNciZX$e1m{6Xv`dLuYLV6~%s zu_#@~>8bza<*mo-HMZ@^RN(%J=odU?)s$=bOPCqrkfn}-!augDgv@G-Mp1Mu83aBu z)qBMR(j3GcmB$4Y(DeOfKhGIgXiN!_XPH(Pd43O`!J<^g$v zr?GTBR97vqB3kLPSUP0+i84Lr0*eg-d6%5j{YMwbe@eKYfg9FDdUDX^*)_LTyPUt?<64?xpGmjGLrCG&veAHs*nyFf4s(?M0olyN3isl|5KN zbIq(|XVgL*L(pvwr_$HEV(+v!iYfN*#>KX^gJN<*6hlfPglezs>vrB+5inoLacXpo ztxBkznc&uh)~`n0@;NWyf>@X|OOQZTYm_T<*VQV|gPj+xJ=0-M@fmJd6aD)SoEYGH z&Yk2)ZHa9Owcad>Zfpc&0$lZR*>3P@-J~PY?|FH)6lxKJ8W4Mj_u&T!-7al(l|Sv( z-M3Dq_jpm3$JlOR!k4Bi8KK66bL|(B%`KWp??L3x9(&mz8v0~e$wCkG|Izb~LG|G$ zy}EZ74?jZczHrdA`iGM4Z$F4x)mNaRYrDI?yvX3^p3}`1Z?`+=vlGc4Z}#oikdix} zo=LbkH)0joexGXn6plB@?%0y8gOO9CZPA(7)u4+B?OmO0owQB46Qw!%dOR*IrQDLv z*(hxMY@;L9QY`zIt;t7TrT*avnPD46H)W?kl;lbnWE%HE-CiYk5`F&H&tpZ`dj>4M zSi{)TZha4%t;jZnfDK(Vk%r-o?ZfBFn+J7#hLMB+gahF7PEEGjynqMUJw@~O z{cC#kS1D79efYOXL!!q|*|pI6*la>tI%GF`@sRusFtMvq{7WMJxA5A<^S zR}ddA&{ga&OJi$)Cb)O|tOI{*&*bn6IGzqXB>eZ0$_;Hh)5L(lOd3>!JH-uNXkGcS za6WzfZP${dte#HEvRn)|$cisOP1MZx#MfjE1H*Sdflq$20j^R`e(f7Ty`=52%x#|C%S)~x5jr;#>UE;3s8+S+Ghh7p8k`!V4cR2A~LZhdEuNcBHamM=J&r z)frWpWoBL6;&qDyh!7AV1&EW!OGra=%7I{7a-^Of<)IPgNsqog#7(m_6;-khqMFg& zZ$gH*i$_^?C(fJNpui%lL;u)RdmY1%eKF)zA`%3X-8=uE=2cYc8%K9}Ka5yZk;<^6Jo8Zui000Xv zZZejhyJDN?mUvE5b?%2Nlx)0~BnSvH6ReegFCL=Z)E|;<9y6nCYqhVcNDKR>(1F0i zq~KA!5E6tfDtk1k(`WwO$q*hE$YK>$1r_$%`0!;vnJOI)Y_qwC5uN_}reS?tzx zesl%`eX}n&A5CDzL+hW;SB7{>oYXP+Mg{Xh3(m~*ViNLW-@tcqHveo^3pEHB#2jIX zSdhLGRSj}R1jHi00;ply|q20K^pfZR-5KdD5YF8D;!n=v2){`HWVKFd@U zqS^A&9>2bIq%B=67c*{QGo%4|kd2DSrHih`3H<9X$}<*zeSZ~?{%Q-dubzQ#E^m)g zg_@Vl+}gOpUm||D#@M^nE{GCjaW}aw$r8Dri;E?804o<4dp^e4}%xZ}ab- z6ao|4xxMPMnW(qX2|FENpv#S3#KWU?kCdqu%no{nZznI!?GvM`{o>pJ?x<6QA0Ot1 zo_WCerP}x8c<$#f(gnx_rug7VH2Ci|ktSbkljRh}++>7W)fKo_y1uQ^HR^BU@}%dG zvnp1&tKG0wj4p<2G(~36l_R%2@pxM$I%v;dvIY`kJf51aVhB_1mgI~t{!zPNV1n~& zSA=dL26Y7}fT@_w98h;Zv0Pu-tlX2&gjj=g1XwU7-8|C0LMX7QvxB(76IZl_fA968 z^Oe1Y5@k`sauI)lAl>InkQ~^hQp^%wK2c-RJKC?*`d-)TI zakcC-J4`uKPIvC=Ud?*LKjR_goY5UZ=VNcR=2SH*))?3Q%Kd#|rS;sLS~j|dEl{+8 z*kkD>k%kK<7Sz%q;1U(+2$v(GQ}TuTqZT7Ob+!`{LBe}2_qDK;b3iPm>vvB zNc0IkwXHC=M1+1kDm?E>gXnQo@T=e-sUa<~g>ku6UksQS{*@^gt zk9e^Q0*NHZi44>`TcPkOxgAOKv))xsFKJwr;^(i-S$qa-y9r1xApRSyldo7Pp@z_9T z`|ij+PLW)13yCdPjW|{Mlmr0?LWX(_LKKQ(+M6gEjukB$2x1N4WiOu|D&^v=U$Ek= zO*UDM9UO4LrzAVm0Fyj^X&Y0k0I1l}C^Un*ft~TjuT+{ z0a#xq)HrkuEtVyACd&4wK?mhW5ev?kh)uTnCTKQA6&|R!fnfozTK5Id)=dW3{P%B{ z4V2WUJKKQ;%mTa4yE1o$%i}r__Ju|@W84GLfU8(htvX@II~y*xsCqlPD^fb|4QLBc zB4J((t4S!INN$%JWSj>Sb7@h}rj4NVo3vf;qd%;qyp+^eWd>6AYll!r?JJ;df~d`2 z8fQ4-WE^Ka3JF__(+hVSEvX}^j!mPqJQMrSQRyfAb}T2m=t(D+FqV|#M;&*=m&FMQ z#hgr7$Xh8=Yx=QJf5d^8Ia+?K06ds~=cTwWltn!U>@KgDqf7CGxr!ZNljm}0T?yc@ za2`Q^Ez>GPPKBXUObePL^UZgSN@w&Iy{@a@a;D0FW6dB_3bT}+0@P7RRkU5R5&G|F zT-^=%CjderCKHM8t`3dTA$7XWC`r8Qh566CSX)nO zIgN^y#S-n2Ae5-jjNXu}F>ZapTe0@I_>YP`RhL5OTI0s&vR*-qb8M9LK7d!g}zX;8wXhx^Ftg zC9pWJ^RP|+e8W_82y>`TZ>f*KyWh(6FlW8=z0R5MMwz1|V?pQaDsHm%95pxMi=-gY zEJ5`6ARY!(bv+6fUBr67sI(bv&IunE*iabQS%RTO;>M>(X55y)&2Z*c%Rx9nijr5H zy-XQq1*3nldr3H=JCC)TF(MtBO`PJzRTZy}+PD&yW19VXAi-wl!5Z zHSAnz<93P2`QzA1rZw0KHI5IeL>U`EP!gcrOM&=E+s3&bdB>CyC!$n!1l$z{H4 zEw?Bt=SJ+`la_LGpLLMvHZ^cB2wg#RM$U)UDmU-N6I(rG-b5>2`kNhq$lPcn*jgAI z1V<6=*F>;Ci-T?_paRy0lRwEq5=5K`VOjW5di|&Kz2Sy zJe#lc_}q&ODsleVP7dJf1K}*Ga{Ah06Ff@r%DoYWXz>uPp4~#ts5^m7?bEL~U!))? zKgQmXHOPnko#Ez!Fzeg0#t(DP;i@mAC?q;6wYi#+?OMUCAdxb)#u~LdtK$A_3oxND zk%60X1)L!|Zpks!RRRc2@^c8T&P4&*wY9)BJMwMQZVhdjC5JL6;D@&e`6oC(;K z5M52JLofVhT7fXnkB>RS(^oEjTZ}u=bdNT)mk-;_*0u)-cgy=n1#TZcE75f)_FHty zR^o*AHxx^QuL8X4v#v@IOmJeGqbx&fDOTf}6wD6e+UZIwPQn_XpH_;XIsvw05NpKA zJ9VSw)73f|DTqtr;Q8IOuL&C;2MhVE|Eh{=yh3zpG*zfXK*(IQE$&mUN`obO#@nO8Cl4kV6bW-t@^?HYWgNP61^slyEhco3&Zt0(duhH77e*xm9w zV1yISUp&tm$6p=c*&+dC(X)kuC4|X1TUJ)rf?Nf7C9idd>Jhd3H-~S0xaaJ{jfwP< zqy)q@v^X4)%e(nc>#G2JhbLvKIM2=m8r~s>{+{>JEftOQ)rFHSr)nu0tx|{*?2uae z{8{G|y!YP~HMp%RcW~t!FTPj~z_#;=oS7DKj2Es0$^I}`T%{qqeYH!Lxon>k+PkI* zVRI)*B#;mZ#l6>&x1h*$+l_QeNNjX8Vj|M?d~RK_b#+E05yMq`FlEsj#99{49Wu^~ zi3&CjL;VW*thV?#3pACs6Rp=v?$LXpjle;bHij0>76S@f{^p2LhkhE3g{Ieawkk`Y z+BYzOMMv^4`RKuiE+Nl#F>Qe$i5%Z@rSJs>r*}OK)U}`sbvZ!K9_{{+%7U)jmYXKI zN8AwXJIs@H5;z>5ESaTl$N)Y@tPE0#3z0{Df+eDwH>SOaIETqE<>gI)fJH>1Ew9Ci z_N6*T2grJ4W#)cEtx2Bh!XJ%Z`GCsr|4P9{qKSZNpbDWOpryGo{BRZxj_<--^mdKq zFXt6fx||dC5onz5aJ{TcK`P6a#u|7vMcwt7h7j-tL?3MmNC{<4IZ{p@NEq3L_v*x6 z)4eZ-*ZmVB%A!<~p5>4{mi^DXMy3~el9AmGOreeCv8|o#w#sxnYv2i3@=w=i?7l0Y-Q-2atyk%20nFhv1cMXEAc-Em?}vByX8wN$*|PW>$!bD9IlN*4PN;* zi3qic*ydkFb*6HYS3!Y5T0;J<@lpsf{uB)RW%5geI$pO$wUFrzvGWmc4NYt%90Gr% zyS@|)=(+Tr(QUpCnN4UjnUWes;N*<=MoRJY2GTb70Oade1A}E2Z^SO5?!%a{MT^4Y zM!|6dhL%RmawjNRtXrRp0;0~eOD5l)#9*WmZ zK$_JSNHV@2S*rczONdXQ^aA*lfRAih#*?q3>8Ft12hm3y4@h=jJ>i@*w1c%y%XDzpBJ=md;$A{{wH-!qb)kiQIt zmPKj0%PDw=qwAzvDo)flQH=BO#8GQ7Ev=vTP==fV*uU1swtN>Z@5HNE4*+rQ7Ki#1 z3MT7oxuZ#f=-OYh1Ic;+PeVS$hVfWp!lyr2`XOlzyOEMbBn1Q@dR$bbh9@TW!{fXC z-kU6FC+TI+*B^&7(R3EU98-b_hF?Hk=|V6<`Yl>#>nyq^S8?-poAKLv^F{>YqDL{C zMjL!dqCAOwnz3KO>pn4rFi%CGieAfr-c4s{lYkL7b9S^%7sybpN7S7>RN8cxAx&@0 zU)}$~N2Q?OE&#aT9nQ`6@1!_fQoDVq8p1dpX~k%`hGH!C2O5Dd z22{P+QTlen-c3J0sD zC2E?#E?gD9r5gChy1B=Lz=4U>+$EBnzD$w=NvN9TLb2d)XBdjyo2qh7y=QoCji zOA6_WXab7Fs~{vg0sX7Hs-Hg&PDcLTbL{cBy_2$?rg~woms!Ks+IT9n@4{-%8MjtM ztFm7`^6eqdX1?Kanps;ly0Wu8JiBi=dVW*{MYtycf1-OU5^Ud`a{twunb_-_3jiIa zymW=(i~y(PYxo9EmGycTqomE={=9U-^GCrkz;*4w$gNS-HZ5wKmHAl#g@Zue;3*v= z(--l)1*_vYh=2jLi5|6xfM=uF`_(kLulxs<41uq;InKLg&rHGJkuj zUT^SuPj}@cDusF16#>yU{B~D5UDz6)hw?=E;*)C#3F;6YA$~-+^Ma8d5qksoL)^AgJ#y&LFaq&Sj~+6^mQD~pRs^Z1 zJMrw{MA1#ujY8@L*n{*wU(cEPCv&}wzL{t-clWQ+8eb|!7wc{4Nj8;xErXf)wvOPy zK#fVUD68zR>868P9}{u~&Ffi=cbnj89TvaUm_KEirv`Dy%G{4fIJVE@>C>5+?{dAp zm!iqv-4H=QPVI_eQo$_aZbg*na+h^Hikra(ic&Y^-(~!1f2@-TEr`DWspMzs3U&lh zSxd2>1gei&{=!#vNV|3DXzql#Fn;^O-ENQs?uY6Itp4(#EHhkGctsn#B}V~y#FyI( z0R(<5V<1Lm$kNosn#KF%eOBPcyX{A>ftisxe*V8~)^H4gkR1L=VRrr?}s1*7% z=8h-6Y}tn5U|J^<7qnVeVGXgRk6EjUme>L$QCY-dVg})hoU7a&MjrBO#3w~6{Zgb0 zoh#leJF1T5RD^Z&*6di;v)I6P(d6mi^a#w&FY5!7<#WMvG{L0^Zr&d@0i0C4S;Fx$ zSNxP12dn;E`wv7(tnlV$eB0}1ed|&e96BA1p0oY^x$0$J7mE!pmD5>hfB0|o-rDb2 zrx>_MXR`-*C<`C8-KglL+B+L`$?A5RrbH+L%ea z9MNdORMr>34nLrN^*%m82;Z^o^^AgH*4}OwTVDv7`N_7X-*tjuZsqgSfaNcSj+`^| zR^M?IkL8(e5iYpKa+zLBP-!I@f85RtX?vfwHSWIrmw(eBA-UTV-IwGe9k)A!t_0zz zF(1_1(6a8}92Ap)rDqS!YyggHG~V%D!~Uk(hOqx#$T=A#-JW62xKJqweJ zyE5MO{Jf`M5+{`Pes4TdnDC*;s%BKD|IEK}$lsgG%Rf%5FHZLR%Rf+VK~OhOGKnYu zvFnYt`r0?MrX-gH?GT-@QazG4VkFe$wJ3NsPz14)eNlbx=&tYVj~t&fY{4UMj@0+} zEe~}y(@G3uPsKRx_-)_!3vMnU3B_8(f`{rMig{CAm91k>w-(qneMytmm7P2h5}Das zBQ}zH0cMY*tnmShdf79lQt&Mv-yHX4dT`)zP6}dc=HTQIU~~Z7f`MlHSN(UkQox`f1O1+F+@T~vFzOH z(xW7RJvyK3!~ror8DY5NlcW<6Z3;X4tsU$%2$YQJHfcIAc&SQp_n#|YNNe6a#g%}F z`4j**deTm2+0(;VT7ROPet-aLtiGgHX$amK{*Q~`?af{qY@oq-uLD}{(>cphDor@N zV;lzwHLl&uB`Lkum506J_g_Hz$Kl&$X?uMlzXWkNG0_ttb0W~qU>M?cmUa;ANYNqM zmd63uUn%gsFDqGQ3@LgpoC0zp8>>0Px?Gp9S#-$}fEniPnW|9lN3;AxVG0`AJ#enF zkIBN$`oId?uXo@%1E<$-(h}-^20Wz#N2Y8>|J{@k;YG=ht`b>71))k2aYqQjim6s@ zqSCE)&sJcI*wV90ff52~ zOl828b>-oRQV%lSQ^i96VO(K4%<9MvIZ`Le$<(Np^CxZ^TKF#mL@)_kohLO(f7)w#H{DE2%U?sRS6 zKb4Wz4BhPwQA`Mj!#Br@_WFiyIJHHfNMM-ab$&Y~gX`W-t# zm0TNwNjz3yA0&m@U$-0C8#st?&1PyjBjP?Qs-I&XO!F38@^IG0IPX_VTZntkRc%jh zo@|-}^Pqsg(M{#gN}{?yYzbEE*im_U>T&~E(ZiGI6v0o6cs?W(v?h*VAV(7kK|>Eu zUx}9{JxQW#z3wF3iSWv~638_&MUec$2wPGHWrb|jIT#d01tcX>J<}d zZtfO%MGw$R;|B}#(6tvI*y}#t?XYAHG_J1|rFRa++@rA^*GQ|`uX&SFvM2brdr_`} zwg_Z9jV)*LDm$}=KC_%Rrg6iG;gN!>893-&vE{J_oCE!WkXn$+T5W=mNI+u`f{{FA z8a_Fvj>09Q6-Iwh#U+pq&HzO?fT(-d&ZZmJab#?)?cx3wlhc7JFE5YlqyweNBiKsB z<_S#wc)T?>gTJ6Oron5KH0EQmlx#E2FZwIZZJRyxv=@|D$cyE2?5gx7_7x>6$8DnPn8lj6E)$i4=@Xs!fOYj zs?NmqMa`DkOTmbBq}8a6?~a7vUX7*nlT!uJL>|V#vs*lY)TD#JV#IsEU=UU9YxED+ zXSKgr7vp{oh_0~h+ImYS6Oxs3q7cRyTnB)w!1QcjJLqtbk|HO;MDa#foG>niCw|rc z=WUeTOQf2D==um$b=&11GZUwNV#%AQl#-I6k)GzBA!9?Mb(Oh#9tnPrqQ=Z78>=T~+RY|ggPz~ppw_q#TkI_<|9~#I>I4h%Zw-6+Tlb8N^IRIWRXDf! zL_`vC&Gs&ln}chwA|u|{dgpYO|H$s6jhYX)Zih`inG)SyVa67XZsP*DM#6UB;n(u( zL$n&ckxC1-ZXH^+{yWqf`b~3DL7BGo{jfPh?;t^}0sHn;F}wkrv^ zPJxqkzkpR{!tSc}7UeZK0TX*BDn*EVW}4rYZ)7(g=yG5+6u*LWy&)vu?yFd`SNqGx zfqubSdeoydbEvpBNrB9)vNl069>Hrml~&-N2Dzhx#%PHD}gr9Vg%llze~3@&nDo&wq3H%ecq@swr&C@f3IT~ zT_-h)m?OA$;Lf`KkgM5#{@lBp4DiHM%a1fHHICf5h^R1}B7ndF(iJUCMD%o{$5Bv_ z=f3xrWqAk%?32X)ydf+Gr>obCWE4%~S zt);(-R=2o%vF+WY>Wp-SkD}pllZ<>W27spT?nYvoWBs5JE@nsp^fF*02HFfW1m&H& z03rwh8ofEmaZBl*bZHC6sHynSyJ!v9Bm+eRTRHzA)zem-jTIl5=8VAI0DS0}jYxy;fOFF{4cL3Y{x7?*U>|BhZv z%FVBS>@tkl@|>h<%8&fggsFaJ$V6YFuO}$ybhv!sHjfLEH{pvX5>z~2$#CqUobIXp z`90~A;w~WyJnU}>A>*6rLbJNrrcFUvqrQ;6MX7Q`x^9Tebe|pNldu>)B0hX9bP90$ z&Zvt2;k$d}*vV&3(|OrT*3-(tRLY5FeOM`x#9Zv;^nFp)8&?mSmMc5CZ7Pjwb$%J6 z-Qi&6*?oAx&9eh3r;pU_>n2BgkX8x6-SaqvqyeIr<^QvwfZDVr;S41r_y@i^Y}$Ih zjm{4;NxZjFV6T*+7~UB$7&~@-&#zc5wJf$Xwq~XeSWsX!vW`*L1rRPEuKpb zBvEmV*hdt_B8f&f!sAo8L(4G7rApZH*8zE50?{PTE?=zOba1ZL7Cs zq-N5YBaxI$XRHo}{f*|#@YZf@W{?oTafX26vf<8?Cm|EvJ#7#TN;_7~#p=KqD5 zj=B{Q_=Ec_M)%x#Q3$^I6KxOJ)tWi!NsWA@fx)$|*8@k(ENLenJ>aBN4L8Y}Qlb(q zh#lt6Je*s8Muz>d=*M!74UNmo>|1;!z@Enrw!B0vXyNK^ zsCPH)g04oK*g;MS^{grFzNYNuer8UP$o<$7%ppp~X?^%jzxiD;YgqFI=-KfOQ!Wz23VY?>5OzZg}cnDWOOVey3e6 zq~&s;`Tev`EJz+?o>$ha+V$2i8_$phmjcB1lCS$u!7tm1D#rlLH4u7EhvmLtjoQyt z9~Vhtj9R3?)qIF~fOqV_8x)kM#=4Q8Cb4((b2-E|I9UGfaJZH#J$XeqV!$x5@5*XX zTdnEtmSKlUlzi(*KYq|UJa_Ur>D0=VQ-SNFIu?&I!gmgefP=>yTo>C%dE6?d*(i4> z$@o97r3}VK^u&3@orprAwweiJLukzYvq*lDSo;;1J42(2zFAjq^hHaS;zGbZDm=7b z@eu*!6%VFZKHTXD(~wTFYUDT{)`+Thq&*uuZo)TrGg4?pe4!jeZdxRLQM#*f+SDMg zTVDQ4LQ!03BzXYtua?^xxKA=)!Ypqwqwf|900BQ1H*&V_*PyU0 zRQ7JZmj^NCXW8PRR)an%C>!L))La(s0mvNYC<8oDB7u(ECQrkH_Q}>|kY1N)xXk2a zWl2m;vPHkCpc2FlxZ&JVd-DcpJh5StR(i@?iER{1H}%5g%Z0M7mEn0CS)CLj5^C~~ z2lObCk{>NoLf-a0oVAEfj_x3!fWM9JE}(rGtu96ricooOxwKq*jn-#rkeCF)PJ zI6<7M4|tv1B2$;ZP9Q--UTi6M0DN!b(iilQ%ew&BTXsCel%O;%pv_D-%lfMVB9%m& zYCl9#=O(Y*2;w}I{*x_#c@sz=h76O_+H>$J&2NIl!BP9fUDW8n&?Xx&TMo*P^dXBD z7tAm8ZQ++I~M#=k{~0*zFo~Y<}B#_&1}E3-O;kBX;g;I zV_zE%q>YG-d5!L_6+Ntn=02`)KbZ&**L%G19LdSPZbrRcg$$F0!`64;+5Gu1Gbl(`0lfqm=TT{{nDs&b>z0nUBoTWW??bfn!*ph;99L zD(p-A_G<}3iE~#gvTYM_FEXGsTmdWDpNBU2s~J&LC{xcS3oU69{}cc^((! zbJZ44zr-epURXZB2bX!W>Kp|ofM3poggONpE6g%+thom7?oq`Mzyk$j99E5gftfc( zZ*o#i7S$?%cfBNfzFBPObDi0J_k!t?<_&!w_3E+nAr~bifh=5Uew56~V^^tdvMM4gG$S zFm^FXRtkUAPzx#5JQb(2@#TKs#|DsJ9|UKh$02{|U;P2x*cGCjuPdDF zQCB>Keuyg|X{PY0+!X7K&XIUe`M_!n0|4lEp!L5^>W;}!G{ppyJ^;vhE!lH?={f&j zRhwb+WNpEjs^@ZWXm9nm587{2xXIvr_X=VIw_BB!Z8}tgY!GhjV#7-> z(W368mN>t-M!Zm92?k9yFsjSXt&MMgNVF*q<&|-I^=nz-E7;=EgEU_QpV!?QBP@wP zWSLqiB9gg$`3X#&ejF}{05IF?76X{9$r&w5G&QYyhk}>7Pj$`}aF#pq)R~zAL-)vR zX3+sKEx|sI@=J$!aZ zd7yQ)qJ4fccGQ#_hY)!;I#{8ozFb_mn#ZNw7Kxc7Zp@kR#|akVTk06qIC2jY<(D0b z79b9K(U^$O8s*97dhf7l51jraU0ijZywb0bYL+K|h{6uL0nMP0``g@yOGYP^CjNPf zqL>YAO9G3ofhU&({eV_8@r_R#u=f+uw*4O^0MEKnZd4o!XXjG@1s8=3DJrPG0jGW$ z$RivUhu~f2Nz*eJW59|a3gbAbY$6s;6(_TOIi?xxrx+=0-hdK?HZ_hr+%emwmP^8> zUscrj=VS>_X#t60z4ml$m5B7k@J*=I__UKLcsGTXO$gunv_9QVKl%_bRtEUUPud6n zk)KNbyt-RBhK8P%-F&7FRR8E(6!6RDn10+4;*X8iHdH9U9H$wPUFVO#UTLr1HE@K% zTv}fDfvYA)X*aVurLi9=&8;g8beoUhCO96c!lL!SojNGYhgB1H-*Sy>q{lVH>|GBL z#JCF-On}n=_efh8Z)jJ4*2uw!!(%>!fQvE;PLvdCQV}P4&_Rg@ ziO+--=;zp^vb1^6YfL+fNr1#!->Xn7o9C;J_Mc#^8=p2Hj?|SFJN?|Utj{kH*w~D8 zIKJmu3~eKHE}h%sM2S}+jt6!BT7+7TTd`!3zQCGvR|15-k7?mJaEM=otRsnbavtj{ z-35tP-ZHy))NSGZ=lW@U42!>Q1O$4lPi_K;_VLNI6(y~%s7&;qf22y)rn2!DO5Nho z3^8qmh*`VqD}p_0bP{ z@ajp|2g2T~-u7IkbZ!dOS#$54J#U22fwq&XAO>_?+h1@X(uJ|^dy|)wZk~2u>`MvT z!^zy~6IwKX>CbwGsV{l+gCOGE;-xkYOQTfD`rq2z!_o_3gM3b^Z=8D}l<(Xq{68RU zbZ^4&L*OoipC%#^P+?}XLXG_uu3{(J`PG{^L#Oh$_=q+FhsU3z8NAnu}=>N=EQ-7!|2AU z>{mJcx0nhPalFjG`iBSkb8sA97>E-PQ^W&(T^1T4LQfKy zx%JvdW#If5xQ!^$q`D03X;rvW>qo-@l7^u1CCN_uxhMP*)>J#L`v)qgJyW-#@GUfZ zt*o5JmjtrS)^7<&pYhomawSX>asY$Bhsbf26D zh0^O8X!rDyWH3G*%gKU{y=`CU=?;Q0k=EsTym%;gECB-Gk1721h=YEM0Zbgn_TG`-~?Nf z^1j+*c99KM7ZEp!x@^hx6Oak-lLGdwyFtKBtokQ1n*#2<9m^J4BVRbf2%CN;FxkeY#U^spk4&MPNlxG%$p<<-N`!afLscYc`L6Sll z*zKfA*otr^HE!2Fsa}AO_R5I6W*C1c*G;CAK5Xa$mPDy4X%_EyG;J!0jT64`yvH*) zb1JC=hj062k(=tb-Y)m5zRG^+=QFJAU~&B^Hu`#7s*?t+f1P*hMJHQzgfh6(tm+WF zLh2dri*>wIrPY-Td=a)XMFF%qozls%Urgi?jb(b1dxIPklp6^wew)=}7J30| zTyRllT!3MUfQ6at97`zO-}X4NjiLN+r{zQId#$nLm;HF}h7z!FQ1#jR`mLj4SIKDG z$H#0A`cuPL2)<&TvvV9u#3ms}HI$_KFLkJq5wmdy>^bRAdL+y@M6Bcy%GR#75eNug z1FAGRLnTAkM1Ca~JrO5!uwXcDA%8#b-DF!>i8x9~N9$Dw zD7-)r4mfzz2fBf0Y`oeEZ@KFm`^*maYpt>?O&zv)$sXrZqpLhx6INz$EEL_`{i-aVq0pqtc?=TGQ)#htQ6@&q{I2{(Nc-6r}+0LdzgIT_07O3kLm9rzp* z_AZ6Br)po@$ig4(gj}h=c?T1QA0BjoH=GI==Ku6bU@X*&nPaC|SHDZ;%O4FYtB|kQ zu#&z_jWu4Z{OT+Tun3up?zD4pfAH4cF|TR4aY;cqt4bOL=BtbTGlV&6Rw9qoFf>gY zUFrY>s0ys$C^gBXXoc?J8(|jJl9UogQb!5CSF1 zE}7{@r__ba^a67Rw}CWrjRr%|l@b3FR}!55*lsqUsG@q_^XhY$fd2Lat8ENzJ*JJ> zTUPALslO81#QIg;p+_6pU3SHbA_QY}r9zJXdz06Df!en7d57XUjpO}|B7@3G42&d^ z$de7C21}Hyn%eHS5}Ts6EcD@LuK1rm(>@0ACa$QiX)Q$F3zJU*bLv6ov!$`J?1JKx z`H7FpG1jaTjv-e#5^jCg%G5cY3~4M_4HAy0AvZgWV>0_*&}XZ?pkgr0x+`mWVehqW z)VDquuQ9qeBO0JqA=TEg4Y6qX*1>y_G#bkaJ3Ca+fu(ii=$k zzH_b^zNvbt8Rj1vno%^!llHC)v=VoN)8CXD)77#Ru%W_D_mZ#%ab?Nb*^DI#6_p#} zPC*%#JT%U$TqKm5FvUDcJ+Hd)1$+B3qKlMc2(&wy*%9ZbE235*6P174x?Noluz1-R zLpW)$zuLI%NMPfT-K=|$m)|ibG;6#8AgGk2#|QNhiJaFj659vCMgsPe&hUiw;Q|{D zh-0?*^$t)M+J1sl@IXK_=qP$1goGZ#I%sB#+oF5Y`=UG|9xUg4!_+nnJ7MF%WNW-$ z)&&eErC4~ZP>9aulxVe^WA#ee zVvK)2Q!yS%j1Zeu-7U`w5oAB=>eArWM8Tfd%}*p?#CA6^X(>+SykOhTuW-r9ot%E& z8VjrLq&(T$Aj`39yHxnalex7wjbGK+$dRg4f|g!XJg}`sEm~0EdA^?iasNbMV;iP2 zLZ|%sP$1D$P7p_ygQCt=E(1GAj&`5lM}uH7Qd;V}?9xCYQZ85}JAp;!7**DCL#JR~ zCF;M0qQJn331um_4Hhr2Nl*@xXNUx*T+5O)u#W(y1#9S;>oFu&5Ajf?s`kci%0eAv9SbwVGd~Ft;|`^5m2`kiXt0BB!bf z{B~Do%%MFB{k@&M+)9-*%y{nU8$E*zdZFt~%SEXtX3SOG$T;pMSn<4%3L`{p%&=m> zsv-`YRs9eA`@Z;5Ec^~{jW3r%my&RPr(CB}_19u!Hx!+oggn$zRgDC;S95%R=5C-^ zGbmm`fA@HC#K{x#cMki?vOuxv(%@W2?5K%VJaD%O4wC6$JQo841tFj9Btce<`X{x}^}c_#?xx~{5Z zNJu<&a!EIuRR+kx%i--t*WH0J@Q5$hJg%1i^!8LE1VPb@0(60cr?Sq(8sArwLO=7%YZqu6$v zQ02qL3EtpHh0R48FPg%&7vXE$UrQ4w5R=km%BJsfA(Ba)hpxt&e|UHgCr`%Y^H8M&(`YW1(HwIpUNOu?LSZBGV z{V2mIP;sl)BZ;?0_hhG!4z|12^Vkbd6&=$URX0O0hp$|p{g00_pdnb^PoLb@6XXMf z=C>*qhi=7PCqHoGn5vtYNIPV0AcqrOw^HSNYq@h2=oc#v`> zP*vtnWm{dXpEp11k$`{y_rJyFbB@`6-JDt2mg5PH7=Ji6EomtK)O?*@ofq0d-Tb68 z?ctWa;h%aXez**MKYF!&Kw12q0EOb@iRu9i>T|KokiC8tK}o}7Z(}36^U33>ZtbV_Uaja{_F||Ww8T>;(CEP!SnV!v?Ml5Y zF>Z7sJgFbg2dVbnMgOmMcD1Yc_?OIV>>By&=@C3saQ1@Vn_Ib6#-daO8O1WD&6cX2 zOy1h8B%!>1+PongmDZF4q}+J_nW$lA>NiF^0p;LpN0P8;FbNX`Xk9!=MA`H9F!-VdWo z+~di4XI~7?l8ubE#6P<+)y}3)n4=UF?ux2>$S zaW;2qY0CjME-HFHPoEC5GiqMD+V;18lNqV!97y9nul;RnCAwtmgQ8%Y)cfu*PZPa- z?{$NB9`6LbI^L0ET8ubl*EbS7t7tRf z_dFBjDn{LGqQA`#CTC{H(m|~iUUt19)6n$+jtp|8B8;yyO$ciBgU1=Nn5>sqA9s&M zm2GhDDyjOe9F3iq+{`r*dG5uJnUmArgn5+qie{2(Ml6QrL>E0$f6lzUHKj*0V(IjL zmPqBFgxlA+KXcZ>M3rT$&mm&3$E1ARjmpm$s6)cx=ICbZ?(M&@wtZfx&kJVZ!W%;q zcy;naIW)(S!KA|X!4QSZJJUo)9|L0}`X^yeO`dG$G`A^R@|gTm;O}zONoAMTOwtyH za4iOr!w5aF7=3XKJWo$+-fRjyWHX=pmEFfX+5-+}8M#Lu<6(0}Be;%TP-N9wfux`+Rq6 zKORg@Ce+^)WL3cqF+O46-MJrB`x?`t+xKWUBF;*Fs>&!b-|LWGkQ$w2W--)r_e4K# z`GB-0lsY!1Eg5T@O_sTLYs&0)hXolBix0@)hpOvt!c}#tysL)t^r}aM57+1(3_(y6 zcf~PLpI(x*Kix(B-zCr~*)k?SU-3C@8G~C;2jVH$hdubNyu&~OAX@Z-GnMM8&KpX_ z^%tWZSxM%=kl0Ek3o$d_6`QhaV;$w}ThAGL`ZSCnq3DkHO*pRxNBQ}0j+xh`F@i`u z&f6r7$PnIR;nnk>2}FW(Q90L+R29A>s8j24{_IR>E4+Eiz4jgZMpjyQXxxm+wd{pa zMJ8#em!BW!d9<5W8gE0qPh-s))1`6qIFs%F{6g`#Ah6G~f7q3?s%88#64f@Yu7H{( z6R_j(pPF_S&YW(>lC{y~I@V;^;GClB`O1Lgi9va7?Tz@hT3l70ST~H`ma#cPbn<+{ znbvo$EiAFhntwSKZuIGWIHV|u*J5B5-zPdS2$*lzbH0s*T)8fbpQLL~TkjW`i3i?rhMMN5DKm86HCe4@> z*Zt@cp~<~M#PF02#Tt4xWhTHsa6`gXS&AFS7&ZF1bKce|_Z5u%6ppjlh)qoN>z6$@Fky3uqz^)HL9r9bKv81E7{7XgiLZ%gF4Jx8IO@(j_Op2 zB6CykOBU%IK<%J>;Fr~eGQ=o==4BmuS@jg2KyhB;F-x%iizwFd8;!wCD4f;fK5|4y z26;lC-PE|Kr{OI!4R&Ml@%vxPJt&(yS$lJ)e=OU&9%imrIZaDCHcr{xbMSkl=?{*T zIj*8S1ULq9i?%!Ul8D7!e+_e&qx?6-Z}$iZ$NoL&>RDM=S;h3B*WYRcXkDVzSV7_K z61Mwko656W_z5Q=CvtPbg@1WBXB5O@I#9Q4L;MmMD$9P;i1VnB)*0Q%UzK*EeGO#Z z|2||A>gY-qiryT#QrK--EZ1yx9;P{_HT?%1{g`yotPu;kSq0#T{9RB5n>&H&Y>%^7 zH45>GEv%k;0N0ag%j#Pi)X7l+&l5R)q)N<)W4ghJ)_rc~Pmw5|E5`Te1^1emSB)#0$wdguXus8Jod{X2YqD z(2wquKU_0^HxkP_V=@Z-Wv)0Z2{#4PviT?8@@;#RV*4RVM4zt|@9^@l7erQ0(bGvnvdP%z;+lHGuVKdKC zK!4C#4llDl<@NE6?8VdcSStC>?)cx{zs%mudE!_wPF_#%c?3OI=3iyLj{f-1A5A2V zOJshPLBY!z_v)zNB*gjsADir+Rn&XaF%05H$YwU_?zx0YC6-RA zTSxVhZpS#OahX>hn>Ag?Yrl273^efuzt&{pV)J$~6$&zTqNcaG%i$gV?&xF=iZ}R@ z%LJ>Ma>%w%BZ9u(rT}WjVG@Ey3q~_{9fR}f?g?>w+tSa14+mv7aecn(7Y2XrwvZd> z)AyvfIqmMSG<3WX;nJXpo9)f6GyE8yVG;w*;K>3NqXK@Rzcmif5Aw$$%`N=vJe>By zwue^tN4C!-*4BWD*xD6}`S;>Dq2Z3CEBo#%mq~$3+5=|V&Hf?`m&eHO-J5&J;;eE% zsy2!yA`24roTzyw;fr=U4d&RyT~puh?5=aG#8w#gpf-X#I!4XdVg|O{&iklipAg-g^$qO&?z}{6gyHD4QJ>1UAsh8%j2n&OpA_URIk)K z$7zOZqYsw<5+6N(Me;S~FAV@aROyXdXij!T1QlykLY3jk9@UST>j)2lmt8dN(NmZE z7XVIRq9dD^oYJP}b;C-aa8@hp2p=>A9R`Y4wWL!7jXwZf(0K9f#58t;L{z9%Pe{Ps z<;b_X)@<=YK%Et%(i_^-(NucAe|m*C0J`CWg7)z&h$b$_VHBM0oZH-%=t{~DSO_h9 zI55R4Zf&pR(M2mcuCjkOLQfeV?PeIAjh)i?yfqCVf6q_UmNlX2`Ga9NcQduktlPt* zsBFWn$vH*5xYe3)yo`c6xnh5e3oH^-45OH_C7@XZ9-=0mhe_h|ChWb|k9R#5`*@u* zUuC)EnQnK7_ve^DMWdA;`=o-eSl^}Q^{SWYL4m7f$rpW|AE&L+_)Z}|q~3uG-foea zGsL3q$2i%uOh@GY5%cM3Ywpm2gZFBtL}WEgMgE4T!*C2e~-WtFK zB4z=kpA(NZ4IE4A>(CgjX~iA$r|an{1rV-ve+HJ3*?I&>*|g$c+e66kkPDt4Pclmj z;UG6g#*|n0g5PA?)FElHV@_AgX`vFX*}|S>?ruwItB^>gZ6eudB`ofxr6sj;BSi1# z`<0Dk%aXq~&vPT*LEpZ;@&r+d`+%R81|2hyE10D~vx$IkDz2M>YQoxrSFJZ~F;;P6 z>2x%YXsEf6xMO6&{w6Lh@Mr33>r`Lo8;m4yz~sjM!p zIcHtCne?<(HWw+ZxPMpfs)tG(T%3{DWh4oQe^li(8J~AaD6w?Je#iHt;Vp?c3e8J- z0X@;zsv8@R?C8V8c?B^WdCrBkRtkKZ4>egAcGX^sX5A$W%8%!`OxNnuFi3Nihzd1J z~3WQ)3xBvA7t) zZaKZ0WSZWFt*WKJ^faLR^adW4s5Oe1m*Iw0RAEyxA_Wr`YKi)D*4G8oU2bRh#S6bW z?vUm?sQ|h}+X-58er*k_=K=)dKuS)DflU@V4FTvPPUDY`h>m9sDd`hm0lmY@bcM?J z>O+EI_mV64s_Kx-y|;_2g@VRx#PTSr_<^HD*`Br=r?<8HLEfY10bXxB*%D^h4=+2^NNtFStd}AqhD}b+C0CTT;G!^w)cbAg{$hr<1wJZB<<;NbipigEzM$YiVqc z(^}u<9P^ExOm&xgcjb)AIK98FP~4mw=PfzYoB|$3@-r@9!iT3Jn{fr1?@8=4sE#VgN}EbL9+Ew>`eajmq$Ot4@crA zp2i1rM5zP!KMNq#PLFJjDBmaNaI*0$o^0g(zN}Imy}wk2K#c7W-hdR;|1s)SEkV;* z#LDwM*6$R0V#K=lSvNkM+vUa6u-skz+>ldn-J~w~nx)&_UOtrC{K%IQtx)j&fS8J+ zG1$;=zHRNV{v9qOTI;vf?liGhpf`fdYC6Jy^2f@HpND8xB{Vrq?tmi+xmO)MIFQ!l z<56y<^h1Jhf9gbheC91FWBVfSV%Xa_^22&UbkEzLa#b-!Lbk%%D!9M<=M_Bi90%RS zeU00{sU+l5Qjvd3hG-8Oy&z>&tjg6D-T`JKCe0dKHS}OY)%=ge>jn6-8AU{@VsVI< zJ`cLcDqkRw#))x^>l_JkC7u;~|KU>8kI~e+0zuxzP!PMJu(w(j6q4%&X7wP^_Kl46 zfRY)0z3hPouI|3s&51tE7kmso*c`3I*GL|#@W6FPI}9!$UFgq4MiCpGzl>L(6W{vT zXVksdamn?y)<(us!^xBEi%xvdsNIekKm}Ob)GzeZLJ-KeW!3rbU-_dyw)z8l&*>8~ z?tl?nPF5bFsnmjk!eZE%NA+MT4RGos+me<{I0#^FX{}^Jm3k|LDPdE39g#!n)IRsq z84V%Ceo%IRE^*Rgzr4z{87S4;uU;g@t2H5}IBC>;S;-G6Zi@x@v+x$uauUjBS3?Gq z{2)m}JlXIGEN)2I|1ajra!hirU;9!G;DFp@e>IlS0Y|2BFZ{i0`b6?}e)o9Q@XSlw zxRYAgf=18yNIah{e87?MCn@A^8k%7yd2D$+y0j)Cr}(SU^IEQg!QMQU(Y6$)VR{8G z=gtcdn_`se8C0zF{DNh(5Q9pJQfe@fYhSgoFYk=K|9o9D9g0A4+>9$!H2sMXOLkx` z)#Gu3qOoG3JfkhuyG;c%G)4uvSmu|CX!?fDV*Kt-z82w};SUby5gNU17cHkd4g29N@j z9z}Yi(UodqU9bk1c}@WVyf(yCs`eZ^1SuY8^SHTOTo&6xKBU;kb$pl?;~noJVirSb zNmM=14kZp+bjtbUY0#fHdRFU`xmaas%=hhp}LEmm;KdzN7S)T3G z$YmDU8f7ilnd{f?oXf!6lrKeF^{7M_S=+XRrl-ZfW5R-h0^F*qJS@C45w>jCZt4Lk zo8EwBPjNUNV_#HfO-t-htT*lpzpCYIdiWk>oH8HNcXu8V8WWRL^y55H&5X3& zba(rttARg3I5V>=xA&&{1I%=cI#8#^{fOEj<6h6+rCf7z*z&hK<&As6% z#fM~m>9@bASN@YkIB-eXTcY(?%cHCse)s(Kb#@yIvHeZ4E16q|H~{JwNpw5QkH6n! zkd{D0U~Yb&Z8jlc9G<*WNpymG*^Vi&9l}^*fcRT)0ibNq|c+@IEn-=?j7smRAlS~`^Hcy=LDPX{%t^n9k7V%%1$dRc@C?u?pvg^~gEY23n&P+!-t zv^RiHQ2MZSdU@8~x;h#_`zP-4(&Yd|1nmB>lX^|x7Rx3>M8|`1s8Tw+EEbHJbl6yh zWpM7#qtsGgR}CI;!~#~d%r)n%zP@5oR*KBSeSWRf)2l(8 zhI+wgEC|{4g~p{q%LvF<;vdVp&rM@EAE4Rg%l2!ybH*GANLb5~B`5Q*hh8L!*h&KV%RsXzjn!?``G*;gn z->n1SH&L3kKL~7r4`9pp-yD39Rn@i>^uF$v+>MN{?iv~}UG1(aw6PzTENRmYBYF`> z5={R{V{D#Eb}tT+PGm76H@Ude0t_7&~KyX&jMTDd(b9z$Ok%b z{0&9;=KF0UlB>RyfNG<6<*Ste`WI^{3HcBYI-D8a%pO}EUYj&3cYPC>X2&6!0+rV_ zTwu#q8Nik$R{0lr?9WWY{(i(|5T1lat6S>AQ04rYx{xpv;Nzw98Y2IG>t-(PA{K8; z>-x~02M}@7V?5yBp_p0jN!Zq)Zimm3-`^Et*vUaTU9~jTy5HadDeG`WA+-KCP`To( zWZz^dF6^90V~LrChmTfx7- zd3I#3#5~Fw@W@N6pe-=IOs(3>WGnec(RXZ`lq!0^L25#k!DCcuj}HMnNEwtuPCgDFuDZdEDD`d2f8vXATd ziY$!6z2)ikk!=(Ac>4BHWWjwiY)jq6(VKkUqQb-0C&r$rw{UPChJZ>A`RvftMP%C^xp~3VL6v;8M zPyX;zgIVhodF7c^3jz!0(J`|!K>wOd0Sl-Ke*O&~M$Ba|+RrcFe;O{+ww!lD6cO3w z^+X2^Mf)h*)wWZTBI2`{G+nV(D67?7-2^`rdQi~l-5u?&pabX|K&>=rzF<;bYec*< zz(+AjSSdhP({a0ci$IUqlF>{zQZN_SosDH8op zj;WX$p3US`m+*8i&Jm0^Fe}sbStGSW4HORY&2*e!2?{4~fy^W#WaUeTgYqn|%U_P3 zmCH^~a*^JtbzaXGj7sE__CZkQkD&8JvE<2@C259 zuJ4EjWp~6x3SUTCiNgq4an?Kc;O@%iH2Q!$kKkec{ARKwwQFq$FTwk&?rOi~v08VS*R!zqAQdH042 zO$j{CTAl@r)0lU+s|ss9?b~Z->JGr6rFI?W)c+J*zw*@0b9cgP%J3qj7?D(tma*0+ zrV=*AC)Omq`n*_yu3*E07+Lh9Gb^4%44FVdd%0V#v2J^3re}qxAhhZ7OF|i~grW~u z=MjUB&QGD(+#5rx+^uN;&I~p3$ZZ6|}>4kN#RezLHKw z9|n)R^X&P){Xdi%tv|`cksb^qsi(YDm)=o);SD<}SykB1pC9|($?2RfssOMr2Clu@ z@TM8(e@zk0`ps?+$G4+gk_n;IFRL1EWW^D~@PBl-Tv({f`6%AeB@>WIPysi(E4-^n zC7k$&-FkT4ZBR2A;paTNX>oaV`O$J{+eChMO?mDw3bRMJn-NW>tuNIVTdB>37(m@~ z9^@DHE1DsDm|h@PEo6Me&33p2bGu z3?p+{$An7>NbBra6sXT*4Yntp#a2Z#8X9x2phxZ8Q_za8H z9HsfjkB-5a)b)2Z@kp~Oq;)*|VMB9=f|hx9TAznl{}X(Y6iQM%cfIw0Ob;Xvvf>rS zlli7b1c%Mii3m_-bOmPCvT<(f&=MG881?0f{|@Iv{U)e;YRcc`lbJn=-;4n6frDqK z5=GX`{j8LUZBjD>P3`X+jG86Hx`YE6{ljBX1yL1sc+|r`t*gFm_56UqEn<+CkP(g% zS-;tLqI^++Aek`(ryVJ#l$PCrGnZ$bk3`&ZmP({oUg~~V@n=2G&mU1r>Xy8ah=C;Z z_{<0E1w53&^9mXlhM)hO0qOPAjLNJ_860HwB&R*@nnCoFug6v~6B~@S7&p)? zM&q$WbamnDGDzQK++ZtbWhn-O{hf9@B^_YFXh(N%2=4v^J;|wskTkS0@j-wahL+=3 zy$Yyt_`JNJpy{3jW)g!uLcwHb^rgD>qvi0FXET!Qv-@0qShY;C_-5KQnckekdvDm} zSN_WZhW8focG>#iy4)npz;~8$Yt}u}sq4uMe2;P@gB{jc*IMa*Q;vDl1Rak=bN+Ic zt+7p0G<}vN75rW%Z+SG}N7+3y{2Oq2x}#hst#8=}q2n)-R|iT(Zz4IR!D;>5B_M|` z%6DNy^16sV=Sp%6%Iz@ftNeTI8a-@Ts@>niA2k(`L4WF`VhXvqxv$&se!v7)cuDj% zHGpy=;GYY5c`@4O?!7A?;Y2%O$OPWcTEC!AV7&3!WU6;P!kg&_o`WYOTym#JcjPp_ zp!ks*%?w50Cc4X#m>oHynAf~JyS#F!WRE`VX=Ib4(@yy|xJZ-YLz>JBe>Q1JojIjB z@5`Wx7p*yyFMIpAI1)>*t}&5pXJmVs_&oxReHDCZ>+YoiSvHTJIX3I_b={!^Pi6r2fqKQ^02ffdee<+G*}Ln&FquT zYmYOZv-1bOH#|Y<%zEAv)zCZDyB>eMU$A{u7BSc(c;#sJ=FPjPe%4tBQq_g==#^&s zSKVtFjwjejYU~T#C(+nq7~RzzX42uo5|&cMlwp_(V&d)8tb_vUbEkcS-`Yn;{Kjtzjxeh>rU6Q2eNez<9RKaliqXhq>@|CZD+o z>12T=Bb?Kn=TvcXZ8QiO!y>L=lcPXmew_~5wWCv^lLj=Zk4oV9Lp<}sgsC{-(2Q^FOc-%j+!Mcj0?s4gCE08?$iwz{tWP$MKmDub}B+WxsHEvQ(t4 zH>X_agIdXV6PJSk2YnY!B)sD1C10q5&~SuGle6&)HQee~b3M|_>pwXKUzM$MXgV)< zdYdQr+)5Efp>x}ab$?EO&HK{--z;5V;DGN)h#c?i8bn5@ed)*)%x6C;XY1^QkMtO+ zyUimW-Pou(|8pQ$=a@x_hhlc~_^ek@{~EHImK0cf@@&?BP%zf@w@2VOl5wEy{+BWH zX$*w%VSPCeAx2Nisp1Qg_SYoma$iSW8dBNgBNV_jS?cFRMcEgI7yKved>q)eQ}0=l z2DzLIP^r;ogz(rUU)(KUtUd*M|>b$EA1T{9vkb$C0YmyGNf#n5e3 zY>uQ#r9a(>*Gs*(mO-;p=?~}ma57=|I&G+CNJy{j9-F88*ajqvcY{q$5Z_D`LJa!@ z;x`go%ZGoXS(9KNo>ubZ=ge+h2c?j^(QK1Y0@F^kF(TjqAFUCVJZ^wIHK}%T$&&zp z5?}~C%gfi&l2Bv~vL|HVy_jD#b+o5VvqA3bnTg!47et}9V7Qd0NLyytg}g}@lJ;-& z^p6tuk=t?8tJI}XAJYpNy5XqCT{obI63+`sTWY8S$)PerJyxxp16H7UrzM$Rz1bWd z<>sb%V)wWVi>LW19)sAG&dKQ~cENMUvzlpn*u!=WdaY+*r$Qa0$I(Xp{pEk+p4UvB zXEp2{Vy(IC?;&piCO-9#)`c&CVA)^ zOLR;Z`hRjW@P9@TbpJH-LB}#=ZmkcH6<2rr)OjXaL;Zq1&Ay1NCzNW|Bg-!cKZ8oipQh0v)kZje0UH^x!x zyiI0^)Qt6FDuwFEJx(@Uh)R1jd1)aK9Shj>S`R3L_BouK6G`O|wDeljwqLB;$?$MMW%aB61p}TRPg9f1lNi$b_^`@qQ4UcN{Ut4nqbL2p z;kxq%1`4)^(B4UxU8XzM=YF=6jO~LF*B9fY*KBO9NYN`z5I^r~b>2{JiI<-SQm5Ei zt$?Fmh9LN-kq$*c$x1e*SNTFVpM5p1=6sY}3+RIC;!fX0mWoE1FR!xyBbOY%g20q$ z`1dxP{N$kF+z9bXUDoSXD$V#P(qK#x^>uH}ZL?lIrq2ldpZo3y+Dv+eZ`Hr>?(XU( zXL{lar%vTY%5d0q?9b%>qUW>f{ZG0CQhlZ^!oBMkVAbrXm-P)XfoOlg_jaFo)hRJ+ z6U(tp6gjQD;{T}oMJ*f&BJc4YSdrh9?{vyoX3d4lP_H;?NFU74VaU1CI|W(%R(Nh{ zsalpNvENw&tghU8!D4XEmA&3su4LkxCGWP;#Rq=fc&e4FZ->m4ed7&%6BMOeb~V?4e_#Ba9kO34dk& z>n$-r_pvzA7n?70Xh0j)-*Wt7UEipYiw4QU!m8Pe1H8inxPzCT;e5nLjf4g(0<5<5I*=gAd!-$H$iV+FQf`YDoq$Vw#!%%BxFdmdT7sVAphrvoU>pXj82hTbBU__v_el zpl%RBeUykseL*12YkRq>zk(^%EJXpYaLEqjR-96~=1m-m?bIC$(C`6w=rrL2RI7AO zXSPbUIXHh^+Y2$nAt3B47zFl@EZ=(zFBXhq44JSATu=jY7z&~j%8LFRmr*yRL9JNU zB``cW!^u+YD2W(Bd=%IG1q0pJJ;N*&M_7J?IC8bt?BU<=SB@l_?zq3&L;n-wv2wQU zwkuqI-D{X5e>(LZR=4NmdyW5Za)7`*Fa(&uf6MN@I3%G~wFbFISb=T?m;-r*Ne&Z_ z=|eL*oiw8g;CIoIYL29J@>HJnZ1BM5+4h41MeXP1+j*Zo7SW|^>*zFv0UQRZod9y; z$}Qtj9L?8+B8DBmN`G3ioszIaC-YF@@H8Q|cyDpdyE{RAvAM@?wtpr)xHiK4Z%~_J z{-36ij{t;K#5Eq9Umd`?`NOm!|GWy&wWU;&#kP zuId&z@s3ZH8g4&GfvzgBM^RbrRCt?5T8_nYL?16Kv60ix4r=nJ+IuV>V7Q%pJm!4q zXg}W2*O0qyP=)3=Xe;$`x0!XyPvJ`=X-G1I_=2kV+CJ0&L&{8XVwIIo-Pc6?jZxTH z4%?+QwA=pW1{3lKxc@8<8OR=%4Z4FJFT&mb6t$z`H0)XXMiR(sk;3{}11_1v_bQ>} zefLuU&z)SO{sbNUx9bp-~2O@lgo?zGPXO|q`%^7%&wiA6MT9G#g` zKPehh`Jox!Z*Ron_HG!Yqvb;L7!TPa@7|6s`kVWn_$!e>t)QOc#9*w@6Gumt|J{q7 zdB@)vrM8UFdfa$Vr}5mqQpf@)9)l)OUnLIg*FPR5dP;smZ0nCSM?g4lUfp5YSzG`% z3i#_(A>%WUXAil@)Eas432b{9_{@tuC(!W!9tqx>5$cf5un4tT1# zSVIHO&@1-lhB50kS_0RHnH=SQ46IH6>2cOwMjC?Vnmm|qE_UPB1v%OD6&c3H9n_SV zuUEoJ-xGPAeVAQAYN>T4f5m7(uKoFi8Tdc7E&HXb%iu%kXxIk!_0HdO%u3!qy3wdF zkW^bG!8t|oeb2*IXkrr$$?LAo@IL09#ke^ub*-x#KR}LSj6J<5XcXDGNo#%ax7){o zQ8Wld+lNG=;K) zj5u-{H?)Z9&M%)CU{kp<(g-5}7KwsyBR+oSJTfLwKA!h%{RbWp2>zMS&6%}h<|;+T z#LL*g|ExV7sGU|{k6@hK2aH~&;B;-oJDX2DaRVX(%yz)Le9^Ph`*lR|fA{;N0VjsT zbs63BHirue%z5mx8~o$RR9izM)x(^RF^CJx;Ns(nIbn%V+eIFD{n%rL-T!J6oJ`f# zS~p#6#tx*qH%3u1IGJG)X-;8@6R@qtq($J1)G1vY-p3E_Ip&gxcEjVMD8;zjK zdfL*)cCQDZ>f&NP-!1WrTuY^xBwmN9@7z0gAP{bQ2<_HC7AsYOnFXC&lhKs2BSFf* zC@?YgQf1UOyJU-xt%58pM}0QSawDaw{E;D>KnfrD7sisb#Y-R8cD`A+K)2$05c*iR zVC*n(Z?yBkhy2{QwqqO@ru7*xf}=Iw!DwoqXtu_jLO@ z_Z3DV;w~}Mr{?GduW~3`Pjys?e9DtNX@4B4q_7&_%H_Y}#X+R1CjVM28`D?t@x@cx^mfR~tkC`h(9qz(I^1 z|3{KUX^;oWV>?C2w!PZ*d~BXeao2NWdJeZK2&)d>bp8d@B?!{D`3Xzk>jRe)0CiBF zY~#`8JPn07+>H~VA)kbGiHa+5xIW|(tu&F(Y{Q~Ldbd~xlX1frnwt!QR>R{+dY?pMP3lN#c{ws<8_-j#-jPq7 zCi4v2ul@K$6}MI*i~kAFCfDj{dM(%b-L$0X?zsJ1)3DUSJL-|vR7W4`frYb~Ya}}t z>cigV!@GS30Lfo9E%6+105U7SL{8kU7VVp`7uimPpxlrZPK?i=muab1BceG8(fCaY>f`QWT3H((x?nw(6)cQtGHV%d7U zoXLOm#^K&9OmsXhYS4|QFC~_6Z6Se_z2$wg@+4kuM4bohf-f>{WEFp0vv^4iImJBb zn7vR)6q=~A6@1;oNaQm?E7OD!souARq&SN9biD4w0WFd%@ow9gYxD{%-~Vtnsun4g*zV6 ziAlL0TF7qfwrm=hQ(!iUi+V!+s{lwefG%^$xp)~j8oTGp0ed*1d|1bA`DkcJ8BUaK zt~XpqDui{m9vsLe;E653_R+uRMBVoz^?OwDozyWiaZr4QE}P6sPQ;YF(C*xaTR0^2 z%4(`sLqpL5d@{S=&F7_R|G#Ye_GtikeF8)Q!Qjo|k|*2W(X&s(-da<@2%a(%l~Xmq z;@3N#q;NAmrOt3qT$2Do&E<$)t;xNL--iz)R{GXXx2mf3Op*4qiO?}Eua!GNhDt=Q z4DFS_R}2{Ggy0d4YKEt82az~cqeOlE4Ae#t2|u~u-~+bO2}_80ZY{dp!teGIf0ToFVyS}wcT1)Rjv{$@!O4smELa06I=*nd0EBpS0dzB z54+*AB%BswSvMDzlVG7*d@iu0Ea>Z!aBxF1mf1Ll9irZ~6{y1Dp|+%!aZhXGU9fZLv znqO$7mChO+r=P$bl6z)HJ-goiPs`;0u2HiZ3$P)pM>)dETA9wm{-X$j`DXS48O_I6QB&HWML> zf0>DabrlZcb{VeN8FWuUKF&`?_LsUrj<3vehN1DipSIxl?{7M#5?{XHaM&(Kq}bG( z_Y3+TS-cdg63dZ3xW*@AZpcUSgteZUVIgthvQ7o*0Txxh3@k+F%io@4!)D5`%%>aq- zV=k$Gj|+Tt<8a~}YZ2U>D$jR+VD8UTpdf2P1TS1OiO1UDdmQL0*#B10lEQ*|q|YsJ z@cM2ucSE$Se_1%i*9-Exu^oo3-i$7f+a^`lXW;_C!?a7C(I)#+OL_F%eo>=?toc3S z2e7}bwhs4$2m<~$AmUUWB1BusI!=4CCJoL$d1LH@>_lujR9DE9Jb1H2NUbnoz|A-; ziQ4tDiHp$J#GsKI{MmxHnGD`hexV`y=O*u0AV?r~x9c#AjT|cAS%$GeSc^JUp=nUF zfT|SeGRm}fV9Y|s);{8(J926k9ZigDbOeOZ=V7VIFHzRA1KQ}puq4aL+k>|zEJ=SK z{V68X@ZUR?G%~~g4ILJwkdAZQPE83LUM7Fo|7nPam!9izagJZ>VqD&-z)b?UQ^!k2 zxH^*URLi{?g0iiu4KvbaML|w_63LJtD=*zOu^7 z`LWTT0$vo0=}286P03z+#*^c`d+GXf12ZmYJ$|9|x4o#Vyq?UxRZKft!DDY%Kig&q ze-&Ybkde2eMK8T{0hZdit~05GhcjrOjr8!3_N++xP7b^tr$VB@X#Cij9|Gb<+LtjU zr7=I?h#FhrRn)q@EBx!E{aUetuL2yTk{H`~;e3QDQ#wj}sazt}#YF#Fd^#tsSXz+u zJiTjiPQ0CgIC_-Z2|Q1}y?UeoDiXb@_YAto4#b1pGr*gq^Hd1~|5_6Q@oTo=vmbet zpp`VQsc^M{bH&D?*!>3mI038o2 z=ZMp(-*zqD*ua}ZJNjx=T?M80TQ3slr!dqQuHWGsh5s4do}M&KtF8y? z8`Z{UON-(=2_qTb%jWd5PlB7ImaTh~n3Z|j@j)?6)c}3{JO?2AbPHar0AFc$P8#jYAlP+Z5UKfouxw+lzsXHP^@j{K`lWgj0pJK3fLivX+ESg*mjBb2G!(oy(L2i}l4> z-$#h``~$C^z)rVzt}r0c_0_%;2S(`-NBykMGH|;pbnFVe2P01~IL6{+ES-)f2;8LL z&AmYX3;_LI+=QwpXFon%1KLBe{s|>xklW#{>M&=8dh2j>xPl5e$3JTwhSQ?SLfr1s zCcycCgZ%1*Sseow-CQC;p)lSTBL4WUs}?Wv2zpj`G(tqIcsE6Ftm)S-^T+dgJ&(V7b>E-+zV7Q>=bZO-&ikB;&qevlbQ`{& zQe@E=(-g)@gA1HF@zQgFPz^*<=Dil;;x%NbJ(F0vhBhEpmzH@{iRl{0BMs@nMJirGH=P}UDLy6GYpsY|2LqB=n|joiZTKbx(p;Q3j-U-@!P zRu1+U^XW*USy=8TaW)TEu?zxqG_0t8RV<`ZXPsibIDc7^wL`L_=}+Q4S8^=tL5)BaZFs9w|`jxwN?CC zLGwj4s;akbn>92p?Q~V0&$)8KM9$3izQUy$H9HTV6MOX+Y4#$u1||w$tyM{|@LkPx zA0(=H+LoRDlwGWElg=WGSnrLrqp{`Z2!+O=H)ivnpL38g)4%vEmVGdoHwS}3Hx4rE zlS({^QAwg653|nZjy%6JS#lu}3GGA3a`tAG>axl;qEJ=erB^;{-`fBF5=*CBPK9^( zXZ5HeQGq3=HL9fR@2)QNo0^Skj_QxecnHGEGkI9#^ZMc}&-AkN+-V4XqcpVzuX$Ab zy>H)J1am*Td1kH?J1+2GK|Y-}}CQwDcLB zY^u(MY$j$W0pH6(J1RH2rHu<&MJ8+9vd&%TU`O7DO$A8yr0>;P3mr^Kx1=IvMS%IK z({bv@WJY>b3}r}I9ls<=OJ-aG@!Dx=7&SEH zbkFmyhvP3zwf?T)lw{JX;>z|Jev~dKkVHl=OSVKqFMHJk7E5sSbvoJQG|#e5exmKB z8yf~DpYV}b^39faEB7ea5_d^|<;F{^C(V@o;wC!c6Jiz@SI{4&D{%{oYmySi#pk#$ zTpCGKsP|@ST`m5DZZiE`Q({nc8fszyw9}zAR?(E_*ZS;QptvS~24fUt9nQ!8@GQQR z1tC(cmJ6fN?OQD#@7^N!UlW?D`HHks&CfC6Czs4+l>}_O#VpEvM37aew1zwcS8LBh zF~?pkSFLB`(ePNg-1Gv#!J$QcAANNsZi!yr#~Zj^>)ER_`?k>AH{ixeTD7QNq^qM_F0CI@n{QJkB|d#y{oxNoK2Ql zjlI>Sq<(ri$CCzvA|bmkh}2_Nj>W@-PTvp;t+``3)hT!Y#EOK}0+PuINz76GW^GWO z&Pj%w;>Ml$vl*k<=^2L(%;PXs&aCdzi#Y&=q8toBFFd0lH~!6RIq43&L6POBZ>sd}GRCnu~ZLOf#r2m6_ zY~d?>z3U@0`0rZzwOXcExPn@y7p3Qtn9dTcjg2HLKa|8p%1myBW)9y6`rtIGS+XZT z<0-Zo=ibSGl829wu%7bM?A5vdOe@~1$-mn_uhn9_>a+m_L{9zGa~jU3GASbE7mo*7o;Sdbp!QGZl)BP=wqXB` zK;L4yg;oUh5(Dcxv;tf*MZfsH?oVBnmc&vtmbLPr5g6_vzx9bI!J0=h;KoyT;%&~{ z^Qfw4B_zIQaCsonbH88g8a~|>Ra;eEF+*Ei*~hFsbiH)4>RRT|yv3z8AcGdeSmhW= zN-t}(`fS6f)o?EBA*CwS{vH5Y5&@p$VZqn-rgEYLrI)sgY|*K)ZG=E_gH>ALiE>6z zF4O%)eyDcUWXW(`ZOb})hdp+?sq`1`qwjbKtc&T&v$M5EH7LSJor>p(<9vK?2Na%B zUCXRAD9y18vF9eRQ*{-aEsA-i^;PER_imUd9jzvB)|}o(a99}{u$s)I0>z%^7Db-5 z>)8IGPc|`T4*j|KzHWE1Zlut%vct~)KxWq(b&Tep)X9*16E&UxkojB zBri)G)Tbx-s!k+b`;J;=t|kV6R4k{fo)+qe@tOFh5WIz!V$PRnTaPp>F~vWG>i(gg z-{8sP?224y1alr6b)MlnMGGkQ439v!!+;3J(K&?Y&FJa~1#xLITcZyho$qQ-Gpgx& zy^>IM1vy08T!;MW^Ep|bbC0htbpR!&&K z@l?~cZ*R!6$`(TnIn;Hlu0%TeJYFI`Ue$0fU4Cym)plWe!T4l=SD3O&5?5+h#kC7E zNZz)Ws0!7GhEhCA)%)+_Zb`|k;~mBRdGT{yd(STOtJF5wuQv5?F27iu$pJ&jlz zb3qA;B>JxI@2?JbelgRnUv9oaNMY#;iwEX60$yP zKV91No;W3yiex;h_fef!MDDlabBYU6)|BZfUr`nUBQfcFmD!(AfTtJA`%_dR6YUn#%TCCnpE3M)S)m4$foQFzQ*yRiVb6~;jS zM|9|}xcS|3R`LgT1=tlIKM?4c38b=`do_XY@zm8W-_E?dw(B-DE241A4mYhx!>6z~ zOG1+UL;m9P5{*Yt-8NZp2^yxTS&TKmn1fz66`+Obb1Gb?BbS_(H3jc*hzc`gKp|$3 z9n+fyH8tN}x_W}b(87~0QfBW7gl-@4gwvN~IrL9kggp9?0S|gkCndrNEcEE4`i#$HS;M##p)w zd)JKyEDajA|FYc~+k1~s#AN;h;<>yB)k|7qaUPE#jz`Hj6~VJ?X?^DzB7{FSe~w!Kljw3NV&H zqtyY@{ckS6a+_-27?S>a^h6b#Tf;PAnPblI%=>!f*Jm`twwO=#CgXml(GU{i$w5-M zggkuQD|-NvBzQ|0;0C1QhZ8s4I(r%d4&`AE!-vLEi_O9Cunq z0&dx!xBIoR=;;{G?Iu)oVQEk>+NU%2p{0V2h0zUg@{J)E%FMf&>?mY$*wt+tvvYal zq@M1Qt|wLYRUVriAAGtwzq$~@TAdR!ULnJ1ERfCNqF;z^P@w$a`>vN7Q8wwHq!Dp}5Y7;1| zU?pm+(HZl<&(0(m*DJqL=#kS3$k=!GQy}fd8#qXTa?t8VU6$L2PZ_Itp)5Cl`PhIm z^yP?lo|Q42e=?eul;LSQf0-1<`8D@j+kQhPAyp`8GlgqLqNxTj>+wTrC_apr0= zl%X#pPx9VF+6eZz`2nW>hvEf(aUO5&85BgkdWo=CPH2!>xt-h#8uZ~ym>4y+5pai{ z**4@GO+i((u%Ie0CGbTuNFOE-;v$U`mgEMs!q+XIwDn`pzSTMUTAVL+b+YkVEb?c| zfGYTBc9EcW*;}qZv=_yHn(mMN-D6c3zE*el;l7o0M-vB>lLP18siqDm5xaAaN=-g{ zoqf%Zj&zxceM{(LesGE5Cg+ot+{e{X{VMoaZ@h0=oL$C+GAlEzb-H5KKQ-Q}9xYOe z-hU}iNToP&A?appr)j6X2)}76DKf0&2rgs(+aTu+8<>1&iFEPK*n(`r?NVT(r#4;R_)tk}X z`b#~w{ppaY**7nVEN2{~%wj8Xk! zT55Ie6Rgn8g{lN)T|-}Be4GfG0e4}0Mk;8G_^~)&SNO@Y=3nolU2zfdA|_3Ud~U|Vl%Pur%Rg+w}G(hMb59h@W4>P4<*W3tc%Ln`c?{^?_#Ijj*+ z#+?Q)dULEe$uV6Ax(u|)gida+i@Chd=Q6^Hdec z>Wv;@QhyrF4fB+cdYs=4%62&$n-}bc_e{F8m~n*?7x>qGDe(G? zNUNQ>9!&c9?OFGo^Kzs21n5%;2}z}@YM6M)hdKEq@5RU8GOskp4h#%Lj97)R5>gzG zG%QxJvSO2P8I2@7?u%%iyv_$>=TyhO{d&GD;T%Ry%8bb$d*dn2&@om7Ni{gva#M{q z$5{n<@D4(8>PF6aZB#TCOm!ET1AZ>Qrnd?VtN9$v{O8XfCl{C3ng(&R#h8{hv8aT_beP1NAB>10aiEXZMM zyH7=Hywk1hET-!+fE9`g5CGIzY`fVgGMxvlTb#}n{THI9+HosEAPFds_pU= zUZm=d2Fu!yGjp=>uO6hxd(BfG4_Fj-9(&$1+A49ZfEUwp8lASIPbD@dS0_YyrG|)F zWVh0UrR7MqE)t=2))i6hn)c21(jQH4HJJj{(e~tJ$>Fv3(sRedBZ40OczKmxIwB&X@7go^?N#dqv3@x@xfd&qyQ|Y|`wVZ> z1u}cpalY8*&D8$avqhM+c?FylL4AH~CRkh^?6_g?VUSp}At3k$63jm4q=O?P^o@-d zO7aK@2v)yDP)tq6umP&thqiO>Zv|?!f7;!Qjg38Td-@r*^v3iVPgj2EDw1;J>2ZO? zAOrypx_VPtdsoD>&d$Wv#V8NxN`RHJT{rT{^gUNTGMyiDyfsVOlai6~CiDbD_%ZOT z*vn`z9i#T*ibsFDC6p_wd;1ZpP@T!Zqs?|!>hIz{Q50pfRUoDE8h^c!6~kM1Od68q zuloztJ$}p^>@8UBSB-R_x`&MfN)DB`1v{juIu(XPp#j$2gmj;Z?{VbN$&MY^FiOsoj{UQC@y!Q(a=g55_bn4#FetSzLf9YK4BMP^mt=r|(-p6@d$diN6T1rBt)i znb!8}2S|IJq=6J##hBcINm5B&d;jQ++f`0f1Fv{6uO|%`dir|wLmrl8+?$|7T|n?V zeI9&1M*Z)HcFxigON(F%Qsnulp@+9r&{LevixvnHF_AE!{v8Wkrc|ejsuG3A#uZ2y!Lfhb97LXUuemr*MyAsC zowUO$qY6b%MZFyV%-F#e+7N@SV5)xMbislGNUoGD-y_DsqvSFkEi&ZrXq$O__#)+B z+kQ6vk56=4+zYLme61Q#=Z%Ce>k4cITP76-lmH&!cj%1|t4TJSitQTuYhO7v9vKm9 zafnJ1Kxag#8%>5f3M5qbdRyghag@}-Sj`Fk<#t{HM}C>y%3j?-582P7U;cg-&Q2)bBQ>O`i7^6x5vS;EX}nNtT+uHX>muN)vCq zpzmH!b@Epalx8ELqi=MfYbK}c8#kDIcV1d}QTyQ(=<+~T5qhwfe;v9XYhv%Lk)NKlJ{VFbukd9x)}hg9uG2m5-GcwT)4zC$&U0UiXK^-bSL z&u^dWwxmsFEY8om-;7&%;L|wo(6Y?NiyPZpgRnqLR3^K)p=uWio!zknDh216Hs)&j zc8&nugtq=$TwN*#v_<4ofum!)MdUb09`JWr7UP$f74K??4^E~KLI9;u7(#J(-Rvp2 z3mLzvmZDm)8`LFSc_kz8MY`b+zIvEqoN@A zQIubVGy`M)odR#0j50SY^!sFKi%EY)UVAJQReLr*@LQMyg8i1)K9r-NH28BmRD73MC-eg5oosmlUg6l7sn4tm& zH_A|~54wniKG?Ag{H-dT6%X&hTC{`1&QSqw$`(!GzOmPJ|=?yy^ zB%IxLmr!;7+k0pYqC@|8$_otAOZ60$Fhs~X1^ZSNr2-N(hybw_PP-!@k>3P? ze=Io?)T-OC{VIn~*9kK$ZvF}*3kju1@*0ZR>S}Jgw~O<%{ap@Dae;wDFQy)_022>E z(N$8gg#fgY6g>Q}0YIQJi6i#P4jHCQ0Q>Ww=BYv#Bhr7D)9zBoU0JwLn*1}^GWx;J z@oz^8pt2D43se@7&&frkZ?qQ#!bG4k!`3%GARJ_pCe1D<->D%+) zz9~98^P{k+t)pXX;Q$82vb=$EncI@HUmA?{Y|lHYDCwiobrCi4Cb& zoO&dC=Tlr(bMv{q-zy>NCMI{kM$BONO)3y^aq)jY>{98#w_Q5MINU*aBU@4DN=zh` zpCAB)YtNWIe*8GxQCeEM4Isw-=cgNE3tT92i9K;!o{tzhl$f!tfCtES!KGH4$yoH! z>otX&edqII^bHKcc9-iYKJQNae6Sf@dzqm2@{Xlil#=|c9w)Ga{~5~}r=;%MqU)*s zV};7#{Syd8QIcimiaBDXAXFD3VcH3Hmvv1MeeCvR4EFZzC;{{5+P{Jdc-Nx^(9g!4 zQY&OEmk*)T@s>kEfuuLSu_^-I^eXNKCDcge1Z|nh{v7w53+fpzHOHo=in{K$L=dj^ zxCPUUJ$F$yGaF5lFaK{WIh__dL;0VLOCpO-ToyA+-*on!WV<{)3gmoGT=UnFcQg2V zT9hn{XNyL3>MWKTcb8w4)=c<^QBXvGbS41u0@u(rLqlpLuc`|3Mm32&mJ%<9R6^(G zNB(B}souLRG5-Ne*g|QMxq7kFm;LO><$b4VVG6)$xjQ9G(v@mrI_Cna0HGd`?i$?j zH~;Le6xP)-|NI=%v#T$q-Fq)lzUack^& z?k5Lh^Os(y;^N}mpWyQWmos9Z=M?Z?wR|98TUfZ(Mx*y(p@Rlhxmr?iW5<{W`bI<}qwl$b)7^Hyj#F0W zdAwfC$7<;Hv^&NheJ&>===6?0Eq?+>hS3+cdWnen~C$Wvly^~4Z&;H$!qFK~@Ome0fTsG1- zIv69Dd)uS^nbyDO+rj}XV8l)JAM-z)yA5I5;Jiyyk1iiN2~HkLx^Kyl2Q5ubUMW{p z^(cv?s{LmNbvaK7kZO%QFkxHy^^{g1u>th*-oG!#!WKP6{45h=1!sH~~cQ327*x?H+f69UR49nW)ZRV@s1dJ@@*rOgix)s+Dm6 z=@}$c(XAaEP~HG46Y0Yl$Kb%XEi_@HfS-k&3t6E=HH14AAVJn4+To9a^7pNV1ot=>hj?LYwtn}>?ch! zZ+Eum8<`-kqQQXmJX}yS;c$FVrM*to=GwBJR{!WAkBv%i#NnjG->a57o*`|3Ak`Yp zK*h~#x^aH$(fvcb-l+Cu*fo2rZ|{l6ckJN#ADnFo7E#86(a{%L`SzrRM6N_oG40%2gQ9l&cW zb9^j$^UD%ZP?g2f_Kf!uN4cv0EjCxdlcGoBjd@JjRwgdRJ)f0_YeI($@woN!Ej%!s zc*tGtBY=xz;n=sf_;QG$w%X=06ent@S+u2fkLyYhSfq!q{mKvnb zEXzb8rqqfPZk(Q^26GzOp~TnEd59`itXGLGOb1zA%FBOX-Wuy1bNJ0M_oOB1z#b$2 zAm;L*Cd*LK;a+a0CA~~b(N>cPY0i_J+$hC|PHZQEo_8qhb!!0q%o2`VV&n8k-&pCZ zt)C<|5+u~>K5Hz7GtlXe*{z;xVYt+@tphdWRQ+No;GAk_@!frlAUc~3+dJ`?*n;c2 z0!t~Jb3KxuRa{(3x#Z(e%n&OW4qaA^$5U{l-WzhJ(LMe01M7dY@THWG5XQ{hywZE& z(sQ`4NK2%*H3EqCi#;KsNUoatb z=Pmo4x6Uj>KTgrE3x`mDWQJsaB6!!fKrUzipko0YyLSS-r0Q#!5ESOOapS>ONo%U} zL2sFA@A9X*e0|;uh@-^|Vw98uIZ`#OX0nb^ei85{nz8v$$cGtf`dqI`1lw^F?2_?H zeI!cN2sIF5mufNoUh$yY^?bf29J@}ozT#pjogr^%eaBh)MHyXS-5Sgd|GRXdk+cIEnW*H3YJ^1tu^SfGK- z2fI{uMBzk=?m4DV(162D1Kek+{{kwB(vorzoWVu99jHP#vQn*LwXgU#GJpHyBh6c{8VtNmlGgDu%_4;^>}g?#fuc@rF0 zjRrI!Mk=B!qH=a}ez_nRxY|NEydMmhV(E9~!PS}k#YSCSUtcNn!GpEZ zY4f4%0GOz_M*UsbRJF$rm1E}8p<%;<#{XiuR>$^{_Z_v zJ4+8$g|nvhua$}&CqC5r|J$2*i^!WVE`F%Q6{=r7BF01pJBggYb&dN)MH;MhwZ-q` zcS#{g7xmG(y}22&wY6P^9^9Pr@|NHGI(r)a>@1*~U>UF^&Tm<7^4+~NQCvb|iXvni z_vOo%eV;y&-PVj!XA7d$O6SMQh+#wtB5*?XrrA7ZI;qel0D;6X!sE^DiOfahiHot2 zvrWm(O-6r!-n9lh3ueg)U#{h*{a{j3((|lLwZ-T~*aWay25nxf2Xw!>x%s-LaZ_M^ z)4y1mOLBO0H13X7u<7?RowxuqCztD1+68LH!$&Lw$k;;mgsG{i^-WDrc{|NSqv&&S z^cjg;<1g+B-)XYF@5W876v3;wY3b>jtV|!N;bawfijt&{1)Dys%`I+w1XT9(?J` zSMKcUs+3QE`}MB0CpYsGza&I3!5b6?UkYn3*S+mJ-cfYwC;BHZ(t<0|_-NJzi~5ddWPOtBNP0B@}(Q@7`KRzG3lw@|f|y{okMW zC!t7DRYe7liHQmQzsn&Z3dq5U41DGF;TfWiV2IgnO!)EODK^8_^pr>BjO#@3u1o!o z_gcQ7;@Fgw(D&~RS>yv5G*9Y9d}16M539@A=Q)eM{=xO|>gb?=*A zNGrP1A|vsTa}EyNTZ$S~tx@b1`WPIh+0{JFEDt!+UX$F0nxM~{Y? zm8uyT4c=W678aKBnooh#yz9Py|E>gCCtEwa4(H_f_*Yv?Dm?4w5CrF!hKi!m*cxn| zHE9_?GQ*NaPxT-sAt_B(HBj;J5N=|yH--)T=l%~LVikUW(}O6w_okch1U8ejH^f@HVc-)L#;vHzoS%MS~~;lKQXGMqgjVEIPVE-UJpjuvq0tqxw$ z;}8;}@bmNQ?Cw^91A9=%TKtMraubwy5*ApPT45NNL;x2FfJO|yS9Fqn^O8YxMj#PN zFrx&7gdU%~MWP3`RN+6wdY-}&tTK5H8j{?;2)g>aG`cdu#LqEPfVkUyJ}C6dfgsLQSBtCv-s zBFDN+Tvd(M<|nOfel0LFljNbb=Mh5&ukRd;-fQi3(Q>vQjQfHnm z!=XcDs`4fjZd^z3?We^o;lwS;`{16+i62f02&lx%1gOVLKkK&>F4Q(@eRG2}T`jVz z#fU7@f$uFUWcg_EwCwD24|c;3Jf9D??2Qi!I)JW!aa?94j~O}Y6jQNB`DYD+2C;c+ zpi)gyF{VAo!eU{fmo%DL>DvCdbRWF4uL=wyQBCHYGyA)Lyr(NY1^bVzxp@5Z>tj^@ zkZfEPZr=JP7VkEv)n<9VaJE@zq=^t3!lyT)azXam6D3W};Vto5-@nH_(34EK`b*pd z^N5QN7MT4hEKm$HmcV2jd$f+Onxs77qZ4D%14dNI%HCaz(j6EO;uce+-2 zx>gcus_ybhnB6j_tfzxFZ3Hp%l6qzw_R60kyP`Aqx$us=BaJB)RaLy_#)j+SWo2b) zm8Fm+y}qY$Z~SCmd1de2;V={Y3-&u*B(=&AUX}H=RSvX)>!jCP%eJ`1>7i{|*OB8H zhTn&j%SQIZG#%fR+VAMTOWGk{@5UsQCx#EYu54oz;U5_A^K9CLI(7K?zk@8lT&~Zi3vT?< zMV8hn&5Y^8*;sbZ;26<^XZ{Y9{);>PHUED3C>^~6f`t>^ZiKRs3m)p_*(kNms=R)G zi;#Rp-%Rqock|qHuc$nCBF?aAEsYzWmX%M#jb=q?JUuh(>}2 z|170Np%6P;58tln{S0@5D)Cyj?F$PFJ1mCHJ}hX?p46ZjBk1_a7O-IojZ$%a^Om=ZHM*ZkaFJs_*j#Fo+N@6J)qZk*`; zKYTXMGK1hKt&To_r`@@sJ=-ik0Ta5E#e*Dhiv1N|5hO|SVNoNsr_)0Ll!Gy{DAd{> zWhMjU-#rbF{%ykj?>ObPEgyxtKlw4JiKCkB;`owhBZT4}AB6hED1T=YM{(^-s{hYf zQD-?&C{^^fx1#E!1zXdO>@eg&as4Fy3w7EHFw6CR16kJZqYy-7{`H@JVO4(%VC4g( zP1n9lKLPq>`S*bIzpu}q*xqmG_(^_kmFhw9V$xTHyx?KmzBo-O7O{sB=WLK5F}KL9Td{Gwl-nssaN~< z@28VfxPHA+sK-@h6#Lg?r=Xj-L3Un$wcn=f5P| zck52Ln9E-v&b_reE{0+DQ5VYj7?O8eOSFJBqZI#OuZk5~NTli7e+&ooTMMW$i z-;tVL>dR!&UEquq42|Wgii(tT$6+-tUe*$7bxr9q&-n*sTFgjIjgNCt4-3edIz8(g zI@nn>vGn>Ni$y{oHGJOb8C+Vs4OmmbE-iSz$;ru7{X)_Pm@WMRX0r~5&zZ&mh+_*2 zDhEO;2j;MghBg-}+`k2xbQyAE#^jR-$`V6{$;tSzc1M+HM>C%hI%3UKtYdc{Rh#Ve zOns++P9{*j>E$mjW9>N&fM{X?T~@gcZn6}r&%LE4n^|uHk z+4{1Q`oR=fuXdL4GxvqwZRmOzYT-%W(m1Ow5*k+)z(zcCtIR2+hh$F(;nf**UMXcV zFa;(y(vTETi~(woGCWmmkiP80onG6ZAdvRTMX-%91SF4f4k@_B#%Xl$GEuf6e|T zS&!3{{+?~2CXzm=s56LIuf6l%^qK}GTlUv?<_cK5N0Fn;!f)Ac6Y9aQkM{qy6ZFKf z0zG{?%?$MPQZVXLQ|T}Fy5;t#;hy!iTz#kBEmG3b6*n_X>tCVGn_p{XR6;bMfk9=q zSptDOEaLDQqGSJ)A93>;;=QcO-M{kMF*b1z2eg&eJ0zP zFHQ|e@fF|gl|)Vr9}i~Nv#k;|?|hfR$gAD|kLWe+BH5eQ#g$S;vNi31Cs;rl!UrA3 z+9fM7%0T0M^o}5LQ&CAwsy8w*q4a+Vl)ZkP=bYG?GfF+K({e8;ETT?|fUncem29F; zwUkwDo*U5gyYgSy{;{MIFISmml`Jx!rip|(Se;FKs%t-t$Ptzb%=TG1|K zFfyn)jh8TpS8J-i|C4N_hgR8xv)1`sY!)0TejY7=cfj-5(dQ%5YV9Jb2E#NOg^PF$ zf0`1JtDf)b#tgtV%7_%QAFr(3wY*U{An@Zx?5uZu1Cr9Ne;Y5PjN#YbvZ5> z)8cuihEFN#&3N>HNVDun7`xcixK6@YVbEf1$}2otxlxcC6Iq=7s3>S|sbH)oI{3nR zYJ}vxCeNav_FPMns=Yz{z-qpF{qkisH~#0c<2pnk>5zBhKlys_-78A*XsMWXS8S(b zon6=CPO%w+3W>RduW^%zD=Kfc?Fxj3G`h|OZ=wQ1rU$IF`h)VYEJ>K3&e z{sc~ip3QAp3kwH{A=4H}Y0;P+X|f+}&5&#eU=>O1@9+P)1uZW>XSNvkqwDRVf5YvbZ?pe5+;)Yw zXr>;+&&kUxw|DPeq=6DJ*)8OVq|7%VxD6cCm6P_+#(#EhNR$(?){Z!T2dT{-p~xX9 z+~&JntjJ#;3I97BOaBsI%P!Q~Sr6UveM3FD=0+R%9(?Gpk?^lM*WcpV2Fx#*_rIe& z)I%EM)Pw&Cq5KuT$8R0D-5I#;l2PE4&Eui@iMzhv`)&nAMIHAp;Xp6lt{C}tU@LUu z|BK4ST(B9n=XU2PGwMOWFE!|IAM9VKj{iolY78P5JWW``X#2yR5vYZ))=BlO z0I0pay|UM?O_7s|~H6a92EYA!IZ054a(jf!tbN zUY^gBCl{=-SFb*@v9ioB>oCX+e%vMk%qN=Zqj#Ke3naNe?Ii;oXi(T)8lu1~BSVP+-( zp&Op-uxpbvt}g#x+AlN9wm`r6&%CxO6yN7KlV&&dxn_spbZ*iSdi-z4qvo+Wr9#)4m5z*WR5BlGteLv-8#U=M(i$h;XZ!+N{hZr z+Q*L{vkgCNo*nPf5j1anZf8tG_7@qih2m|$I@`i6B&2cZ(4mpeye!ZI_>&q^+OkXt zO7YU>Jw=7ZvRf3rszzQiB8hW}<68@J0YdsA5k}gwlGsz!2kv9)BhI?tzRA9^pVsI@3~5F}!` z$O{>*i%*|EZGUs);n2Mt^QLn*KW%?@@XW_(BP3%k9gB6`bPsvR?Wg|;#5Kg8Gn@BZ z;N&-Pa;K-y1-%F9a*jDi!?B)f27sW(7;~<+rb%OZYLOg|1WBgC<@~^>`>BuNl4YkT2&y7O4 z{tK%8@4)V_cl~cwv5%{T4_BAxEjr)bLxLM16QkEX=3oK*c&R^7;dYL8Q?k-=x$tx7 zCW6MiV+!)7cMia?VxTgV66<|LSyS_63+49h+osTfX%{$UKsT58kOwhQ;I}gP{$ePWMcY)N2M8}5}ESDaj znnESDJqM$ur3L)g%S0PMn1PiwEytpTi-RK}4udo^ijTBd8(>&*X>nCVy66H-3^rmN zZ+7Fxja03y#Iphdh28#$dQiIDFkf+8v~&5g5TQwV_+&*{C_cgmY1m2DCs+t97BD9( zupPH}g7@0GmL66b9c_DMQj-=&91#)RXy_5V5<4u<*N9rPvwvEcDNLMt5)d%y9QE$q zDGAk^{+TnKs&EPwB{u=G=71*m#&AOa$_&rDg?wX15BKU_rMJEe8J>e-KtfVd=)13( zMwPTtyNA-Udbol7z`4g9;M}*3252lloe~i#rx>NF`@$f`?)%mBNcm7ULTbA!IWGnD z#>BW~?>TxX~GbYB|j*te3_($RtAle zog8Z5uB5DtdnFJ`6?b2G2e?9={YC2xGoJ`9>0PEPBijX=iA7$fBKqrjh9;pvYz=aq zwA)g%kMUbV(S<{r)6}6(#nH5y52sF?kUn!}?a&fw;k$3}^XD-L!<3E$d`v>{<%;)i z_qLPpicmDOItuU-K|pGPX%h!pouG!0%lN&zmxB1> zRX{)J;-jagr<;mBJnYH!QerN1LF{?Dd;WlZnizdr{qAlU%0ea^h`)U(5a z711D?r#`Rbvq+zrxkwr(ogZnj{^oo?K_);TATW?Q89V#VhH70ILd-U*)PTfImXsVp zMkxy=3zGrYa3>VjOrFEg@(~x{W=w^!>%vK~#r}9%h3NQlHjk?Y?s=D?D{OhRBVdF= z%!P(^DBE_(K@Pv7u)5sCW(IY@$KPK`y6h&e9!Yb3sYM39RR`W8?64AuogE=?U>M2q z0phhlqoKf@M%Oo{2&krM-+ms=i;5B5F1;Q7Pta^8g9Ek9^rhEyE zR<6P8W#1^#7MpF%0u8@mVbY_eI;um^aoiU>6vY4przt>Uyrnh6LxrgC2K8BDpuZoq z9AXe&{G^k@c9dc)sJWh8Ms;yJ(=;cGUI_?o&uz0zNVl!z&8@tdaY|TNyGBAn$D-Ml zE2$y&_3K-3N{EnW=WcwAcTZG^G+{+H|e`i$VksR zpPsFpNS9p$FRAyj#A3WI>e7?#BW@t|P9%ZwgS96`lqxTbz0uAw$ByT> z&CC?YL=RT|;kkV%&(sakVyzq=B(#BCI}j?=g8YecoU8ZTwG>J7Jx*2oNxMMRqh~xE3cT)3&q|s7nJMfriA?)6oaiwis9kcUgskG<-c;#PR;n{8&eE z-Dq1)<$%qXF$aR%vPDzz`cWX&Xl3>0dveihe!E!~l)lffzmph zH-uMna0p@X-<)IFraj*^F9g+1F4}#Gt#E0?b*wE*M?peNk6JkY)_vt(F%0a8>&wMG z3Rgtkma^;X@s0khJ;`PbCrE2c%s(|AdoQv}TWsIxKF^pNpCFaU_GBAxhtUOT$DS?~ zVv(?d#M-;!+Lh5!o@YnIWLKBz>wWpFTV}{v*~V2|y!r*TSzjORxOIS8M~_&vMACP< zf!EHpiU);C2X{NSWA@uak0u5U-IT>?CbRAWDL`Bz-ep`1UB63c1wG%h9>Z?7zUGEq zB9pPbUe4I{)dA?t0!`S+v6V7Pcp8;SNB=NksJ6ogUN{ zxw)ui8OOmd(P;{B^w7$L)nyMwNlCeynQ+1VV#P8t<)==bl)>2z*B-mbvb!lwr$(R{ zK*%RRA$h?VI3D7sQN@-Auuaq8UcF+U9TT}E)MZD&0=7|FJdUAZo|{tNFmxcVs6(&) zh^*fGzEQ^z_nx^n`PTHsRm~1RUwCaB z{gLOD15;mXs;ld>%@i;=Q;JOs2nZN$ z(7szl^Vb^uTF~f%T+|@I3zB}8C8$Y>J0qfJ6EA9v1ei3i$QJ!;W%$j`^)?FNF= z>}tfWh;}%Ae|io{>hZj!fVlX0`#LF3Ja-k+07aZvX-JZY|L|}*CFXs42rscdS&9A{ zu9;{+U^2KK$_`ZEx`NoOUwuP#170j2V(JFcBU|@uB2kNqIeEv>v*d;=&zObkn>2D& z(zJ7OOqCwn*eU?1=Wp)3h4_$Wg=v!4}KGJ3>~k|s3Mtb;Mi?Eq+87Ehx?ZQr%=^r3ih zNR)*M1?XUXLEyQpfC6{K%M;g~RK;8zIPWpn)t!-*w>m{=8kt_pcPz#zDb)dU^rx2o zgp%BhL}Lm_sY#y}5p;(aj_HyI0x`2PoyDEX13mH;B-PA8bL73ZmX_-sV~!5x1EgrKNf?Z0@Z|DJn?_S;RMt4So$v3y@g52Iaglw!|J8ox|`n^+w5B zoIYUrbWMvX&@#0`Y~)CDT9R><^2(~}CY1XpWV%!aq_}2gI9|bgZAA(uosSTZVNbc& zcLLuqS$+8YC_iZ+qyu^xM=)e%B+6l0hU@wc?7h+O;<)#0fQMhhMDM&=b>z;~G3#41 zbuw&Q`#3kDxVFQi8u~%E&nYUZ3&SuC-Fpc|>+TvUl*VKw&=rZ=|eZJz+71LaqM{^L#%g0Is3=;o*b8)PhLz5R^3$bA18JYPnX0k`0h4 zDJxqxqaHklYJ2)Ikf%E}V-Cv&)$0Z-&{i?1GN6tgxH}0ttSek~KUm~syb@W(uY3nog;`Z_V*pt*n&DmUmS37tf$zZuk4u$ zs9De*YK}}nCn{mx-lSoWwoEPVVH?{FmgXq0T&N`GYVN-1a{hzsz0UtbO~1F~$bjm}2|6;GQ<$`g?E%*k{c`pR>! zel;@ivK?;oU5+r`1p8sIwBNy4;622aT+NTSGiV0SruO!B(~~v^IK8Y(gCc)4)|E~S z*!n`}=sr|alNam6^H}#ZsUL?yta~7og2b+Zo%pa#(6sg_uU6(&V4j3K(BsmKuI-zg zJR?rr&%`v~Et@L-Pv>0>`f=Ws<%>cG0;E8n1Jh(M9R24(JpOGuRE&k=pgQcGqOok~7lYQDGyA)}E z7IK)w%&-z#IZD`mB~t-uX}6E$;xjWdktGCO1cQUZ8T{f!?ANc?(KwJdM%%Iw*#vBV z;!lDR5hS_d-|cI!;ouD;?nkqk;W|(qgn<55`aAXrjQ)=vnrUB?sDpTR$7ht zCqP;e4Iroy689dLHU-^zjfW2PEuJay_3q&{Y5)B0f87W(X!b55)a$s+Zc4L`h)d!# z)@31RJ=LJ$u6us0WR>KBy0rMH6HG6ti6)~^K@IE@i=Lw%^DIT^jpBtDub?O}Wf_;Y zWe#xPX`51ubxTSROM)h3(X9vNIYn0}0$r}+hq3?<_EruQN|FWSGEXkbL~ry=e_#NY zn43Rhg+-dAa{oKD^z<4xZ{At$S%0Pk%MH<@UASe9pd|4XJQV>h} zU}Sux5t<_lz-sJpEFw^*A(UKV-BtCcZM&Of*xX_u<+0LFz>gYh-t;pEdU4SiJJ;r6 zPVJ`JFU^=i_Y>GuQRxVq>iMg!S7P6|!%Fflpa|Ni|uO!!Gg@HA7fSeS#LL*zE$#+$=RHn{|0%Vg>{$dC56^X= zn~KWHtmyiu`-{n+p6*8&>??ue7jbs5+)^?!7#ui*n27WDtQNl1)Py#f&qlu8hDw~A zb+?Q4^tgeX30OX*jM=&#w3id!CCAVJv{FcgT$}KwaG<|@`Qqc_QxC&zUNlmU-B$%G zDQ=C)dx?Q36=ZS7)sgAXm|HYP1EDWxSg+WO3QB#qzT)Mr1L}MH=qSrY7O7|l#_fOE zZMSCZ2YJqJ*AdtG^1O;W@-SHCh1XDQsbs;v!1v53c+T1S`mO8Ge6rD4`&7 z?;Bw*?m+(>oOqwC+K{c)505|6^ZahZ01z(t^)HP2<%n(qvX(9X7uJeU#jTRN{xPkC zPPrq`HXpj~*fbp>7$7{By_W3Uy`n~FHlO&GZx2K|D12#Ij!t6DDS?q+AM)h{OfFz9 zMDgWcBOJT(*d-)Pz~rs?+QM-TjwU7SQlzICKkCxAJ#ftg;Wq;rp5utm(p7scvof3; zfFV6U{n^qibFz;0n69aNz!8%pCr=6~&W_x}lfmbxk|>&J5*-x_y20>6`j8Doso=Jy z+Y%29?Po$$Dz8ovS9#WE)T6T-5*xF`+V$FS(Zy)5L=z*{&@L+tKo|~mZcn#dgpf6@ zlN-wO8zRrAzAo|hs1H;SPX%EQ3+!fQAe6cB?Hw1nmlMfFYgmGD6;>VxFSG>kd8B?o z9{TJ{IFJQFWP7qVzOi7D0e^=Ne#U(MbxU2mH?lBxpml9qao%)E>LqACWV%C-5N`GH zFU7WV&QmIQrT9;uu6B8>Kl!;BlMO8K@VQ}qq(cOgx`EizaEYv3;bkDj_pygjD2ZbS zo$sI9g$gc@lRBiG@!bt=Qgj7uw_KJTJ9g|KjiiF;%B%BM*~%&%;kz6IINFYFsG}D) zkY}LNmp@W0I6bcMb=dHt*ohPBxS+s5l^bvb4gEa0yf-i)t+vuahi9nx4ykLLwnH3s z)CqBDM)q}1^xeG>6dld<$sBB)l0dV+v4HS}!O2Hnh!wODZ>r=kKF|-&*ZV(&C&;{c zU40jj0+GH+J}ImAEHLaR=Pfmpb}=`bRRopK&bH`)C8h*dt((4F%GAJth9lLJ3X*ehcnkKA zbf}YC@`GgiU=j{YHvfU1bV6WY;QrBlTv45!ov?mZ$Sxxqpx7<&K3@Yd_>zolr~Vm;DJ>mcLvD{d0faeK(3glm|NQd<_!ry!uB8zIG25z3_)d3$7RVE%v9}#| zK%ckeXFo@z6?=-h+gzIoSfrm@<&T{QKg^j{BQ;vSPz)U-Ppv?WXguN#4b^kddV`@g zgRyfvmVw#E>P*R;VR&;%kY10y5jXuYI9=-(g?sR2l%!DP>M>QrsAQhZ~c?T}$8 za{~p(A`dH{M0)?9@e!_^{IhgcQ{` z7&@)xd1?!cY_e~a@6PM8f_qegUWJ(L(Xm~6(;KJBn=qGv7GECFy$-Gj2^0_O2)5EK z`32ASGcqQwthn-OzPUs#nD!TT7){3T<*~OKc@@x0tSG{Ti`Ev(CCtBndJ2NNBQjs8 z3Kztrq@;Y{?uS5Rq`=98C;Wetx{P`0^Je_xIn1;jt3DFT~TCnU#gu*g-7VntwHVuipxfh=`7L z!Q|um`ufIfGZn!*?W%6eb4@T5f+kcC=@35fo$0^8vKuT%(r7dqENzY8EZ8Bf|5D85 zEp~o9EuUJXgR@&-ou`0~)f{16lwv_FRPXIA=&S*?Nfg+AVMuw#)D>4?_KGBz8ojg>{C zU$8?1d1<%z6?KB(XNJciZ{D1ix_WgOR)A|Jzz5hJ`lNj1^G|0qaO*2W60noNyUd@J z0DeFW&Z=n9I8DD$TMRn`)JXX`_f9c3*NOW@D^r-&S)vDs2ea&GFf%j);feL-BVEe{ z0qi;V3~SGssVvZJWntvbHFLi=M91m{*i3-c9{5xtsR4ohn4T>iYHKo3b?ri|{Toc` zfpHRr4|GNi9PBzX%4%zbcMn6k07cbDI@X11dif5sq7K4QFu;I;s?;6=ts5j+ec0aI z6r?%DPkbM5<%(Z78lnC4u&VgmX+^6xu02E zn=j^u9SN|N5q=l1EfGMrc?B&Ln#GpmUVaxci-1w`aq?G}j}S`1FJBImIcYnJJ&J!m zew=PKFb<}KFrx`tgQeJ#k(t@ofQttB#efQ{0z0Zqv>`;nj0XE>&QF#XZU-YfX=SRC z1D&Rm+dM;V2o4PFTP*R9L*5Q36E=Ce@N8fMhKZ_cEq_@Xy*9kFXxFKygVj+4sHa7N zYP8@%j*Mi1!XX107#X91JQ6%5 zz(+N7Z-y)Z8M(&e9s3vKEKUv%DWt^!6&eYtp`s_+Z!J>*x4_;f@Pt8aH?kc0U74CZ z_=fA}tvWS1y)%D8L?Z}PMF0r|7VO8`)rbXCKG1%#0(gc%;|h2b16S6=k`sJqAV~ zFnXVDy%lgTCr9~HX(_oi<5uM0h!^MjPJN4~K)5=1&)d6}`R8-J9 zJUx~K<%sm{0o^%p&DI3}EM(P*Y-Mofvr=t4eY4*rBO?RZ=>#q7YgpVafsq%Hk*{H} zYyq!pMW&brgg_M-c&{z@tWQO&@kg?rfu^#>V{MvE5Y_>-E?NhzV?>i0JicelAz$BR zOt6V)U;tE*_NH2_MO&uP2ynI`-<(i=YzlIJcRJrG!4o%tw`>P`Js#2xX+faRL!30Q z_YrxGm=GQ%0NcDvvS+IC`&Nh^s~AXZ1RmMRfD-kff*`hGJ=e+7%jswBz`~2s(6eXg z?Z8BPtmSokth+%OYM7o+14u`oDr*x^^*d>CW>gt+RGpCo8z+!Glt|P7 zx5z%BfPYO%&zC=`fLDcee5i-7ahj*~goPsmA}oFPvC^pprd|rT=h@RM$ z@5s%~?Mtt#5A&}m_vJC6F6W^rT=h!J-PNVp%jsO_pag=`wZ{7~fXxVaCF)>n&B6W( z$-oA(DxfXk6$1l90nf+D5+sVZF3C6s=b5zG@i4 zIAsNcF)uGKGL)+6>lmnDMQ+Q$IV!&~Fn+8)atn5B;3KTm-%_zOC2-Po*fKlE(B&kB*MYE*=V12-{Tq3cVlMVTj~#(c&O8x*o0r zlTT6#>;pm;9=SBV$HGJKcqIZmF$X`pFm}fDO(iX~@fdE6bk&x0y&Sh?hbq`41PyI6 z3^k~W3WDxJ;2!~x@Ml|zAP9F(aq%MMwafH0&{83S;g^%qsy#(+O>lW&i#iZW^F1EL zh$zV_;y{!C_4!PvZ4DVV-r}h)mcYcIV01PlAAlTEslkgOa%auP>}sOx@mMc} z5h!Jc%qcCny{0!eu7E!V(cWM`Tv@F9>JsR?0_%~yWgu{6Ln*B3LpT$>GZhG?2O6q} zyz&m|?0N~wiiXcO^pnA%9jJmOB!Ded4V(cZwXqKoCTG)=akBuf4i6Rqjs8tU=LP~*4|O0OUe6XFzM-L^ zNyt6-xG$$6c>tgboB#AYmKvfT=xJXQb`gAo)OJvx!%yFMpv%X^Hd^@RUY~T~WEnFO zVZqVS#7B>45C{ikizXYK3E?+L2RAE_KrS#P!s~7}Ses(1DSXDY|ACU{>dJHt6u7O9 zl(~AO`~VB=0>CiuA_;0VPk!`fb5BNGH^{!P^t{2t{yEB#u;UOIHVF~u5Ml+u`TfH5 z$2NNXB!%dZD&szVE&U5ny1U)x^^k1`;TGw=n%oadyy>Ao4qwrmMs~eI6KYq=pg{x8 z1qaYAuw^PjMRR_J?FG_~n0zhIM}Ur_lH&5~e=bl)gUH^${qF{UY?l6`{}|zyk1l)U XiJo4S6vj0mEf!iz{%X3U;e-DJlSf`T literal 0 HcmV?d00001 diff --git a/doc/LectureNotes/_build/jupyter_execute/week34_93_0.png b/doc/LectureNotes/_build/jupyter_execute/week34_93_0.png new file mode 100644 index 0000000000000000000000000000000000000000..49c6e7364cdd160e78496a6386300a5075666900 GIT binary patch literal 24893 zcmdSBby$^M*Dtyh5Tz_S1ZkwZLy>NfZt3oBR1lC_NOyNgH!2D&x;sQlKtQ_T%p0HY zec$iwz5m$zkF&3HUC(nZUF(i{&pF5V#Td6Cit-YeXoP4m7z|TN@})8icDo7&yLEB* z4*2HlEgE6)!Rsuh;jCh3>g;CdXabWnbhfv)bGEj4P3CIi=wxAM%g)5f#L7Ts?(A&u z#K+8R^REM#>>SOQCk^*nz(wxaOKLj7VD}B7|8Ert=Uc#F21QaYpTBZX*_!k4{-twuXHByrtSqHcUceYL`sQ^QX?o6S3FivoAsIXnG(0-x(F&c``UzQIzs-|REy@%P7Vpq=9unJ_C{Isw{~40gaT1e$@p3KAtE^vOzqfvF zjUrMYD_17oa<0*3W2!3O^XPXWp1!wIVq&6F=YsQy+;po?(`sKbUL8K1ZhC-Ts-~vK zk%u^m#W2vec~{(JXI`)-fLf_2Z2UxFDZ5PUZmz(-3_$H3IBl`rO>%hDbd0ZY&sH&^yZSVW;)Htkz~GZz+PGY3wQ0yVUdn5k87U6l=-{ zcO~J_FEI*24MP*Q4J)|sUObgjX2BnQho@JtHB(2=&aSf4yf1Iu+Z{_^<8$sRczv<& zx!tfUT80Umew$p$h zw{2TK0lVohXsy!(y$-(P=}W?W4~zBJ=iC;?!0H?>$0%%+*3Azmaaz67)-DcXZq)!c z@yR|fC@dV>@jd2m+3)8EOX1j{bG94lSJK1O#K2~{(CU-KZWb}`ckRPVjD_;%IVJ(y z%en<$uN;Pq%1Y(?f(KFKrP}$37#gd|3PY`G^Jp&X$&8|+XD!!P=iS}imiq(v7Yx{0~aq;@Tn!auNLg@+~9v+QbwKL>SgX0aTQO0** zGwy4tZXce%xkF%<;t<4YH!raCIRJ?vN|Pxg>adFxpbtN9?=(%EJ?x22^e z2>TDgq`s5*e9i^C-ySRAZ;qF;gD>YjHyMovGoFi^n3!bmv|a4=gfF%S6G1cl5bYgw z7EQYMFus6$#V7C6y{>vj?OOadcQD!Xz@&89jE#+TXqu&(==h$lh}+oMEIZt}|HPU> ztGZw;y=_s*Y#?oJV{)p>j0l`!>8x#Hl6y0zVHsvl*WZ&vzt$L-nIk{!pL%!X#_L(i zrtl0+R+&W_;sh*M!(U5riumNVGv9UGm%=+z|0^r(g8|Fr zuTKL!OBmT18K2!3-{C>ywQD`?;JG{)2?u}3lM}Y1#b%rLJ-0FHih3#zLeb&ZU}o2Z zTB()^1`$`U`^jQ3MV;WqpZ1QfXo~P~B5rZpWMRM7n_(VK=tq9@nX#!O0P(@O-MCp+ z(AGI)+r+r%-2lzBS-Q_S3r711NDY?Q*w`<({n~>tZ|w}r1@EovF*QdM?ON+0Xy(G08fhK% z2?z+zP6zzTl z!rhos#grs2qe^nNPCUeX8wJ_bUdP+i3o~tgelF|7RC{)jo!I~iG)es~9CAjVCJhfO zyMSpLqUNMa!M2^Mq=H{iYFP=JSy&WOt{;pj*gZ0{u_*;J%_s!`u2`*1r_k^E*l%JP z5yJ^i>=G{pOU>;u3)YA8?D!r{3R!;DVb4jm#mGSik)1gLP6bWsXt}#C}% z*woay-#Zq6etcoW5Z!4xPNSp)z_+rplEi8h6vv>Mm6Svn{zOpg`}gngulqyA>dd-+ zmxpI_JHGh?!q+-*=grc(+7mB2dU_RF>`=-|%peQ`^74t|%F0UWg~yvt#3Fl&?C6t& zN)jZw79)8UAYJph|H)V#NG}ITPs44_B|eTxmvF?#o14H!!|-z322Nhu7J+TU`3X$% z-udNpot=J;+O8zX#%S+eH2w%gC!F*B^W*v9?phxz-Jd=VK6&K@R+^O+A+~^9mdIK^RHCrcuTlm2iQNqp!z__YNBAb39Uj?@X zpE`jj3d^o-3*`W5i}Lcs;m*+R1+nP4`yI~(q$$Lf>*lP}z?iPvT&__BuQ=F{^2EhEsY52R}*Er zE>kA)7JW&aN!)g6N858mr{)4mS+N|y7uxinMAGajSR_MP>*wo98B%n%7aKy{qwJ=QGFkG4Dy$J7tO z9^V8kYg~r8g;Hu`uBmRj_1qR>ZqMlHM?ki41~6HqUar^O(_^)}+jhE|Os#*unrufN zA_Q~F?gjZ2VsD;DQ)bz|ee2HMOS|7*y_r%<=C+F%4VPYwt>?5FHwWpA(brl}L!)40 z!_sE1v1Vv1nagI_rfv>*eSA<4thud@G!7a%I+awq&yi;K%za{>3@{-f7wGAc{_}lZj^{6!feX0%VPvKHRS5FwXb-d_e|&cCeleu|VZ#Lk8d( zZOWQ#w~w(g?zo(9)mdGh9l~=upLofFY!i*mWL6KZ<$|D!f7yKxuuJs5M0N={;2xZY zUy&~XBEfUI63o%i(3YL~4tG0Axf?I5283b;*22D_ex3z1HQ8gokGsYg@rWSkVMA$7 z&P#MWrcn_3gkv()p%a^9Akh$nEEeXc4Y&H7JA-5EPJGmL9Nx+EEFue@{CHDTS{jS3 zhw1$z0ChB9fwW<6^`ye0`y(A8uVaocm{qHt`4;$Bsyl5LE6gj74dnT@KHwQA)b)LE zEO1=Fq)IBsnXBu%#0GvbFgA*(y!>q@JhDC4bQmmr@sy2wbE15r-1m?^6{H|=87e4h zgaE2{p*1Y*O#GgzO7=!xNOD-}SWj{4A0oXz>68q&CwnjmeyQ=;U+oGii=|iJ0H_n* ztZ7le-#;)AkF7`L3zoXZb!)oIhZ!J%ZJoAJXN1ksE+B{Qx*jXBo?A5&)s|z=q!!_^ z5>Gr{f?N-w>rh@gs)!}ehn*`iq^R-na{Y2Z0}B8T1&ABfv!V8$dDph0d)3RLIrMdlo~a zwoJFFNWaY&8VjY=NWKD?nzrlX>jOZVpp6e@a-am_B-5wN(mce=TaL7rg%pG@odtn> zfq$isZ@b*GQ`OTe%Z}iaW{Q`Ip_Us~Ro5N+S+wnUIpMdS>NY>rpUOX0s$Ex*91G=a z?gej^;{`vzoPpm6n)DK@vIcQzg9AO62T=i#+y_nKb@QtF`DD0f?M0W@di(mu7OsxC zr51&%6a;qPLiEaIYg!fD4J7bJurDd>@If*#9Y`w`3&SncuCv`3$_{QlZrND}*r5j4 zy-NpynK{VPjez}AL=J)MJLfuM1H{{lFej2No*B(5Q_ksnmv&5U^Y!6eLLU2!vW8Vc zNNN_Vmybib0q{?-zE$~rZrde5f^hI^od6Cw3^FW5{s_U8w+rB47PIvZp1(4K$&cF{ zSGv1)9Kg2o1RIucE)r~_^s56JMvV#)z(~M<6cQ1m5TB4bTkt*Kl5|6=n<*6`%h+A& z+>Fz6zpju0MOb+|z#p)- zkMZ#n1rs|U3r%wz!{9)IiYRkZRslHk4yAbB98Sn7ji+NDu-$Sz|0?+QXK`9~7_1th zZP$_u^ij(Y08$$jz^G$lzQV|cAUSSR>2+)iqMSzgz79~AH~aJKa0C!c6=D)t6RTL_ z^LEXNg;K+w-d@>h@w>2)I^RprInT|~=x`yL6#GDsT!ja(x&jFsQA;$d#DS6n^qr8? z^5Zs8kcV3{++&NkDhNg>t3XhU)jKRHlR&-*!&^3_|4{!u942=lx^uKOvvPbs2B-<7 zpZONwK2|0n1jottV$)XvKp6v>qXMfuwEqgr}$YFfjh1h`Lams~n zo4}LEG?3UqG_#Cl0$b$sH`(Z3`9wA`l{{G~7kqG91-{Bh0`Ljjdj5ZGrhv2mPdu3a zzn#{gLIQ*3q!@OoXfmxuY@r>l(_f!>BNioHUcvnFqk3;waPItk*PY@KsYAs5(iR=7 zOR4X&YB`yM>Rt3L*z4zsw_r1SOnL<+1S3RLmGI^1`sbSLc|A?;hN0w0uy3Kpxc-6! zL2PF86cV)9hdmLBk_7i>=?TQ)NJy}16yRWK@4brS){GW}d$oId#Hmx#k;6p1p!-sR z;W&_>!Xz<)-BpXdF~VK81cMcV75yA{v9ymv95wPYK}5dnVrdVK{ScNR2!e7lrad$1 z+zeqa-iARapKrP273Et6GjIgHnRO4hMQG8QP`QY_ZT1WjZ$nYqxqy%9tI$2+x=YrbI*X^?X3cgD84h40j7Zr2FOFaw}z9$hXxbjS2Tq-)&HGsk;s(j0}zeV z{f_}qO9Eq=U?iWHWn+7{XB$-fuJ+SaNFbAx;p;2JZ;(!$K|0Ro!vNoCytRN&xy^e> z0C*J>6D!05z9I~F5txE-X8?+jkWAutEAtl8Z`$cxVUx3sK(+R;tfuM<@pb0zf&0qA^}^z1DJwUC`oGWdnCfyqSj>oIDT*a)bRp+FLI?8R1!r znC$qH1|gA%G``Py2M5zAO|3%(4P*{SJql>PZSmuCT<7v~w=fg4)Wa{KH%D6X5D%C8 zF}4#3lCY-8kIf&GCASTd-<-mSr;{KSujp`JOd3Od@%K4@U%A(P-d9)Q~of5aBQ)0EK*MqVA} zw)?`QFJYcGI+k*x1G77cRd@b7x(YVI@3)WxU!7`53!VS=g;Lw^yYD{ker|yqQu4A) zVnQ~(IlF?1M-0vs#lII*_hvDRds{x0LCX|(;#gXnt4%~gAf7nll>`cm4l1AE{?Kh!Ol04Qwpt46viWk0x2VA{ z4m^1d3mZ(5u^R&9A;GEhs6GP8-x-=wYTUn= zMK+g^mHY)_l0XO{*7n!`EVcumVMM`q?=fJ|tboLbNS}7D(eseye^-9<52WGaz!HK% z)`e_mKx~G9To;uZ=HF?~hU_+t36N1C9h_tu8ao9Q zgRT~Kj;#Q+@Bwo=Ha$HZ(6Pyg5QWJ-xg_$1Z-~Vc*X_HV+)M);ySGG&637#6Se#a6 z8FlIv0ZL@8_NNwq`0(LXF!E<=28XY4?N9ymo)=}qU@8vyK@EI1sBfNM0$*&KkUqJuOf{4_GW z!`){F9iRZ_K8anPj42fzHro36!gXsu|Be9UHxy!_itw>xtmnk!xNIhoI{ge8j@RFd z7J?;2(RE`0Db%cIx4Zyc4oFcp#$DCF{KC)g^&_>ph0?QiZY6hPe7eSTON-=W;abH% z^5k*E#0^%#3t}}@D68=F+%s_*wMKu{&NaDN0h8FKG2MVAN5miRjV?emc)_Ww%{gD5 zLq~17M=RCCatKF<9_E?%0jaA6(CIq#;yj#telQ4=h}_5Q7MT){Av&z24-j?8Rl4zM z*YnfW3Ve||fL!75JotqzaffWq;VUjoBbkodzeXJZ8>z~q|IF4aSJylHHjfbrc7UZZ zrd53Zf{bY2{5Llj1Nog^N0szMl5ykH<~1~#hyP=2F<&q5^J6GVFT~762Lx3HuVY0DOTxHyDLe^0f!1I}po{T!ZxBuii0PQYugb42vRCFS^wL&v z773?o9eI`q=*`WXZ6v^Ufc1rBKJ@ND3b8=GPbpnuQ~lbuF7`Eq9UR$=^cw!E-=@Xs zS6#K~#KFP)>=+F_GAq+ZF2P!zFq>FgZ~(S8Q7ItN_2V@ySv42Z57bO=B_x>*-3HM% zYO~isVn>2ht<+d*a0~w0GF#P4|9r-rb7HTj%KW)RmMv7Y9)afj13hy*pgzuC63~T* zDgwvgSmg${=mxcjN{?(UtqWJwLDb9auhsFdt_~)l$BEM8B}+PQ;wQif^cVVE*rUGE zYdEu;3$ypvZE=i<7q=XbO}l8*q|y2us-ADR5tQ`aW})Jm|IF~HCJ-gzg%43_BdOl# z;^o~>I(0R!HcdNiyw)%z18M8#fW(u(?EeG|+JaVf!U%wD-nR~-1bH~o7bLZqqD@x7 zY-B^ixnTy3;Pl6tatLc(UEQpoF#nE6vwU}856nuz*ST_$uKm1g=0jOxA)zQR!dYanJ^%{@vMpy9z=8gt>>tr=k;8B}0YF57VIXf7F zDg{;Pm$Hzqw-SL3zHQGG@io)|6AXLAJbFBtBdC@fMi?P@q#+2>`dN?uRTijAN<&3{ z$ne|*z!dWSK5P&-=&6EOB43}MSXyXheYrnOgaO__P%?+p2)V)+r~8NL*Qe>8peW`7 zYTK4y0>7m+pu@~{oz^m)elz>$$ow>|a20$3e0bpW9*(Lgtm|K&#bdM&Mv(AB9`FY6 z&c;CVz{4G$TQElRb%6tr2&Bxn{8c>?WS#QYzH29Q`Oi6Yl4!Mf)^OR)GqkgeR+%Y; zJ@ty0j+jp*+>ug4SQl(2 zDxU4!<-GJEkk2oinPj~YIu30MwHUrYnHdqffj))}kmYUFMPUB5A%_A7uTqmEklCOF zNMi}~!#_$HU~Qg$Eb27;RL7wWKSr{7m&Uu8{!T$Ht~GjLWMI{_<&}nD^K*9Admz zqI^Q>VYE@QSiu(_D1{NXr9eF=9dKqxI}OqXh_+6jKQRWkDhRVBK`{Ib`soae0JW77 zJeQ$TibOFTj*04~OrF%F7Gy&zVi6?p%cD7It=^z{u>fi0pO88(y2)aXH3 z^@}BRBqofDkxPV;!{59hECQGrB;pR9j-MW-&HqBX4YLvLW8jPUd64s(N&xde!GJuO zqVU}v|5|9EiYaLV#QGLCeQuc=BI*xd=BUAjIRmkuii*?%L=J?JCE-C@ZOFL!O*cp| z>DijRxk>*{{p-j&=1aalH>>?|9rAPCbG!?RQmf1ni`F4N;H&+YQhnQ^l#+%GF8`d> zIrBYA%^*y<`z_Hc0E}T-a$z;!hToCnj=n&8ubJ{#p~_Z~DiNjRL#%zz`_53y3Uxcn z%v;|m=R@)nBTLPY@S~*rep^bMg}8{-m*`iSMcp`bZA`ov6h5wr_O(@lIqjPBeVplWUw4wEv|3}a_UD)|X^tmTFT-~j>*Pp_kz%o2`h3;?@NSX$`syx|9#$on^`Tt6y53u4H>@1qrX5q)EFoi6oT%dR=H!*zHmWBBNHq{(K% zQ#~zHTjVI#RPGZ}l*lQ@{RT(}6{x-gN_=hRpgIbg5(Xg&0!-&zJ`UK>#<^wv(Hnjh!~L7t)Gg&A0o z@C~jFKJ}Q_XHuAA^N^8Q*4xb_4iJek6gSO# zqY}fk#GO$GVmIsFrP=L>K9y%{GrD?&BWTWR_n+ZZ{1F#oUp$1IDG&U2@6cA4yAM-c zqb1N*u|6s(lGP2+9HZ_abSq`IhfN&){5rWM9u^C5@Ib@H>CxwS&Ufi-LPh2} zLRAGiJwOZ5nBeg&MP%nl-cDcl20Wv{K)lxtO-?|YkVjFguI!YRCau7B?UKxvK8`MN z2T7F`bJI0LTKaR|9sZ@8VMqxa5v~6W%Qa6I=S_7CC(ZvY<4YSGD;0d7n`ERg=3_c@ zszZ=-2Z7@U68gvq`pBG@{S|bL1l|(^epgP18)GtcbFSLSZFP-jQGw50Mf3RMdkRZR zn0nnpJ^oQuGkqCH4S37x-oC-4G_m_zOw6ivh_zb2?ax2`D%pr=njvqGixjwF8;;k4 z;cDiX!J-kg>sgDtwDxD_C}?3gw9nKGdU-%Iq3KSOVe3-gEBdOo(>}Xp$A+6Ofz9(b zUmlbSQqELAc-^W(Z@+DfqRsdH_bfu0c#Q^Up9DW9zYNJy%SP47{othjs^rI^4L2Cy zyBCgrjaq5Lk372L5TCr6O{Gwsn?#l>o!`6=qs^TZJTWiP+ehsD$U$Kmp^&O1%C!=$I=&CG}Rc zb>nHf0vt&4E#Zo1VaZ8)uxYEPWV0G| z&|*{F@&aWuNmHZPw=w*^fnn(_38t+?-EPAUDAO-x*%u_yR{5V}i1vpQEi=+b=Q|e) znB+Qc%|3K7|8eqoS$Na>6pf}UU7vR8SN3CEwJiA5$9Mu`Q&UUOOHhc_zBfJmZC+SN zSW4-Z#ka+%bU8hkYH+!gzgP${Lh8IxDsk8Msv2PDnO zlO^dCOdpcgTD-mwgDuZ~x}2#DfSWkl%;UT0{CRup&;&`$adeDsKO#B#BL!EAx%rsm zOZXJWP-YB25PXHaptqu~nR3|SbBnDFxXscTLSKn}p&VU4vob@gxClMf~7UD2V&UiLnzbf!@JKaZHVP=4rXJ~whQHW0I z=MvWAi>HA1<|S6!fywa4ArYV*@QOX$s*FRl>{0LrexYuEkvO+JzAD4G2X})nV-W*LS-;S{0 zuMO@=>sZrvTtPJ_e-L%PgtPGE)gQfFV$}Ow1g7pSd+9>dm4-cjXZVFG`Pj(BaS^LIQWUt-DPj{_eaP7q>^) zvt+c!3hlP-^Q2fXFSlVo`$1V~bXuS|odcQZ-n(V3ea(97jOv$wKJ&IsEf^;(IzMCa zepx?YWf*;7w4c-4yWDAI1l|7RFE3_PAbddx6lv?li?m^r#Rw!Qf@UV16WTLBM;R`v zPJAT?K3Y!~-!qh3*|&hwBr6*}&3T&huszDEy-{+?zpnaa|6EfR7G{cLkvGbsO6|od z3R|TYC9g&`Wa2(hU-uq748~cYan<~Spp!UeS#s$c zH4zKsJ~+ho(_ntlO3}9qqb!@ReAK{Y7s|)PLq^kVSezkw6_{1+ZaZy(ME$R!BqSTH z_1m2{$1z!r${aQYUPi>XWza&4TcX&u6oY5{NKIy|_Erou7<~&cnBMQ7+{y)34t?}` zz#iWV1AE-(O#H0*3@sFULzs+epe?5)Zrc9uEw+ab!PCYW`NayXm_%c0Z*!)Lli+Q5 z$QqB0>1*T|Y=_V(%))E5EITn1ZdO&YGbSiXCjw~I;rNW522>V}A)vp`)m{0R( zsd{c;+}AJ0F!O^fSjd}gM8{%XzwG#(^W)3$Dr%>i(TKk@UucBCv3F8Fc9@vIJX^)@ z=%glJW?4`wC+y;~S8}452G0fCgS3@-Y^<>7VfzOr-DOgCy(_Rk0Ak?_Det*I0(o## zGp13?vK=!4tNp#Q_iH6p)fzP9tcmRZ=F8+WLj}<0O|Ayltq*4|?d!5lxG!p#j(b}6 zvj6N>(g~+ibA2shY3KIQ=QLAK~&38UcJW3@5OI!lEU%4E-cJUC$ccz&V_SCRAEV+I zEh>stJ}~9fJS?HJP&bQ#wu^u4)bgxNPP_(mHNKh$l>+Tin_&a(K6)~72o=TIt64$F85aty+e2yO(}l-7HqRqj#|nqn?Q_>s54ZYE z0xND|#2KAqKBt?+Nq!}v0}KE9!sNKCddE9I8$YdG`6PQ|L%3~lgQyCZdtCBtyVzfW zImyv>j$C2A(*Cx#8u`RHqf#e|qvHy5$^+Z@SbfTDw;e3u+QhOo)7eWq)xMSnfrHP? zvAmZO4t@e~O^t{|Cf~xSlGM7w2Zy&4V(2G~+LXqQ`)gOc^NYtD36Q;x)%x1oYy?Bz zv$EPAFmlV#RI)FO{<6cRo+$2}y`uP?Rl zp&xEAuE`0jj=6BtKd`m5^dG1mtQmMIv7>9qYNa*JX%*)wO6p#1>XGo7!5QKAo8H$}&Ct>+ zx<>Ze7gN`y_14ynJm;&saD25Dli$oR>yY7`IGbJ`!pGz+TW-f~je|>?y<9NBu%yRp9#N^bLtr6Q;G`G$A@LrG7Ww-#&O7!rpgB)e=KLQf5w=T5h&c1K3ykACj*w|P;{c{NrYpwlw z<r z)>|9&_nP`cO#Wbuj+d(w>oaC?6dFV!=bJMh()EyUJC=z0=JrD_ z_E9L1?wN37p#`f);iD%Qi{tpDz?QAoz5yGb^;z?tmIebv^e4*b?(%pHJGKs$1f52l zUv^wRxs^e|H(7*umPB7{|F@LTr#li72ZV!9D|PdRmKM2!SYgWfF*kuJPbyjpmjK{3 z!b}R{rgh7m=o-WTk}0+6gJ1HzcxobGU+(>wCjW}Up}L=C`D{bRLpmM-B3;2FB*p7k zoP-dcjR9N{4CO?CK*B}AwnT&wyr1Py8MYGj-dQNKdk=7L%+2-?tILgItBL*sQ6!r5 zM?&Pa85&m?w+RFzKqeD8qey(`0*aP6Ce|qdEB2Ca4|(dWvC5mr>v|l0S{OI#KYxDQ z4(Rgty%jtJ&uRjt1lr_%1Vf3%mAA>9Z?Mk7+@bA$IC}*aR)~2prJIkv?KR)<+L*Q$ z`NV2CNJ!`R--Fr5!=?j9@7wRlWPDKSfo%Eio|V^27|OYCLR1;@mo+xt79q=sPoQ>| z=`Pq#KGUW?K63%5M{eRV5fzksVIJ}eF-6*;RfoJ08%jbSe~rJv>T8@0O4lIefg(g; zn0*q9_GX$pm0D>k+-QQ^M55m<%U2_kAIl>`OCj7g)k~6ZX)hgO$htU zCQ(Ri!MF}(RR#8cBG;R6c8jmFe!qXFYH+$qN36>iNN0)fZ{4r;>~$!y(<52<9iQJU zn^tME7Z=xcIYpz?`Ofp`BeSp$+YjcXj31vLkFa@Mmc45I#5Mx%Ad#*qVx96FjVdEG z4!At0)bd;l_f)BeqRZ|_Bq2dWCm!=-SIgN_ffL@wyzbQn#0h}-Qv0*4^84^SUygVf zhG1ZL7tSsj`?h9;?`PCq?_xPSz57o-uC;WZ*&+d&((od}paL|-D@dtm4NEXw;+E8W)FiQVW{Q1Qwi&`sUsFul3D^DAIZD1S4rV4+ee_saQD#5X1x`{|r{Q~kAsoq8jYc6! zxL03(^ffo1Tby67`zd-m2`b>+Qco1JMCBroor!*Uggc9{q4(^9mJLr}9_^eY>MU5K z8y$luVwkp|(j+504kqn#JHvOGuYcZ`YaKMhcs;or9f5SEl$|S|3MaH=qb!<%4OrzM zH4=PwByZ9`9_v!l>h4J7Fw2|kuO5-kFi_l9P%BjNC~Q=V>!ziP&d$r5TFl&@bNMA) zHu8nYF1pcKN-u>j>em{{97YsF^=jxX<~v?Lews0xCfiM0e3zo7oG?oVNbmWqd*#9u zP`;p%!q4xc_tx;}+z*D>BDM<${fW<;Jvcwq*U(r=6{Ss>dihX%0uuOh*q0~uC%KAf z@^{%*ytyx)oV+4)1-d$3ze}YNf|RUR{XXYOSWZ04pk5HXkER;0PYY^)1ey}LWCl)> zmUdg%0_Ae2XX0N6&2UAeb2r|Ssbr<9ZDNR)9a>RzwtMyfBcO$TWAHPbD7lYbc-9M- z0YSZBNn1*)2543^uz}5qfh)Y<+>pfgw-Y2)OGy%*lFZc&B@KkKM>i;$UAc@p1jO7y z=s(!z`Z2%^AF482wz94B@oE(pt-F=s0-Vr5O<%E?$ANwE1=(NiilY!z` zUJim}Y-9-L;Wp44rD3*6BAdV{M3xT5k(0ms0;#h<74Z9ccy6B%b6{us13iyYyLcSF zGJQCO90gV|mO@$)sI2Rb9vKJaG)k@Qi$AIrXG5oyPAV~Xd=cftd-OXxzd7gX{p6#D zwllVkdV>WckbQlC=sqNkniUyIG%VQ9;!fJC+Kd{t*u72C?3PzpP9Z`b-TPGwLv(vj zd0ixScCHc!-)LTUdA%}AsxQrl=SUr)qK7+Qq1q@5l!w?~<NmZ69ZAP%nGWm!DaZ4a7D` zX^d)rMxVlCn2-vixF#Nw1%QUX)%cvp_CNcJ9Jt>+4rKUs<0b0pb19M^hu&Ygi<|_c zjB$5*JSiQ~T?|1v#|GsGEcEp7fAMlW@dhBN+t?-S>J6Cb^XS{{64Fl;&~A$t z5C7)7O?ex^AUn{_X#}qKO7QsfTe&&)ree_9<3LchDw^q=(5`c{ZTxe-qPMG5ud8Me zB0$OzzciRh-Tpy7b(vJKwsl72)9q~e_~v z*|nF4j^fJ20%4GMT6?kXg9|>zi8*7`sftX;j}Q!a^j4Cd!=@;7k_>{Nk}h=Etk$`u z|0lA3jCp1wGS#r_@HD9_h!MhZRaLoQUSa^_Syvg5RskyicU++V(g4GLA14iOFWbWV z^tipE!O=o%SQxhuQw^47N0RjS4j%dm=47~GSIto|xRMoYEg$1-_kv22NabY+mOjIFMg3Q`0e?2(B zNdM6exAt&w$0k~E=>acTz7uM?3tfSAO-A33jMYYCctJi%)vW99m7 zZ-t)V(QD|1K9bppCvy5dO=agH1yrrq-1LnW07QW{iOdhkfzI@(Wyy#1LIJR8rd99t z$&)q*ZZWndmrb=f(6mp;V&D(nCQ*J86NWSVyy#KS+oNXi4tbc8#^mn#2a3+tmc5Vh zK;*~8c3qw({mZ?>2JUW-qI_AYdI{q@1u}oGFsgFMf~UVV;jB`8osPww`a|SY^e^U}N6P_?}!SVY3eZm)ep?c=O-*e;zryalrdqe(yMi zLE=+=aVoYbR4(=yTVR(`rvG_S;`-s1ZgZ3Z*`r?B{qP7Y#_}ScZ4?vgjN|~&jCH3c zo>{??jb3y4_4iDyaQoe`=ajb+w> z%zcjzK9cz7WH@zQZ^2?ue5gd0d*N3F*tjnb6XlAHYayfl%TX$9wTw2#s1u98=lq#L zmhf4c82aeSmDloFe++p-OtC1t8AAB<0$V(e?{BBnwD1C#LBm?OA+6QHAMxEMbP4>5 zV>{&^uC4ZeJa1UgW^x0dBprvS`6#T79m!&PC#{w$c8nV)+s8qwMw|14 zL|EnUEop)&4zH#=&BE<4Av*#6GIdjoCCgSF=IcQvn7wSmVS-VGTs(X7We({QRBo>$i&d6c6 z-&ulWy83m&kn&o;#4O%G`Kz1a5H}xrDSXQLYbKWY{bC`z*EZ{qYE<=Z6iho&t#o|1 zisnY3|KT3MjcYRO&$B76C($PD3Nj~K$>o^97t9>^Xy(9&Acw! z*(5+i)T~y4Tmoty&$^c(Qe+2ubWwsLZ~S7opLN0yJUD)tFjmS=@ z-7YVLBR_n?h&j8zLszV$T0`EpN$VZziDxR5Qv%okgk_Xgcpe}8WUbp5%^2G3X zXD?R!0#Rq33oX;rH96WG`TaG7irTZayzo8mPvZT$f1Op!lc>PJ*QfjF#>U5zjur2@ z^%g)w#*9oBT(cWPmO)h__Um-XNwb6PJT~A8m9Nr(Z(8d|;-d58Tg=xG;Qe}^im^rT zQX`^o!+L{aHR&tb?{V1hl!m(Pv9j&%cbpz_0e$xF@#Me#R*8VDdmGN%_pANBX`Du` zr!GrITi;Z!YVp)ew|f6+Y^D4Fwea>g=PRX}*^5Mg4qB6L)t5aiL;D?{obo|A22kK_ z)9GN8a}s91=Sua?85J6tzMVZ=u0<-YK_7~=v`lWnlE@Opq&d|W*8jfgklfb#Sc&&& zJ2t<$b-+q9WLDqzp^Hw1!U8}WeUKq9Q4>rL@nHs-lvR=>?TV|tURiao;mdy;g!kl9 zWCM*HBQ90{m85q`3b*;tm9q|_fV7fwlsiz#`OZHugh;`ZLpIO4WthbTUFFy?cXOuIji4(=S+)dJGW#TDQpWdwUGD-zQtke<;g!$!tnf9*u$ zZ=Au2r%V$xQbdSZZg)bO#22f^w7a^G*dIcTcGUz|iODW2L$GAa;GK4UxDO3n+29*3 zVoA!}yn0(I3b{O?_V^Fh-`9y2kt^`Q%SCB%S3kracOk|C?hC28M#`=>QC#>osQY3A zXkv_ByOid|BC%xa{o}bpD#+z6U=5r=3Waz+7ir6ai%(%~Z_5*0e`uKey=3M9!?3bb zzA9%YXy=`cT^asX74txs@E&BuI$Q>on(Vz&>(bPQ+)`J17E5PP?(jR%D3pl@OzWkT zIKRM%-F8j{o+Le3V%~)uIQaA2%tX}H6F(%#kNEPH0bcW(aMNl^ zn5kSea)g?A_;wB>;FRE><^V54)YKz2f-36PX>bQ*zkc#qS+7!p~D)!Qj;QUTDF8T$m)tfH+YIk^!26#eD>kIjrE(y zpHukVytTO$8%w6%Qn zn5I$i(5zn)0U>WC6!zD^gj`2pV6tqf=SJ(8YO%@@J|5q_{r*1aoOuqlPN*n7p(C*e zb?pdR)a`ek2T{Vp-U20U=7E7Um_P47do(~vpNBB6<%lb1I7t{*>Bf(n#wcjB($q>ZS^iwgR#9$NL28Y5;u{6J>jm zB&6H>T8cT4;g}o`UM39rtgn&5>)SR#FE%X&4+P+h+#4pG&t9Py^+e`X#c?XLW7 zCAJuCefIZljMq-suo2)P`rNbz@XfB(fHL=gw+p?$syzpl!%i99raUqOG6$fY#BH=! z{$}c8@KOY2HR$3r#yxt7h5*vyPmvW6Qcw^r{4GA(Pyc}czg_?kBseK^6hlxZsCmBJ zYJK`2?RX!F{gts3un-Kx@sO;bzwrtiWz+T^@g8r7^Kq~kPu8L0EU)zN6nQxsp(ukl z>*jwVG~mns^y7X*sTHF`>wRyGyWS2Wr$TSO(mq{X!6HPb$rV_QJf|OT3}^^H4|w|{o2y4f?U9}bqb~Kt#0-cQ3;|YUt{*+XWGGgLQ=s&M z_nPc5SQ_dYVl)ER@?g3lutodPq3)4ffZJJhm+N=6bzme*gF`|EC6t4n>c zwSlFnTEOFY^O6|L*k)4$H>a{Y&d9KxC83(C+xFe^qKKZb(^KNeg)o9OBI(J)$kW<* zU_x`lH7R4WSN%Z_0$mPNj^$T88>Fp@;GeHp!xIA#s31h;7$7_LeLzJi4>VF)xy)Xn zTxXavM93a@MsRZueP2I*-FPbgl6?Lf=njP!7{{?8QI3Ckv)8OnU>gs4D6yo7hZtN0 zxEpa$RW)Zx1T>1=1bknWTp#{n1mkN8S?3GAVf|)lC4YC&Z0JoQ#4v!m^X~pLTKeeo z?EA~WX9Sf4(=K5lW)LqGlAy7Ls;(hD!nNT4spQP#pO%0UIW^8jc7-aUTuIjm68&ul=_rcAj^RRaLGFa6TW~ z!`WN;)8oQyYo$JMU%Lh@^Auzq$K@XO{Rw3dn@QFyEtFb^GlH;5xCoU7#do6X`Lpx8 z=bU@h&05-86Xv|zAh&($5{||V4yc*Zz1EGaQiJrS?7x*7oQy(23qA=?Ix#4aN7~Nq zU8eWt+c##hkwunCt=O3!&cJ+q#e44{PVv!i-&JsTp;RPm|3hXXDy=ZhU>RUD8SvE5 zvXPKb7p2P9oQsRNNhwrdCO)zxD=09>*>n~}5(*B3*=>*$0kiv;s)G+LZzykhdF#Nk z8hkXD-1teX`3382B?g>bp`gI{dA0y#I+r_83Bn7@f0ZCyTAp}SGdcLJp?2&+zvBZL zI=MvFsdwVpzV|}x$0y#2HGUacBM-w&;M$drT`*TAr60I0wnMDmlwS<|Q?*--tf(SU zRKMPAJbruFklLTZDI~|$HL&&nG_vbUJjO^uEYG#J#GO&(vd%@^!~L_M=Q$9Rsa=2gxclm9@v~bcltf`_szoF60i1RhlOJH9uZ=~q+J*~8>KwDT{OFpRhDXy*FQul)IFOf!V zdr3q2PuMg28T(~W9pZ($>Twp9DKBP|o-TNC23e(j0iyNZh6gi|C4U)O^J^UJQAkNQ27`6%JX zmpl(D$TNSBKcv!Wq$7RsCY$YgxYCJuI=T`zGdpH5Z3~2~HHFJ1LYmsK@wBJ+U|hEg zTmG=(l$PP{JP{s((231d4GFk4)7~!}k@i(_l@g*x?6^hzEv(A6AAadW91 z%Ns?OnlACRx&Z%=2xCz1HF_oF1`P~!b{TQSfwooBa}k7qW5> zer-yI#9#(@ED7og(FPgSIE>BprRkqr8bJc;GE_3CtHt-gf8cP=&kcMACv4y!$=nx{ zN_a}uA+l=xpe%=Rn7i0(EEK$rxqi44LcrsQv?$0jVldfhg?w=?Z~S)_p1>v97TvM7 zWAJ<$;1=50&%K#c3QN^PR>1;z#26!!R+YC;sDvrtR`{6Bbf>i-hhaDvFK6d+_^G|92euV8Y&w|7ui+G~( z{UHcbn@1p`Oeno&7};NUxungaSsA{-|#C((wDEM>)22ZTJpCT6Jn476|W}@ zGTF0mJVh{#x<8Un*HBoN5MbV)8uX{}Z6kY{Or8*p9kfnyT6LHGMFddluKdlG=Et!!u*1wn75IcM^`uO$Iz(*6IT9{XB>+_26JXYF{etmL*~gI zr0>k%>Fg8NJow5yek){X>*OIgH)pT2{Qp#fE5%Vz;Cd!7K%DYhR?Ct#*sMP1 zBbCpIb2iU3IC0Th0+o((ly_d#YrY3kaq3Ux&VCBpoF0O$h!d#KvyoCx)!+;pkdK%; z8-QG+z$+VEn?}8)%6zTa*pok<&pPr$Z{DoFDKcchOT^@BbGRkR=!XR1`$FNSfwEROI86A>-TjWic>f|4!=6Okx2~_L$zxe^mDcl?1yd^cIXs7-PFKSCVMPL{|+%xjq1NVDNme|jn=kFe4&WkwQvjSv8aSa=@p?+9|tKQu|HJy5-#A0p7t@OKeZKC2~6YnpQ zId+SqqoAXsq(wRNJ@9LM zI6U-h64?eITy4QzUu}v3=2T0^4+D3~=@L*vph_5|t2!Jh7?lg}1wG-0KxqJcqzY4Y zdb(u@qFL}Lok03gOnxk=d*MF{Sp5fjuB7%FnbYp0Uf-+Og@Re$LrY+l$DpCLkI~^U}JzZLQ+)psCU=(5L3wDZDvTbR`L3 zoC<^$zqHKETD5C`>a9i-hPoyqWFbB4K5a5MiEXz<1bd1fZMpTY8{KwOpiHZ7l8FSqCp4H_Z#0F-eFi|T(RjKyC7|iTTA*}d( z8yL6*tSDl7t}{hSLc&n4{JZWSx(4j|p5zd2q32M8uQoYOZ^>|5U0FuRUW4MH)WY}P zw~-_esFCK|Mf5ZZJjsKNi5aFz>dNYRgW4;7+WfBvob26`#u37>q@)A^>cq0H_K*QV zYwqgmiW(5qxM3EhwQ!M|@>`9xL!O?Zt-1MQr?HqUdgkZn5vL1Z zv%~sgZ*>^rFXB5~Gw2lUmFtSS21Ix^O3NEwff&<-2vYjA&ex!)ccYOrWc!?Uk0y+d zD-Nq}>GESX0%y!n_q^T`bk5<%S2jf10)DN>o4%ud1sBNK+1V_0D6kD~BLpYlUWfl^ zIM8xy9XsOJNNC{l9ef8Ck;v13wmSn@qlk0a@rTR3m07~7b<$yP(;MYRZ&e4_rzY{C1_~M!X@MUY6`PX*@mFZ42rrE8Sjb*^}%&)|CY`20Cyxk00|3 zCC!wRg}6gZ504%^5ifs)c+jyDV+S;BB#|Y4a-=!IJ;5YAhY+v_@k3_D&4BPDq3*&` za{TfuqcGHmlQ=+Y0?18u6gX(KQ7{2CI81oj_rCVZ+|O4eurzv=&;r1N=?_P@h%4$< zo5Kur;;BUdXEdMzjYUk)NC~h3(p)>}OrqyfX;(&l{AWiytwG}Jito&&?9PVS``tCM z4j!SZdq+AbP+rp9fk)V)y*l4?rzWg5%$kf_P+VusuN_dXHQygoTfz1A1JJ*o0Ypw^ zcn)k@&Y9dNGJp2;9X*o<;<9oca%|vmNDF8V(Fs3gx#L2?L=KS@_8ux^8 zfTP;(sQ&WW*KKyJJMv2B zYhQ1yvs|+Y8p0&%v1YIkIfx_9mE9c4rQiYV&47+H!{Y?Q_6(LN`5j-=B4NpURap3b z7J+!JvQN8$3qnkXS;PS7Lg* zMPm<`PvEeb94NU{dgC6f#7%(c1Trwt{%#|@9>7@`Li8}Qh6q>)fV~r|OzqX9BAl2;M;Bhq@qrBO<*Z}iI(w9`V2ePj z&EnpKIerh22l5F(BxL~2xODh=wh?ws>JHE|IErxc%6tIE`UQk12NL6>r2xY7qiZpz zKzNCXxdEWtDI{E9sF4eJjHWXJK%#*>{0dXqim)&15%3X_lacqL42vRwryPKv@)0;3 z;6+&o?Z(6-f h-~Rb@|EHI-Ovp8z;!3Lda2~Z_tZ%CKQr9u){{Rvf{Eh$s literal 0 HcmV?d00001
+
+ + + + + + + + +
+ +
+
+
+
+ + + + +
+ + +
+ + + + + + +
+
+ + + + + + + + + +
+ + +
+ +
+ Contents +
+ +
+
+
+
+
+ +
+

Exercises week 34

+ +
+ +
+
+ +
+ + +
+

19. Exercises week 34

+

FYS-STK3155/4155

+

Date: August 21-25, 2023

+
+

19.1. Exercises

+

Here are three possible exercises for week 34

+
+
+

19.2. Exercise 1: Setting up various Python environments

+

The first exercise here is of a mere technical art. We want you to have

+
    +

  • git as a version control software and to establish a user account on a provider like GitHub. Other providers like GitLab etc are equally fine. You can also use the University of Oslo GitHub facilities.

    • +

  • Install various Python packages

    • +
+

We will make extensive use of Python as programming language and its +myriad of available libraries. You will find +IPython/Jupyter notebooks invaluable in your work. You can run R +codes in the Jupyter/IPython notebooks, with the immediate benefit of +visualizing your data. You can also use compiled languages like C++, +Rust, Fortran etc if you prefer. The focus in these lectures will be +on Python.

+

If you have Python installed (we recommend Python3) and you feel +pretty familiar with installing different packages, we recommend that +you install the following Python packages via pip as

+
    +

  1. pip install numpy scipy matplotlib ipython scikit-learn sympy pandas pillow

    2. +
+

For Tensorflow, we recommend following the instructions in the text of +Aurelien Geron, Hands‑On Machine Learning with Scikit‑Learn and TensorFlow, O’Reilly

+

We will come back to tensorflow later.

+

For Python3, replace pip with pip3.

+

For OSX users we recommend, after having installed Xcode, to +install brew. Brew allows for a seamless installation of additional +software via for example

+
    +

  1. brew install python3

    2. +
+

For Linux users, with its variety of distributions like for example the widely popular Ubuntu distribution, +you can use pip as well and simply install Python as

+
    +

  1. sudo apt-get install python3 (or python for Python2.7)

    2. +
+

If you don’t want to perform these operations separately and venture +into the hassle of exploring how to set up dependencies and paths, we +recommend two widely used distrubutions which set up all relevant +dependencies for Python, namely

+ +

which is an open source +distribution of the Python and R programming languages for large-scale +data processing, predictive analytics, and scientific computing, that +aims to simplify package management and deployment. Package versions +are managed by the package management system conda.

+ +

is a Python +distribution for scientific and analytic computing distribution and +analysis environment, available for free and under a commercial +license.

+

We recommend using Anaconda if you are not too familiar with setting paths in a terminal environment.

+
+
+

19.3. Exercise 2: making your own data and exploring scikit-learn

+

We will generate our own dataset for a function \(y(x)\) where \(x \in [0,1]\) and defined by random numbers computed with the uniform distribution. The function \(y\) is a quadratic polynomial in \(x\) with added stochastic noise according to the normal distribution \(\cal {N}(0,1)\). +The following simple Python instructions define our \(x\) and \(y\) values (with 100 data points).

+
+
+
x = np.random.rand(100,1)
+y = 2.0+5*x*x+0.1*np.random.randn(100,1)
+
+
+
+
+
---------------------------------------------------------------------------
+NameError                                 Traceback (most recent call last)
+Input In [1], in <cell line: 1>()
+----> 1 x = np.random.rand(100,1)
+      2 y = 2.0+5*x*x+0.1*np.random.randn(100,1)
+
+NameError: name 'np' is not defined
+
+
+
+
+
    +

  1. Write your own code (following the examples under the regression notes) for computing the parametrization of the data set fitting a second-order polynomial.

    2. +

  2. Use thereafter scikit-learn (see again the examples in the regression slides) and compare with your own code.

    3. +

  3. Using scikit-learn, compute also the mean square error, a risk metric corresponding to the expected value of the squared (quadratic) error defined as

    4. +
+
+\[ +MSE(\boldsymbol{y},\boldsymbol{\tilde{y}}) = \frac{1}{n} +\sum_{i=0}^{n-1}(y_i-\tilde{y}_i)^2, +\]
+

and the \(R^2\) score function. +If \(\tilde{\boldsymbol{y}}_i\) is the predicted value of the \(i-th\) sample and \(y_i\) is the corresponding true value, then the score \(R^2\) is defined as

+
+\[ +R^2(\boldsymbol{y}, \tilde{\boldsymbol{y}}) = 1 - \frac{\sum_{i=0}^{n - 1} (y_i - \tilde{y}_i)^2}{\sum_{i=0}^{n - 1} (y_i - \bar{y})^2}, +\]
+

where we have defined the mean value of \(\boldsymbol{y}\) as

+
+\[ +\bar{y} = \frac{1}{n} \sum_{i=0}^{n - 1} y_i. +\]
+

You can use the functionality included in scikit-learn. If you feel for it, you can use your own program and define functions which compute the above two functions. +Discuss the meaning of these results. Try also to vary the coefficient in front of the added stochastic noise term and discuss the quality of the fits.

+
+
+

19.4. Exercise 3: Split data in test and training data

+

In this exercise we want you to to compute the MSE for the training +data and the test data as function of the complexity of a polynomial, +that is the degree of a given polynomial.

+

The aim is to reproduce Figure 2.11 of Hastie et al.

+

Our data is defined by \(x\in [-3,3]\) with a total of for example \(n=100\) data points. You should try to vary the number of data points \(n\) in your analysis.

+
+
+
np.random.seed()
+n = 100
+# Make data set.
+x = np.linspace(-3, 3, n).reshape(-1, 1)
+y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)
+
+
+
+
+

where \(y\) is the function we want to fit with a given polynomial.

+

a) +Write a first code which sets up a design matrix \(X\) defined by a fifth-order polynomial and split your data set in training and test data.

+

b) +Write thereafter (using either scikit-learn or your matrix inversion code using for example numpy) +and perform an ordinary least squares fitting and compute the mean squared error for the training data and the test data. These calculations should apply to a model given by a fifth-order polynomial.

+

c) +Add now a model which allows you to make polynomials up to degree \(15\). Perform a standard OLS fitting of the training data and compute the MSE for the training and test data and plot both test and training data MSE as functions of the polynomial degree. Compare what you see with Figure 2.11 of Hastie et al. Comment your results. For which polynomial degree do you find an optimal MSE (smallest value)?

+
+
+ + + + +
+ + + +
+ + +
+

previous

+

17. Recurrent neural networks: Overarching view

+
+ + +
+

next

+

20. Week 34: Introduction to the course, Logistics and Practicalities

+
+ + +
+ +
+
+ +
+ + +
+