From 998e88b58a58dc99c86d7ba0734a710e3142b619 Mon Sep 17 00:00:00 2001 From: mhjensen Date: Mon, 28 May 2018 17:38:49 -0400 Subject: [PATCH] Updated the introduction --- .../How2ReadData/html/How2ReadData-bs.html | 4 +- .../html/How2ReadData-reveal.html | 4 +- .../html/How2ReadData-solarized.html | 4 +- doc/pub/How2ReadData/html/How2ReadData.html | 4 +- doc/pub/How2ReadData/ipynb/How2ReadData.ipynb | 4 +- .../ipynb/ipynb-How2ReadData-src.tar.gz | Bin 113365 -> 113365 bytes .../pdf/How2ReadData-beamer-handouts2x3.pdf | Bin 413838 -> 413826 bytes .../How2ReadData/pdf/How2ReadData-beamer.pdf | Bin 389647 -> 389631 bytes .../How2ReadData/pdf/How2ReadData-minted.pdf | Bin 411715 -> 411726 bytes .../Introduction/html/Introduction-bs.html | 312 +++++++++++----- .../html/Introduction-reveal.html | 334 +++++++++++++----- .../html/Introduction-solarized.html | 303 +++++++++++----- doc/pub/Introduction/html/Introduction.html | 303 +++++++++++----- doc/pub/Introduction/ipynb/Introduction.ipynb | 274 ++++++++++---- .../ipynb/ipynb-Introduction-src.tar.gz | Bin 212 -> 211 bytes .../Introduction/pdf/Introduction-minted.pdf | Bin 203760 -> 213987 bytes doc/src/How2ReadData/How2ReadData.do.txt | 4 +- doc/src/Introduction/Introduction.do.txt | 276 +++++++++++---- doc/src/Introduction/make.sh | 23 -- 19 files changed, 1347 insertions(+), 502 deletions(-) diff --git a/doc/pub/How2ReadData/html/How2ReadData-bs.html b/doc/pub/How2ReadData/html/How2ReadData-bs.html index 91621c57d..3c6bc2386 100644 --- a/doc/pub/How2ReadData/html/How2ReadData-bs.html +++ b/doc/pub/How2ReadData/html/How2ReadData-bs.html @@ -447,7 +447,7 @@ factor in front of \( x \) and the normal distribution. Try to change the function \( y \) to $$ -y = 10x+0.01N(0,1), +y = 10x+0.01 \times N(0,1), $$

@@ -459,7 +459,7 @@ 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 succeed or +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 \( \chi^2 \) diff --git a/doc/pub/How2ReadData/html/How2ReadData-reveal.html b/doc/pub/How2ReadData/html/How2ReadData-reveal.html index 7c11d1c34..1aa9ac1b8 100644 --- a/doc/pub/How2ReadData/html/How2ReadData-reveal.html +++ b/doc/pub/How2ReadData/html/How2ReadData-reveal.html @@ -415,7 +415,7 @@ function \( y \) to

 
$$ -y = 10x+0.01N(0,1), +y = 10x+0.01 \times N(0,1), $$

 
@@ -428,7 +428,7 @@ 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 succeed or +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 \( \chi^2 \) diff --git a/doc/pub/How2ReadData/html/How2ReadData-solarized.html b/doc/pub/How2ReadData/html/How2ReadData-solarized.html index b17a9732e..8814a5e31 100644 --- a/doc/pub/How2ReadData/html/How2ReadData-solarized.html +++ b/doc/pub/How2ReadData/html/How2ReadData-solarized.html @@ -404,7 +404,7 @@ factor in front of \( x \) and the normal distribution. Try to change the function \( y \) to $$ -y = 10x+0.01N(0,1), +y = 10x+0.01 \times N(0,1), $$

@@ -416,7 +416,7 @@ 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 succeed or +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 \( \chi^2 \) diff --git a/doc/pub/How2ReadData/html/How2ReadData.html b/doc/pub/How2ReadData/html/How2ReadData.html index 80d7429af..6a37184b9 100644 --- a/doc/pub/How2ReadData/html/How2ReadData.html +++ b/doc/pub/How2ReadData/html/How2ReadData.html @@ -409,7 +409,7 @@ factor in front of \( x \) and the normal distribution. Try to change the function \( y \) to $$ -y = 10x+0.01N(0,1), +y = 10x+0.01 \times N(0,1), $$

@@ -421,7 +421,7 @@ 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 succeed or +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 \( \chi^2 \) diff --git a/doc/pub/How2ReadData/ipynb/How2ReadData.ipynb b/doc/pub/How2ReadData/ipynb/How2ReadData.ipynb index 48fa4fe11..d10084e44 100644 --- a/doc/pub/How2ReadData/ipynb/How2ReadData.ipynb +++ b/doc/pub/How2ReadData/ipynb/How2ReadData.ipynb @@ -268,7 +268,7 @@ "metadata": {}, "source": [ "$$\n", - "y = 10x+0.01N(0,1),\n", + "y = 10x+0.01 \\times N(0,1),\n", "$$" ] }, @@ -283,7 +283,7 @@ "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 succeed or\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", diff --git a/doc/pub/How2ReadData/ipynb/ipynb-How2ReadData-src.tar.gz b/doc/pub/How2ReadData/ipynb/ipynb-How2ReadData-src.tar.gz index 532ba34d7e2c85a21b685cb42b168d9969f4997b..950467d983551374ff39cfe9f0a0ca5f8cd205ca 100644 GIT binary patch delta 21 dcmccmmF?`c?opfy5wmNn?$%@VH*h$}g?l}A0y`SzwjWPeK zZ_ZhdHAemMh4|kJ@f(7{VRSj1bR}F<8r;d7E5McD{?=%RG-{vrClq1k@#d3zaP8GI#0t_FU5z64MTWTki^925^u2G|Ku*{Xt44N!&9l=6 zzLc5gbAr-)Pw6bW^5Pl_-<-PYdJ1>H!AA4&`tD~5f|-C7z{!jx^$7s*zXoDiiOZ0o z!23(9?n5_%8SZ?q?sYtMTD?>ca+pYE$sr}==kUq@>DziC*Vvy`%VX_~>pN%cDXEOZ zp4nI#hwXeCfSeoDn2U?}m_o>Y5#PHF^x9rwLRz@*)|@r<0%>3yo(bmb*N~r-Q@EfC zy_1?Dm%A7^w1U*u%fUN zkLey;(qV{fcgB<7VNj9K*>j5ZExNtXhFa2#uH0qVlufZM7=NH6^kGNU&gqs zFJErKA(qH;gBDdnN1;3psbKs|VIy33se|lD3};P)w5P%2e(Xb0Y`P)Q{gkY2fNrGT z8`0s8?gdPaRP{XtUh47s)8~oxDXT&#_Zz||SH86|JwrfgJ?-naN(lKlAfb%JiA^X4 zxdeM3`@^F=iQ(cY&`^M!Rb0ny zM0VB7@RW@1bqahA!Z-b=`EIYE_Eg&Q45{*VD||AD10`cdDLn6R}Y!%Y3n`eP#U`tY@5_%bj* zi%>}J>cc8y=bwaeJi5;r7ab@V8RY!A@u^$njh&R>E_h_Ikqt8e2$|4PCC2mWTOybTX-%RCB!dG#0e7-4g-xEkreE*+V^|0y z{SOarS}q+mzMN=;m7yh9$xwaNe7568FZeX9``>h~FQE8a_A|baQrZom)Yv7G)`y41@MTPUfyzV|zb&<{ppx9Gp@bpD)c^C_PkZkI>6i;x~=gipY#b&#cIr@fL&N#LZdqke5E! zLA|G*kHHzfeSpAWuv~=DXmDd70c4jZ4YIma=RRr|Y?F@E&bJ4qxzvG~$HB9_RYf77 zBG+eWN}g&gczqG9duo~r_HC6tRH2(TwT<`CYs{nC`$ts{;>k}S;o5f3Z>G?h%(vRk zgWvRej~H--WebNFMYKp8rHNAbbneV%tz3(ML|SP;Qj8G?3t|d+FUls|HBs9;92Bs- z2u&=+&S*2FjyJK73Wa*OYhOyWwDPRs&2hapwvbL^T^XUSA9X=ecN-zudqMTIOVfLP*Xy##OuK)?lP{dvv%6S8d#7=P3`{q~zH#Ql z+MgBlih-W=KG`_7@>dV0uS(T73qUJLwfZ93c+{2#b8YRR4=wx^q(xNJ02R%XUDYTM zfKeXx_VFuHkvZzAa4VC)C7l^S)1Pw=m<=F@Zg;V~h7AXni|yJNhBc;TOf}(fubKLX zc2Hk~TNB)jMnL*CaXN?Oc~KU9d0d<}p-H_fh%DwWMe?B)vYd~=WeywM4<`Eo$@M3^ zN$3@_{4uJRDTc_dj#`RS_^dx3m-5Vr&28qFIdoIX)nj1e_I<9;_~-d0`nH_!xBYo} zc6m`1rok@YC|O3;fIxb+QD*$xR;qv9^RuEUg)>4_?jD=OQ{>FKq|jRgTKEh6JZz3G z#G|BCC`Zbo1azF}jm$d1=Ybv0p3Up1-%^Xi2|5aR6sUFQhAA?rBs* zD1Y-gRFT9!W%$<_;Tc^RzM=W+{iHFedcY7|UQHCxN5h+zCGrzM)$uuR%Tqo#(VlBy z21?B|1~(^v-vnLKN&BH;6>$_aRKCK71m}N#;aMhO4}2oHXPR%EUoJMvd|c_$#aa>H z+XbM1B6H1cILk;xIOVDiCj023*&q>FAysPZ|Cz^+Q(le*dMMJUTvV8r<#9pLxl)ap zkrx18cU8ZSf24{=Dz6b7zn1KP`f;NB!`Qp;1_FjqZy9`^huermpv$DaDATwX^4MgZdd4`|A37MD#d{@aeoaI3rCNno>%)dzU%M148?%HN1 zC_Zpml-6%{-;Is>0f>-BeME!7TT=8n*5(9*(DCUk5hN@`MSXULW-{Qf^R^$t_iHu^_g4U8dSDU+Ro__GCSDlJx|LNQ-5_jvsb?wmuhgc|+;n_%9 zDcgg~R*V(pgk_xBE$?=_Ur%A=)8B35g1@n09QQ62FJoUBF2H+#8ZIErsT$RVEnOm_ zADiR<0nNty=<{&haNmbb>06_)K#IQ3x<;_VMXoxfG8}0G4NvYg>CMy`f~bw|0N(ze zJ$mea0@mwSbFZ9Oe%hnyc18|%F!Jxqzw5d)L(D&H#S9S5{IN2gQb&!CRUjB!U0x0u zuMrAzmlcThuoZ&j59qWoQt^}f{yef_)ol>P1MNRpD0OLiY{BB2gA`xR12PiYx;<%q zwlMqglOJ~knDs^Hk>Ogx*E){uhu$;!*+=zWtLI350)%|naLfz2S)Zzqhx5R;Pr;I3_U`kjt>sX z;0pQ=9hp)We%vIn(LIOaVI3wwC6??>h|;|cU7*5h054G{R|Dqy_&0XI8W0|VDl*gE zM0c)k3P_LO_Fzb;kx{%kv1C3Gd4omJqIYPRBnu2fXu_89_d|V{D-sf;B(_G`8Np=|&FHLqC~R0!83h>~rMckHzGS8bT~uTNT7_%=AJ@H~PGAu?a_8Y^C*~C(FPI*Ttsq ziVrKLxg5Ml^_a1^o+hRe#voY;&CNzc`wPteIpjZIMTW7p(Z&$w%<2uAa>_0@>6`hxa|(g$j@ z@)?Z}XQV?e_ke*C=v{zNLxNDG4puvq1LC>49=|JC<*y=^;jx6zRY`x{-pQbradb{NLSXZAWD^GMS`Z3J;J-7>y8N&LwEY`P!SZNaHvAO$uXX@OEE1VqB6 zp#r{)$^)CeI{y(v`bCzvxn%vbAZp&{)b93)7HD2MovuecE>@T!5wN(CG?U>Qq=2GN zC4ksay1`-L-B>p9qU0jUFq>yD73@6bNx?!ERa$_DVE6(gW)y2EigXh&6T-s}*YQX4 zugupYczBKuXt%;2Xp_|n3;@{90Nb!h5;5`7cj6usAxeo8D`*QyNH<)Ulcy^ovP8^k zWKUaRSTptPF^j_nsOs(j@1?#lKSoSq)B*<}F}6`b_wTHN3_r!p_`JX@q!<{#uET{_ z=YKD^n2i-AB6owM!ZxcXNjPP;8c-eV!Kb@jZ3w3Tnl1iFB5c$($3TOL214J<>Gx;o z>ac$aDO2MPcX2Gm{RPTh$^&;pRDC9!R(vw6!^fUZX$+2f)?!`MQ-(z80g}htV_KlDGhl^ty8(W_v25JB~(?>MNY7Zaj;hhGmxb2 zubDGqBRVgKllJ@B^<9>eQC+;5DaCwlrgjbL!lxx5u&0|J8*Qvtj|&Zqv7Z1 z0Ns_vKtBT_xe9=aVgeztZnwj7`AeSAZy712%A?q5>^T1G7x%8IAUsI>5j9LEH@_HD zCsXTnA=5Afq_aC!2k>Y;E<@GqlCPYM$vdF>?^Z*QCkyyI>nl3a~(MWox;)t%ceo_>NqgC-UO~9G;Ji}StWH74UfmfrB zv>NOYJU-a)ZIZO~FCWyY#ujWgxrK_vfks+c+msEF%inixrUbx1<+GWYloYD*aGs2v z>;ZAUD#6unE=@*irjqZeFb~=CW)KoUG7cKL9U3>UXE3AkQMJii9{_nUz6p}=}Jf~ zE6&M3giMxW&~gt(=DA#pEKjvVpfD}lc-Snj5?|U}65HK8R3#IG+AqeYhA%ZiOr+Sg zqem9$;`O$+Gqzi-r@JRzU}(m^w3V53KO4lIx@3|s)`4l&lge(qX@<}I9&A&A_TFF+ z-CRkf(I%-|^QLxM8~E#r3vGU0M%$e!qJvaQti7r?fwt(B@!nC+YZv4*kT95NQH;2J zARp(0@z?W$JCiA1dsgJ^2AUn zthbSjCeopWGue{iE9kLD7yzsQJ!PMWnw{ypkWV=KWiYwWFdmg3XHR}&tAxbW_-(p; zF2(OjJ8uz}tBF@tY&MI@^~4`U_eQIR#)pU`XFw0P@L5*DQ}Og!@n(-KMnE&FF5fnG ze_#+umk(Oup2XrTu_ykoT*i;ZsGu{ZTyx4Hc=+d2it1Y75-34AIW!k+HPB1$B!XXt zyWzH9(G*wAn~*RgbAKT(F)G-x2Y2KxE=b5LbVpd^H3^%4erKm~PF-INk6P zei;wf#h+&`ZoqxTpQCnH*|nsq>eVJbU18V40&UWAzDefsXQKno9dCQ=qdR(Xx;6)v zBRsJ?FGG^wBj5B$h;V{X> zTF)^GobLeU+6E*Hc$C-89Fg zy9F9pO+r2Q#Yl7JHtiGeK~l3&w($u40h4Es+)N6*ZWSu(zCdTB(AB!pzt2#U`uu$M zDn@1ANcjazvLl%t;IeSeyb%HSps2svI|c8nc0{8094yB;^nH*!A2%DA7nrP_t@+tF z6(QDPXhIJnB4G3U43&qoeDPbYu?pu z85uXc?}L-*CshM~Rr-u+J>L)Ix@0>F;rc{|#td%HLafOZh8a(^zhFGpgd4?$cEQhcqC zh1n9TW}=lPOHKdn_BpAC_??U^vJ^3umJGZ8}pLOZ@sl``J-|4%S$>--zsT+M6OFbr!p}rqiUbGyQu=_SCB@fQLDuz1z|^K z)x-EOt3rR+6OhAK*-Ob@3Q=gVNiio;5+ zalp>8p4v*SgO81730z`?x$lAUhkacvVb0g4NTqYiD(}Ij1X!?3^?=qaZ&`)rb7>k` zx1r62%p}%#)|xfkYN+5OX?_fr5egizAusQ}K-zjYzl76B;xG|mJv$d8%&Aasog!`L zU}4AHFw91ji2XybVG}~IB5|(_f#Z~EIYJx}Dy%I8&XqkLI7L?9{m-F9qxy#t| zv698G<`~2aA+9IC1}l!P;y6u5zLz4KNpFw|aXDVAxJv<=mlfMfB22T0>5$VCyXt_U znk9|OD-G^;A(ev|hlAZ8ql5EoD}zOpRUy>U8n9FGRVX;5+Qdw}8pKTN`ov5{(DAbP zpnGx82J@;AHi`M>%_}}ms#c#eL2fnem@(;wEnAHm#j2-n4v4Y1FhN<*(qQ3A=3?udM(vl$0U}K zB1-^u{i9FHP6s1v%V>}8>e1ZjP#tEqHs(MRE2Q%G3@+hrH$$y12CfQwJNoZ*MTHIt zs^Fqgg!mBnZ(9ET*eiS+V)KQC>Zin51Lw)UqtbUIsj+2%mXo~Jxl0?(W9>Fi21TUnlcDm-kUK;5wOOCG>X9cVg*Q%0uXQ$PL ziRoImNvt!hY1=_^B88&da=0#c(UiU`?nH@j=}9!QECO#p)q$P!)OcK7%Q~uV$Pf-C zC_@4k@lC!~W;rijeC`P|B|B&jC?Mz)m4eq&4@H|5G_-8@vW_ZUy@FbSi0_X->5*%x z@}M@X!=slaHn*53OFXBuZ+*sGy=N^R=3{&SneNZddM5fu?#R_c59v)N`yf(dLT;NK zsZ$h5N&;iu!6Jbgg!QDF-rdgCb!&EkXO5|pSo8@)76{U&Diw8eTxf3wKyftKowO(XWTt{`QM ztzGNSc@t@7Hc~vvgp@QcldogC3pI3h9tVB?Vs8yg37P7DMefoo{#Z!#|!XF`5gsf9a3=9bUigo9$X zd$eIfX#uQ4K_v{d-T`FMDhr$XgiuZ>@~X@#l)!#Bs}cJA5uIx6z45V#R`=8wzrc%w zi}#0pws4pFM7&C?42!Rbm-}KG;sIJbRg|;J`VHxE4yu}$ zzJ@aDiCB?b&J%}Q4MsYf4g4DAqSp9|wHGB-y_q{8iHt5EB|{l zhNSZVisctU%lU+M;Ct=l-9h#0@0|PI&8ed7t|>3(7}(hF&TSzRj*#PpoY(UT1{zf> zk+a1QHjO@CW`xi+k!I4OW1kr}7{P_0QrEP5UBcOQAbesVm;pnep$It`S+QH z<^9&hK1o#}D3ekgqt8b=>Z`q5dVU7+1vG-XVz6EUrrJFVfP$~&=!Su8yPw}l(7&g z{Jd<_p{qC?Fu8pq^_%*4HMvX-f)lwCc~z0h1=LErKoo7YkfcWi+vh%RJpXPl7I^X> zo8(ab8$XjZ(8sGXHy}E&5h)8?_%YQb8q|b5mveq>Z7SV$fBb5|^%s7Nj(6LZ2zl&E zw-<~z^fe+IRy9fJ5lNJe5btC{?Az0VV4C?kXK~ zR7hXpp)VB{A+0f#Tgv0zXnS3ob%pU>9WT+=Ig>qn4)%KnskQVf;>1>4N@pZ1Tr`WZJ2IHAsNs~VV24aUqPvs2Dh zd4O{@Pv@pIHJ8AnD!lS;UNC5<_B6w9`6F|4wk9yb&x)@6=P-rh~%| z6u*fkE(L^)$i!?l6EH1DBx=5V2f1q-_EhzlE=>6Sw!`kDxbe_r*5)9Q1Z_vR?LQcV zCSd!vXFtp+gj-a6voPFo!oQ5QUkj|2@Bn%%&hh7HFn}*dHbMUS9==HKt9Tmz zNgedplHS7=CBUe4#^~#1(*CUvGD_iT1VDYR`bvxj>d+p)jLm!1BLtHPH|)}negcSu z_NtfG?VcWT$(@Gt=i3a8l~#A1o@wejFl*!q+3z}0C9pZF^ho#6uv&N6H@d&mqaNLX zcrmp53Wp*JPpwtcgV!ZA%&11}pl)X7X0gzu+cB;{bnh&O$J5ZpK&}LPAW6gbXJeUb ze7qNuE{TqrGmcpqercY@)f@x&i3w0ILCwOs8PLu$&bzYOhc*r88oFcHw~@u&kQOZk ze}+)yG>OdDB)$~mh7K>69)oq5ACXh4T&M)BgrnzgDg%#=YvYu+}4njOo z7eW+i%-8o;M7iGq@KAhaDe~!Q)Hb8&8!Q-<>zs9d87u~-rOuKizdJ|^Dgrw3!K7JJ z%hWe3ljR5nwt0}f$-Zo&^YLdBl%ey?$)>dd&)yTMuqH@)&Lk9LDwb^7xl_7D{gGHGCr^%a7%wzvyVt(3!@B<|n`?&X{&oFr68N`n{<;MDKlVS) zoT?3zznwYFXovTLf4BQrt(}nqeEEO<|0h914{q|WL@f*W*T0`m`46E10H;NTMurh* z`fUFy4UUWrbtkITpOIt$Zu{St+YG@y{@Fo$wgvbD#6Q1nw+4^=7h&ZLKJ~v}{;~_T zYx{t^5&wPVuSCS}H1NLy;;(@GE1=qcr$Lug!Ladg@UU@bWQxEL|M%S9BLcIIs3yfL z%EiSY!7jlm#tM*RW#i#tmEz!*5@TiKX6KP)wCF)F$&h>sX>EqcTgeB9hE5UKLI> z3898vLYFO@mMD*(ped54zui$b?HufEj{mZ`D91oHLxr7el^8AaQevitSs99zk|Bnl zx^zLsC9+X#mVI1dK;*vc>d9?Ou&nNtXjZz^Dvs=GOo_W8t60KR32pWT-*3z5f|m34 z^!pmHM|IPPxvdVXeB3Joq&qAV}4M6$$e(M+uu^uywEgJBX-B;G)#0&UBa*2!D``5k|)NXu9 z8n!bUe6?>v{{DqT7cM_iyn58Ud-ULU^l*Q4RQpJh=Dtn7j~VVedHhP`0TS>VBCRkG zLpcv!xv{F8wu>*7t9@&06;_G+;Ebm51 zHW;#wp(iDK%t>8jx2H$_EC>nFMa(`xT`bh-WD-?Bo_Ctb$7qBPQQsp7)mCv0VMe>Q zXg$cQ%AXH|@@83HGZ)5~$Obl>lvz%gGub0ZR#Vq&bScgY-e(8*zW^*=0Zj3+qdPfw zPCVUDy=az1_dXx`n;Eq+9*n)uNZkIv28udZK3_3@Ve)e=Py8OaIDgi5-2aB7mMTMv z(`!G{pq1DmEo{F3%)h^ay}!wFJn`c9aVSchb;!68cZeG??T#QN;T5u$j!{mEmXDoX vzR?zGO5D7h?>vU!C*U3BMd@8J{0R%0Ue=XisRBa=&&Gj3O)aS`h46m>pDmNp delta 10442 zcmaiaV{j$P8tufkZQGcb6Wg|}33qJo*qYe3ZF6F4;vLT9&3*Omt#f|8A753k?zOtV z?y6SR3NFNpFT|@40)yt&vEwx71&UUa0p-q7f%SuhO*-JYF#oL|&?jvogy)C*C-%c= zgZ(GgAryl8C+;EB`xh3XvHTZ*lBEL-oP!s*4?_;P7PJ9wttRb0F=+3U3~wN!bw_Mf zET<7p#jCY4xO87`CtF~J4=2`(1VZ=YEmRz}IrY8wLnM%Kr4edu;1lnCee-y}?D{?7 zdY~i(IM4&2eMsIu)4UAlL$ZMuNwfAKePPFdtrmK2V*xtcH*b)Tb++}Y0VYTY2PK5n)Wf6brk62L_>S~fje-o#u9fh z(H1iibG|$ER7ziK2_{>rp@Z{^6<@~WoexmlGTp#7@CqxnP-{w9omoMoS%UJ@RKF^A z&|00UO5go+{G3mvosUd$6vi9ov5?v6{n~kDg(Z)+XQKzq7x4a5(;!D^X_!OSyOwyp zB81w>k)oJ*V<99UUn)CMed`uTK4{;2Q|ru6^)+OpQ;6nfGp}26>AlgYL5pA=Spcw7 zd#^AiWYfyzyPz`eE=XV^{=DsO;4MA?g@WnidOytN`=$Sj2wGw!g}=8m`F`7m zLUd4k(J+E3SZ6Prmhaa5cJ)Tt$9giI5w9?}r{%3s-Gl8~swsd=&uMmt`981jB07)J zoZU=~fMI9;6`%jN=~ou)D;gg2RU&|I&vkc|4sOx8S5?l?Rci%(=c2Dro3{9@lH?w- z-{YD=A38$bL)44bb5#)*%SL9hg=>bCeNfd$yg;yt_{ehr>Z*Ixdj5{%rMHrZs0pDt ze{9qAlrm*tw;Pf83Dq|N0MXY-b{g%24D{EEQ)xgD?tPry z?E1qo1Z1fw|L>AQ*ciC8aT!FAEDG{rzud>NKxfX-M{NcKe$oh>MEfzWa&3&R34*EI zU{r%Wrmss%l!pJ!`+AEjNL3)rFTV&fFQ5yZQtjD6=N1aK`qxOCZIs{bIxb-(L0EF` z#~%32^fNl)uP|oT4CP=g5g;IlMW!3~7aD}$)O~Gv#QM(3Q_=pTmeU$f4(sb|nxsLq zH_Nz01A(ILR%f>bfk6z%*%wwGy`T#2F=Yx^SblDj5#neLyN5zynQq3r4GhR8OV~up zBid{Jt969HKP_eT54?M=YTw@mp)UHx(Xf?X{3_?Q8*#`x4usoBVS50a>bS?ck!+)K zJfc)CCC)teYjGvygK$o`PP?Gssbrvy<9&7`z3=^z<57y`c*HyR@QDdr+E7vPLR+vyVW$? zjm!b}h(8W86y=cf?E_$`M`t^c?oH=dJG)MKisJ0*ZXeQmB>urIKjP3%`%Z&Z88|Fa!&<6Iy81V60?J{>XyYyrSwL`*;w-k$`|8tQlu zdOI+qvA1N=ujljq=rE2T-cLe}{Q?YSkN+aW0B0g0anGTqxy15q_&s^_Gi8ZV1fLgoal_!zaChj+FVh(@d9^uJFa$-#eiFyqeUGvjUa^yRCny92JZUQK z+_(W5r+3TTXDYB;MZ(?!J6B}~W<0jlT5-77cs0JQlKvC0=)n0y-rJ(0Z)*m6TgVe(=TDYf zILNgnq)Qt)%f>S2uyUpei(Ad(xnc-C+xl%n;RtwU9#DJVB{G(8dtYoAG|8BO+B`%6 zYU-fs7>Y#0Mp_OiQXsSFaj36|Z>+WQD13G&Do<+bH@!@DT$VsqxS%`^7{=7SU!A!S zC=3Phv>bG4i~={E*Se|>zci_G)63>*DFbeFus~5@!|f_2X7#J6Lb*_dr!)h^9gw`G z=&PS+!{|qnvJpdy9ojgkxvp~XU((G0AG0-W=UFprzeG(rF$?W&G`Hq5Bw|tEhtJ7^ za8bfo-SEZJ+-5n%@o3d;GkjB#GOJCYH$@#>auE%+L&;}5Y49|lryJi@IcRJnvandRfRp2!9V;Qv6|DgahS7aSsct5K4 zG@_tACoRl_#e!stm4{Q6O{Hpymghq&E&u$k+r`k2cJg0~BT8Z-c4{GA>$K8+!^rdvO) zWqttpl46)K2nBa3l1Yeg9!TX)@Yqbyv1!UOgJ>znhZ4=(sWuttb)W|A|9u}FmysA2 z$GhNUS9w($u$Lm#kZzGAKHXA_Hlz4fcUYI?e#7ND^U(mlG2!{%tL^lr*lh;z0KqyP z`v)J)%dyLeC^G^E?oCn4=vfnvE43;Q12~R}aY&vqsTbA0Jt8KMbxONyvfOV7hq6v9 z1q$s=qtCOjyYe;ZWknXS!4lKy4FU9L$O#qtw4Jijt_23kNL}=}hh61*DXyTk6|9ug zZ(e&dCQpF``SlZK4zBBh{uD?*Sc~oC^E*GD*hXk#_a*>Yk&+%>_rYfg;=z|Fxt?xs zIeuFcRILh`uUM?ugL6GI@vbE>qGbFK@Y$6BR@|5KU02C#vL&xN(OGaj1O-9cR2!q3 zY)KVCOyqcs{axj!>fhlK?q{2LJ=*9j2;gx+L$PE%sV~CrQim8^ha?6-xMhCZj0!6x zFfqcY25kUrxzKk^OBZ1aR8x%B@D8E#XHZ?(o>SrUF9Fv4Tk(%T62;G@ol!dL0Zjeh zYmhxWR{6(J^MwJYw?S^I1;1Oh4Ff@9qG!pI9)OUrFq%33&#N8%4>b%saijKXkCR(= z)RG%Qp-@qg<`b356a#=q?wr<-NQH`ax$YB}U!V`~7IV;RqfyP*t)GUqYt9zOmHD-L)41-8)qcJN2RE>>6Z;K|hy&3k@;CN1AX8&anE*mXi;C zR&i#uew7&)xTOGq*uUy)ikd9L*o0lf5XXmvBr*GdLZ=4tzYl-9-=Sh2B@8E&44jM7 zz7JlagiHHf#ZPE5E0V-9J@v#N;YxKvzEc={qoo%OKBO=RKCF;-h7@FxL~u9>q7YEj zfW!%yoIVV@K3~QKj1b!xW-|-55B`bSt$iw@#a=VTpX>wBVnJu(m@r5|ELUFVbU}je z6N7;}(*KFxDSFd_MhSe@v?ui=&jsFQd2{WDbtBtk&=9fSYu=J2W7Hcl z7$rBB9$EK*mxT*S#=5ZP7wal|){l8}8!{+og@q+r z=wm1WOv%;31vLY`xvGg{Pq2uSty&DURX=D!JqR=&^C1>vH2wPhvfMezFLq5l@;1(w zUW!J<$nKsDj+6;7to1BNO6jq-WKtEu76oDMulnVHAg|T?eIVa>oTS_m*dCD_d(1E! zYv9mne?nx+NN@KVHF~Dge)HDsspS6S*5w@q`X+eXGMdbC>M#$!;yvIfz^PRX9K8@G z2g&7XyH(zq0Y-)~l%)Ltq2x+VQW|vD&1g#+VrYx8Qlx|6-V<`uT7$>lP)r6ZYl|k~ z8Au8+DC!B<^*fw|b;Hruq6C4b*oqoNanNQ(s*gcRN=qp>(@p4LL=?mX^WqK$o4F4^lJ(mfsp=k3 zvChi-PBS%_u!N9M=iQh z)XB@K0f2j+v^@JLbyr4AADLmA8X`QO%+s->dP(%AFM5@N;+{Bm6&bx4`bp8bIHq|{ zqqgKMq!UG?3J72?03rr(4U%@6eJ_b}wnpd%5H3hnxMle04(rvifj!UxDF6E6O#XNO zeF~G#v5|vtugRcfzm6*_(F+RcAC^;PszGtm4Qm4vZIO)G&o`;15Q1c?Ph4K)++Y%; z1~fwnZ!z;juZFku4VK%?own*CC)*+v6H_g{JO_FZkZlYc?f~Z+6!CVM z))aX#BBZQz;~%OK6lKzB(B;v!ZqQ!-P>cJWAKvZYY?Q=Fn(g+&G#Tgh+gYW0%<-m$ zr1QBQ!(&pGJy_&MMb$X29zW{g>Lwc9)G$a&I5G)TRU4z}730cHniODtvYCjiUS>`pPStqxB^hBy$Z%|OwQ%HQ} z-qrQS^&hgPC&flb%Nw- z2KbIgJtTb7mTgVU>JBx01VO*jc}cIXb)rUdI)9M=z=uIiZb$ zEzo>{@^b0NaFvsy&yHTGD(P6`#z_UU4?}!=&~&Q=X%PL7%V88%MkPJy*})uWTIM70 zM7sLvr3RE@hSj!9jDW42>kW%6Dj7jlo0aE5qzv681W_+-E4s8A=^wgsW+=xvI*6)P z)@suE>8aGuu#Ds$6ug`42>qZ6w-Z6js8kc1=4!NB2{2EXFT`l_$>)F^QB0V7b$P^( zov|}1U>G6<6zal~T@S2!(*}E)x*nfa>-vs!kc)}swGb3FKj1MBi5kdApB_JYS-3{; z^)oCy6J;xn_}bOHRyp-oohxYnq#*3ea!*OT*rrN&5%FRVSg~g486ibDa})-hxYsk9 zd0 zQC&*6>5yq`6bcjO-rFjs21{fi&L~3Y!6`1mfymLIUH~EL1j|-t+$kdGwqLz0=$Vd> zrPC)MzKF&{7NVM);Q|sCC`9Uyr`dfn`e)Gh=qP2-MuZxETtzs$`$rYPZh-OnnA3n~v=;SR+SBVCeBzL6H?nR!n+<&xh_+IsCBYyu~WEA_`Y8_AOR z&U&i5dI94&0zT(Q)!8PKa}m?o5tcigTF#av>`vVbzkCC>GI^lnABoL(67}OyGI`R+ zyQ2cmr?bt9yW!wo&d953giBrpWV2A5u~h3`b0;6Y(;u{UeC{W?VvdD`mzeqrz41}L zoVasG{>0f5@(kS--UCrEcoy{anN3{KSlx1x_yfL0;pG!4^gJw??m+^6*sys*L%nHG zj`$@6*OQ)86f|=^RRI#8Nr&n-pM*mjpXLC6PP8((9*{POI?jH5-$7*CXufgInl76x zD=xaS56Mueur|{Ew5IA1<^9vL59>y|6aHy1Vt%4Y48d;u=DhUKXdQ9FQ9W%%(eZ`8 zj2y7wO1~(o7}msDSI+@51B-49)wa^1>vztn5iWs9?2{Mzb+A zB@SK6a<57^TwT~l zxSX(bhEHV+mQ|~^HI;ajf}WoMP9a1D0Ru}TV66rb2Kz;6dwd5L;Cb{vTt~0-t@2sn?7I({`gM0fxYrES9;g6=EY@S|sW!5pn zZtcnLYlJr>fi-+v;XlT9{t|9$6qL_!9Pk0F1~ungd*TRty9^-XOu@PT~M#U{$J%1 z+z~w&q{1Ed6;U7E7Zt;Xo8@7uv^Z6qOB)tbd4@-4!f(-aI7c%Kx*;3-_b-4A>5!&` z$EMGk2qzNPM-oAG*w>mEIsAL3ZB&nVXZK{pOuR~#K+LEh%(-3r%I<1B5gj^x%kY9d z>c*$BE3Si_Ix`fa0tho&P>_96poer4wk6_>ISfH-2foN! z;$;cDR8s<0ceQIqBrl!=)eUfcgP z;r;bR^a9@Pz27$y=iL!^=xV>+)cBD4^n2AYtBEE$>|L+MU92|N-0i)F0MLlte4xvH z9-QAJ@a!rv>?l$;@2-y1oXnf$^1F~O?I^`4MLn_AaJjw4FY_BGIc5NxuqN$j2)sA( z!Y6~2VbL2PX&rEY+T)Q{9VJjr&zLB5;pwfEIcHXqz~@^aX^06^*5O!ly4cUt>^qg<-0IUvOAFW&AuEMO_QE z(BU7EWmrv%;Klw2jw--sBc4pKG`Wm9`%2?ki#1$8wdK3R_MLyWos33XV^pC=61fO* z{ExBx%{6^I9^QpZYN8ba^&5K&5|l6NWiPf|N68tMSYh(XN1+|X926!mCfW@wnsDG0 z37(XOk&=vGLa*Zo-W3gxABh)^`Qd{CnwR&wsk7j2vLpg!A zDH8WQ{b~sukR<@@x(~#taN@6=dmKv$qUU++W{0hj_dqp%wT-Dsre?1!pnS1kNETYB zr<4iHh7C^UROi~Z_K<Y!z%@{` z!R85nK#R3z00*k$7;c*_jnd|LHV3mu}NT=Tr;?&u-Wxj9QrEbsU^}jZ)4E0T6jR*CtY* zj^P?I!t#%TAWoMl?%3bHZcR_{s@9vo3? zFnkyV@72tTbY#m781$JM)ariXJ)7gfB5BF4BJaP2{6YC5$Ef`1i;XPTdrlN90!>T6 z^>|@gA?XObl)wayN95XE%gX*D!|T{d@t=}=#fy!Na;g5AR=vD$FJ5rz3YxCOlr>!7 zumJmj)+*Y&fzGwG@a7r-5@J0v^07h#Wb`U@#@6C-LG-EaK1`KhrYzIpL^pc8*w` z-6{*c(pcf?d(?zNpK~5@_pl7&qE;}5f{=-6%g0qj@yG?ta#(B+qDAL4d$|vw!lasj zX+~}?IZuX2X(g_}J$>c2sZd0~+NWUU&9Z|AB9-`vMdP(^$#h*>7e~58R`Wj1WE{&> znLK&ilERmB!?WttSYM96Ni7*utg9d`Ts@~9k|^^)$~)Tdt9Xe3cZEf)=oUI?dSL|- z$_!{59(;Fr)(1zmyewLNP9_FWU@849!h}Tcx(t_5ib@YJ3@PxdVnqXYhsxxP?`VC~ zO8+xHZSMT|jhm&@c*3qJl!0zt;>xfhg5pC{P6t?mt_q1Q9pRzz;Z#=z4a1=GPD+P5 zl$tNK4Ba@kliXi}s|SsP!J*dQL0eFaCa(R4Y-)Ae+JIq&DY$zqcDW09=M};iM`7!B z85T#|G@}fv!f7Y_NsqT_M?rQi5@}spDZX2mgXc0y&9-VSx!!PRa!zQ`Yq{SEvAGAqb4NyYbACtRCPI|FRnO4{9~p4ey0 z0sod|sdOL|!=;C^lMDo)97X-rLn&3cL4~&m?xr7U8##9o{yYR{ZEmzpuu)XG{?M6N z`FBosJ`*KQ+ANHYXGE301x>M~t1_x9sG68LQyGn-Ix?0nFBcnbnn)wGO@bF0vL(Fv zur2{gc#nr%fU5hCmr9R(U~jswj=LJq#;nC%AllOmH*1H-9xosTo;VBc?-wBMj9NDV zJ7#V13PWXNUcX{IvahHIeNS8@%6|pH%3h;0$wv7}Lv{d31p3JG~v#~AnmsKI*E6-bk6qm0C(TLX6SU11m0a!pX)HO zKOfcANYC0uF07kpS{hP#NSx(aR6I)0RW0p?ASsf_5t@nQ zF(k5YJJf3MiYBsj7P8hk3`xwAU_YdUONyTO^ktBwHaP%4_TM*+udXk8C__ozB2cev z4CPcM39Bqk5%x40QvwGaBrFJ*+@$Dj1vX)LohU3Rneu z+f~J=cS#+_egD2rPn-B{&DMf*=xwRT^=OA}ggHC30IlW@YuLqO4HKLBj@U8OJm60i z9>8q%^Mwh0^%dZ_61Oay?+cV;TdF~D{_xzKjWM(^RJ1H3Ngy@cd^Omg(-7y$Y?h;Q@NrS10Si?o7(mRZ z%GGPuHF3Li@xZ8lV|dzjO@kqp{XG=;(kqHWj9%%)-O$V!Q(Y$L#G`b_lYYV#TqB<4 z+_o)7i6$Q18qEw7UOykBVg0kEgU^s3)4i<*jgyu!_o1{FAxyiaoXq{5wT!Qg6B5d2n07EClPd9NvHBumR8=8CNjN}r?G-9KDED{j@OMMwauzw# z$N^p!CE~O5bfmHX&K|_^zS*hP4my_lvAPEZwOm1E*nZm!NjErw7jwO31;ALreRb&8 zZ5EQ#w?>g_4)?{&`~(- z*iUNlE)raRl{@+8ly}Yd=&QhbP`8>SAEaH|rw=9fRLr@sy~!KQuAk5!xj)aZZAyPg zmZZ!z;Vm4^-P~O*OdS4}fC=h)|1`UxYE=E#EYmcge-HiNw!em&rcJZh{~2HYgX8Q` z|I7W4;ihTW?Um)9{G(#6|MHtetAG59i_|RfPh9-H>z}YGe~|XS(*N7_uSvkaaLUrb z|5E#przdCA(LW>+2dQ9x1N?75{0+#qgH))JN@!MIPF7Bi^uNy|qW@QJs}q6VKv3b8 z!TipW{o1Nus5SV%pe*_jaH^Y)^Cb6dmLr@lwQEQ?ZiJ-

XPs}W_ zgJ{(>+a^W`^w9Y&0uCYLQS^dK&== zx}_5INEJ(9G?^k_lsY8bq_Ykg<#w_jmxU(M2U=Rv2%SfXy($x31ZQT7+$8yNQl$)? zYK8y_yfL9!T!}2L2gA=ozG&=lzDZvGTy59$II zW4XFKmezNK8eXv)dSttIn51@Kz0SjR+fH-}^+zl?zD*g^TWCM1mKP(|_<~>1xyi^C z1hF^QVw9>-%Mh{VvI{JjT9;Ilw}@j{&{ZH=Dv!#<6lAhX`Vg|G6nZ2K$Y+dbW7u%h zarz?Db#G#;BHJ+A5LTq7kj_w~CY=C))XCV1gK0jYm=uDtlN3yp8to44sGT56i}fM~ zV?Kcq?YOP%>;06n!YPNf0-wHjQ(Rx@ys*^UoUN@(5BE907x}B}V8C!M<_8nVE7}Wy zdc)?qSHkX5 z5@xxiP&7hC%8!^s|8AXt=(4VyD=qS;Ijdl;-^IS_JIZqvi>q^P~W>E0&) zvl2}HW2nd?_iO>4-07T4YW9P+wiAdbzxcxw-};Z> zKZ5^1!6*Jj5MwV5q6?g058?QtEiP`K{AKi`d;j;+3*W==x`hwqn-{yj`SRs|lh41x z=mLda5XSi8E`9zvcIli`~c6| zn*3zfrmXFM{uN3u5UkoK>%P5zxJDWBqBOa<&H(BGo8xbnIGvHLzzkiVlXq=T-lEyN zvR|=R8(2R))voeQ$3AI>mc8gGD7v1Fu%R1mx~{yT@PP-4_c?juiah{Z2gr!s$C181 zA66K6Q8eB>;+wa36eV5|3uyI?x!Bu=Zf4Cs<;RrF_`3Nqu``1Ad7W2(+pZMbtJ!jM z_I8o?{8d-yYx?6#>Y zy3%)iZn1`(ABw;=_~_5uekB$>=M(X--mQvOyktj8mg$=-T;+G<84NorzW0N$rfbCh z>~^TpF+bOOSY=tl12G(XKl-dUMM^S{3YrJ9ZLfPO!u= zCGu0xj{-5|d)J(?=gS8OP=j>#WkV4b6WEcX810pw9L9l{_^An*oCE!?m;)Y>#R*Kf zdm9v-PDdhVUC-Cg$N{~!{-M3}I7a-8O7vHKS z^1)+c@DO<@OvUIwxWIeudkGGALO1i`P=Z%LlYuWuycl@40lsBYHLl?@{cbIJ5R1`WK0p zB|dwCAl5Yk90@spM!U{B_5z@P?J4X$4c`kCn!a*DI`B}0#S=cCDm0etg-EXWx2`cl zFAWt!xfWuK@!CyKNgOHEjr?^ZiHLB*8^+p)UT|4*FwISVQ&we9p4?6x(lzn}5AfT> z7{PHkE>wUAEzF7(Ep!6k^J*w0Pb9Mzk|JusRs|9XY7sPlZ?N%QT+ZT@TNe;x52HW_ zk!zsP!y&e5E4Dco(uP5<<4C+TGLECqZ-z=yOTNrYA-Xzamz+Pq(b2ID+L8}(ahM8m zxG1>TtQSQ+FQgQ7Y!Jtx7iHN__5nksUKU#Nu^Kvqz!iJIv$`q8f#-{b*yUZBTimba z93ij9=;pb9WS-=z-wo@H#+xhz%ync3k;|~1&Lv^4q1mc%X)qWtJ@e(U(P0Q2sRly` zCr$_-+*}4eT?V|Rhetl%xZ%i0G&G|g&;X=SFEp2h0VC{Uy9uye_9f@Kjdu8)%UY1; z?3b!g4!9b0ZYVDEys8+WFjsZWflV?rA&8R914he#0a$N5wio)3P{9P{sB?r&DZdgmI&E2M?Q+5i2LZbJM zRbijFK@5Qk8-ulm@F}35=V$2-wFEe(($#aR(u>QxoY7RYY@?NU0!o8I8MAvWP(mPYpJ1YJk*w z_=+))wM(X?pngY*%f)t(mr2(dEH%-8wn`?HL`K;tF?rWFl&O$4yilGg7efU+Q}&|Z zF1c=lX?pkJR9C%48@$jD$PJ|T2HY?rBm@H9aFsAQ)6u8F; z)i}R#j$?2h7$jY|h+o%^iPF@v@gt*f5nQUl%=L&@PQM z%0QcQQLC+z#(_LK!iro3n|WE7j0tw+Mu6vVXz+W2zgNbWxVKL#6L|I7DEU;}I&LbgO$)HQHH!Py@4&P6z8C zci<+Cs`*|>V?+Gqi$6TMc*UhbPv)u08~!Yd6Y*~~4C2l`7PSK9yL0VfGu%{Qi7Mrm z7#{>DrCCcbY&o>j0RD<@tKbK4+loUJmO$}-Q#F={PogyOzTX|N$tCeVv|DqB*3t#X zrD#4UmM_|k|gJ{*ygrFk3Vw4D%b zv6p5BDN)XYB988qGT@!~ltCX}xD*D92ZA6E)chOD`sfdw$|v@+EOKbMmKf38r2jlz zjHElo$Q2983S=kS?fYWA={xa`{nxo8yp{3<#i5sEmV%&If)&JnATd>79?Mm)Ovl|` z${is-9w|&@q>z{y9uEo=o7Iba_J>A>jZa>XaB!S}d65y}%%n7p1cxyXVEo(Vb%dkY z*XvEwvW-VL1RQS5hWKsCcZAyjr|Ash@;%=dGx>aO;}bS_tj!IZCvOtzCWRSYW5OR~ zjwwE2^iI}9{Hfc2qZdSf91>K&@3foa=j7Xt&#`UF%P)bi5O(fuwv-{=3kuN*P}o%c zy}`gw|CZZn_nE$X3u+dqSZL-^r%d~1-H( z!;B#nO8gI<(fCBo_&e8R0<{hrX69K#fC@)4EEWQ)_a23RIoyvKgm{FlVXlt)z5Q#U z_gg5lJq^r9{1D&chqHgTw{Z+wF#A^jH9_yq$!B>6bf`Z#eBhiEXumC%XP~`M5W1vt zJ=$>UtYfI&mbPQS!u0ob@$T_?3L398Z`+(K(tq{pm>rMK|Mk_YlQ7`*>xt9YZj`=x zG8UZSX!&w~=<&iK>Z_&_BM>q1{j%Pz_f>vV zRGNG_2m@M6Iv<8Mzk2^Gkf=`|3DdWaxwGv-J3AA9QS@bB6_2wql5Y(=BaRrQjkDKE zR6o{K0pkH;;EgORSy zXJCgfm&^%DPYDy3SlbNCm3ULI7aVNF9%e4~l4r@_cy=@N^68d8uVp6_L%!a+*4GSR z*{i%0<(xHgAZy`Rw_o9%S>r4QHgk-o4ViSmPG+^ z%cjS6oW|#9n46%PJsDMj)-6{lLR|ubpuDS4m!fd|(ED z3O*tO1^E-{BWZYa3d*kB`u*wDU$|4UHP5r}PW3kctTO%Yl@RkGCwv^85rcienjpJZ zM$G`V}n&nkSg>gjSJ+_G{&DlWLCzkX^aAU8L6B}l2QB_caeC^{t*wx&=JsfE@et46G#9R%Zl{-a%L zMR(7Ot7t}}nA<{ACxEaOQI?Y0v6{7r2vt|s57g{(zko#Zq3c8{HlIQ->p&!4*>F)U zvJf`>xbwNlbZvI#GmZ3v&KdN>fzAts@bY-31N%}ZBE2xNE(8Skt$q#~X3>hO0dcnc z7$m@f9T9cRK0^*g2i&`1K8H1bh*juHg#@>lOr4QLUP*qFi6{R{bb)3N%97BFIElHJ z)Utyr8!a2f zAkb72D8;RW;TGr3&t&Ooygg5vZp|`)ZDZdPj*ndxYqqU$kBZ2MxNfwE6e#Uj7dP)Ck|87GEjBW zq6ACUWY99o#*l4)PBl#yqw3xmaS%bMXa372u?Z4&UVUZ|0zdQ1Z$_v)n2h?-GjPnROaU0 z$Rx?jwZ1oEo&ti~EgU)!>f>B*QdT8PMJKcCC0MD_;VAxpku3zOqU)R%(YK%+uFIdt zD-c8Cv9h99Cd~W959 zMmA|Dv;9?nO9+9&A(2Q{k#z;fEmLms?0HA3aZ0UTRIF&)f6t%OJ@(@SSR;{|V3E^T z408)J%k~H>Twq9_PofJ@)T-%<+9U;}*^oS(;5JOO^1RLh@6gtIzZZ=(ta4yC)*5xw zi=qSL^JYxD#LBzU^ZG3^qYQgPYI;S(mwwQIAr{+z4WhBlT`B9(D_+D;&W(fN8Lts` zaFLEbEev~yKYe8A=)^q|gj1Z(YGGrrC0G{mH_7sK*~Or$4v!oWYgxezRtfezKb$=Q zlVZJL#RAN8v~X1ai~3mNRB%@8it<@uLJM(e`b8q-#j1zRvd!71UA|dT{Ue>g`d+w^ ztaZbG82bjsz8+PQE!`!QUt}?8ouN$~eO^|Y-XqY6^q1gh%Nzj1Q|HXg50NN8go;V7 z3|FJ&j2Ke_b7$m$M+B2(l_2L2#h!8&69d2ZG1}>nxYC9m9kS5y@?q+K*_jZIENZIS0;t-4kQ}sOfRan95CraIIILbGD^VBqvDVyYK9&dB$y~my@@4j=gYT+gS?fWizT*D5z=^{z4X8~7LoaFK;k15OHV3yl| z)*=7`yETcW8eo*&*j;LN*_qf((YAVJUf#J2LBS$U;E)3h22vn(FphSKZ&>RgEdd4# z5Q~5doshIuDV%-)=1;Z9fdT5z+xq%7`nF^jVk=2VbKMkTN9om>|&|;R)GDL^OIV zZ@Jhh@~*75dnDJ6k{{I$L)M#(sccN|W!?2fzTjz2N0;Y}N`pRC=-jtLr@4fG6Z=%5 z<6ZL5^FvqY6flvqLMQm3Ds)Iq$N$R;o%pdUbfRCnLg&zIa6$q;I?HP7Pvt<5QVs-M zg|uuZR|rhTk3qI;Cn{9Z+LREnML4m0{0 z?aKe3>^C>c+rKixV6OB`J^HRPGIOim!Av0Jp66xJj=g7CdCw+_22|~TMWWSSq@&sk zX9r*5W$dT26J@8d&hkfe{jk`otcqouZ&s$4S%ayWS>VdHjSN4Y;W$iYtut2sOhlha ze6vxR;L^Erd@DdjPqWFDaZA7o`VXE^l}wv zTj@DGqV2O$_%uqhFfmb;f!)efV1fQLi0plG7=CgX{);;d?{n;bXWB1lDs4IM&c9XC zd;CrJ(~yI;`{t4{-Im+ZEq>F@o3`L@)uZIO#ha>GzR@42--9jMR6}3TZ{WH1lc)7x zz|)E?oA?2}rIe9>Jf7X?lY{oC{#}pnpp8E{XmtneeT4d^E$iOpf;_5c5AFFtr4?b! zV42U0P0v2btIaBZ|H9?LpgS~16`v^Z5pPX>FZah7<>`O~^t9jKNzgBAzQ;HBf_;2W z*#Bo!+}@U?{EvX1?eeA|772OvVL|>Vusiyd&0()P$d7K1Y90XZjViI1-sVCnl&e*m z;`zGy~m?T-dONcUmxANUCIVwZ$5+#U`Q0~)i0i9D>jE0DX^?IxhIn8&Cc zb6%7dd6C|@?vTK(RXDCSPNVluRqku<$gP%HWZhmR{UOde$iz!k-<)k(R15b-RH@#v zWGapg>`e9f42qSulo z__>;YcXaD<@7r7}sk)U{9a`-u0_5RX2K=dm%~fuL6VMjXQ8$R{_g~q^0|53Z!kV{s zNldd^M!!XzefaZ{WJs$IaZqW<@GNE2QEwYiWi~{er?B2>k?6#cC zS|)lYb-^1^G#b(AJ>^6%5?*oN7mMkOlHEyLY?-u-QRO7TXN#6xmaQ_9RJMsAC`m(A z(@nOPbdVNl29&byII&>+-~RQvlez+0uV~!9m_!HMzPNSYzNl60O z2#1!s376bnm}Yyrf+eJ=1m!g>HT`O%Ia*3d?8jX(_QBPA#!$c8qp6uZM2ls5@XvrIy0vTJqP*qavH#Tj)z> z-D2YKrU&5cIF@>87Pwnt&YSgS=nJb~t zT=kEU{k)=#-dSyTH!0e6S?5Ce3Nqf>KM}#!6FdDUHpx<--7GCrF2YjtB z5#3@s`;y&23t#aI%E5rXsrabfT=md}CnzDof8v6o-?MC3@{O~Vv`kPoI_TBKQp?;X z;zYFd$j2Ah+I^JVLo#x`Df^Z;@-sLY)8kmPE!n4IoJ(uka`wck4@zEt(KBWF`vy~@ z;O|EF(aPRxf+acH=UIAS463#=vkvbZF}~4M(&?3G1&F;cOYQXju7y)!dC!s)$}N=j zW)S^Oik$WgyK+j0rMPE2ciNb~4>B~}#{drM5)&Hquj46evP4(m?g9tDOQO9^{iRYQ z+D^xz=lh8Xx38V22QP7d5`*%U;6zhTm|87A`{7`<`W{UoV=n_`Re#DxOATY)Xlcqk zOM50wJY$KUrMstwpN;k5U6&rP(vy)*59(JckE7Fa*e5PHj!28FBZ4w+P?`U)YL(!={`bD&u0Ik6ciHs&JB~E5-$xy2SNE(TOtJL3M#2IY5(jpADN#F zjvfi=cg}X#X8%^wu8BClQV&P?LEBizK=`BsEROb@s`y{tR1NlavGzezAX;(;rf3X1 z*x_$Ud@qVEhl`1SnJM606wnSy2q*)6fY=#4GQqUVR)Wc_;|#|LBRK881qcql?U zd?X-xfg6HU$VNJ4czXH@7~#>DZ22wSjFo#F8IKIqIIA>`XLcr zKD@$Nc9M`#gJmRLH9P$ppL50P&U%f{(Xl$)L`39Jm5lh&g$%}(l^?nFV=ieV*nV^e z;I!*4znF-<#Z>HtLjU}#tc~zkqH!>KAFyhSGMQ*lHB0WD?pJFeUUp_5h=%(CP+At) z83f-vpGeS2vgC172$F|?WJcADSqH~YQV58}MBUF&h| z>IlB$dgBijoYy!@-0a~&p*rE60FIx!Jk^sK9MMkqL<}+#Wt4A!3*VCFpKu>Z7yuIr z*pXy=jwHU&o;t{z^xn;7{b202u`^`TNYXj&lGHgawozA_lPICc^e_b9&b!p3%!~J2 z@vJXazs<)XHi5fg=G=3&h%${w9!?h5ItW{C5O*rQQ!yIRgH1$wSNWX`nTpiJre;?u z@vtaMC4ASft|o4Oj9r;FiFVb9o?0=~y$HPKcSW0v+d#U!Dl7v!m+2THkS!XFvB%`C zXSs7lo7}oziCuxiwl09(&s{C4^bTd?o1g6KZ?%L`{d$>R-D0CUZW&Kow+QXNnOVQ4 zhZ(T)>MK{fINiws7;rmOcK3afK0l{v8T=5$Ww1|(?f`m!1T7liFIm|P%N?#4dFQBQ z=jrhR4@#?boaiAYQffI1gly=4;uRq; z#Qn{I&wIAn3y}AtxcptCGbs8o&eZd2SjcjKd@YkS=FX08aDe5akQZJ`G6RIZQoZh+ zy&$_yrDeZh^ZqjHs;*iglzH|KYLDMrW(qvRaj<{WJ6QM&AH7QbYBxvvE`LLB9{=Wn;B4Q!R+QBpcb+oQi9 zRXqED9cebOE=Cn|yf4M8qgsjlsae&ZP1TwnUVr?ouJbbt13UXO$?-hHP~Vh55TkKk zA3f01ZyHYiXm%cuf0vhSb5ovSkKg5+raF0`5P;Rxi{>17@a0g{O~;_}tOWUW@(F~q zr_YuSWIw8)J3=RyQ{&Tz2*Q(x2%}Mr@aum|Ot7Xmx=c8+bD3cC_pG}8JuK%dQjbS0 zV6wJ}>R!%k#ZzPlgth0L>dGNlyZxU;f_#sZM|v?^nEt_wKTQ9U|9}4S$^QX>uTgY^ zaXyD}J^_buJ_3hvJ_CnwJ_NUMJ_R+R1~E4?H!wAqtl$M51~E4?H!wA~&)@}r3`03G zF-AB;GB7nYGC4UoGdVFiIYLH7L^nh;IX5vvI5a*WJUKEkMmR$1v!6PP*rsl$MLn!1%wkIsCWTIz$8QzR0JjQhN-DpY6=1( zVmYbVousDnp^2#;dNA|JY_ge}n%-)9@F7b}%UY?8X?I8;8nvkztC^f8Ti@l&=dAx( z|Fiemd!OGrB9gkE5V3jC0`o;Aecj1|Y#0VPAZ*Tv%4-LNN+GQx1+jljKoU|3(H-NV z1ja%EjDZnwGyAe37qV26MRAm2ki(V!E~4f^(w7zMPmlc~x%SI^T1;@|{iloNnIh>2Z2z!e`FbRWRh-bIRFP2ycHn_mw-_+ngQyot?*= z`}aD#COQw~IS;OfFPw+2I1h99M`=8E$mx5*d7N*xd!^IQ$NG=Nb=(GvRi2y>%T`#z z4XdKM9hR!>DU9wiSPpl<3YBNeqp=cJ!5UbtGH@mucfwt;9@eV7_*693!QHSCHmJNb z7>zb)hfUC-viE;PG&-S+uNFVJEzqs9zcjj=p-1JNlPoPF)T?q}Wh}FRs975P@{hdl zd}Ps8O?=lF75O#tYblkF`+dt%Zu`-*&aqw2Cp?r-Yn|h{u+=$n+4-FJPJZBg`MhIg zz`g&f(K$8QIlT=oI%l46zW&oWTjPAQ-8sjjJ%2A;b-sTcbiS{KgU$~hIzPVc{6u50 z&l&1+ep%yO;A|JJ#w9QwEJx+y-dHkPECWmn%TxI^27obDtN<7uR;Y4$XEYcdHUgL% zMsUpM#_tEC%gC^pW~Ee--*TvC$poQ}gTlzK z%~72&5S!;hGZ!-j;!ZLJj45D@2U`q;#FhZ*v86yXjA$5{u@yinj8qtLvDH8zj6fJk zv9&-Jj4T)-u??z|n&Jz!sV-=bs$I4B>8LiU?rDFGszdeFOHp;I9=>t^x>SdLi)xeq zH8n@ot%-`_s5Wb2Ui@!ak0#m|Mzuu~eeXuqtBF7Qg)0|L53LYQE`2VRs>z*AQ6)6_ z^1i5&nmo2Ss+1-#bw-ueR9$XVS(;kb5mmOP`p!o+OjF0}qRP?Kwc@C9HT}v!R2is& zsZb;f(_lK(K|RcXnCP>h0cJxZG{GF03(cAy^w;#2pXGmw<}&e@=J5p}2{AbeB_%~q FMhe!Yt2_Vz delta 10099 zcmV-(Cydzt;}?(Q7qGb-0yH?2;gcwT?LBF8+{TsP^($E3q%_=+FuHL&mD-YHDRIhn zC5m>lTe7SIXFv`t3^2w)YMe@b`@X&pH0Iz8X(`bcat=%%uit&Oe|B@`pZ(#9Z~aH` zAHn~h;1mBWh_ROj(HTy#hj4t>7H8K_{xbT}z5jdZneSnE-NFa*&C6}yeEI5s$>%RH zIzyosgfTvQb$yn4ksqGn$oDdI_G)qVdiK|HwXTZu%K#;_s;rCr9LKYkyiaFsanlxE zcaCOdQ?rj&&G{uxW{aX?KN-5RzBzyM>QA4)NU%N=M#u|s1}mY{bo26E*6f zO@6X%Q`YuBe}U371grMRy06cFE>VWOC{4~TGk|))=J?w=PG@8*FhkepqQnbg0j<6<7kk~%&8*m`{Fss%UpGG{c1G|%uk&hu)0JX-HCt}Z z-Y)WiM*_+l*EwWfg2V%5&1UF}cGa=bn`{2jvS1(FvXdxtf+dbA zk)L{g6o?_;yXK5NUp_#98lXZnXs`6-Fb=%LPff_=9O$>j9Po%NPGG_{ zM}cxx)Cye?TM|x;Wuw@OBT2sB`#yH9oV^rJA>XGsfoHFlf@JD{zAX#B0D;!M_*N~E z4;~wXhsZ-=Dn|do1>Q^FOK`Xqx|tV;61)PM417uA#lW)-@ZFTRMLkCMHm?^AC%}4G zU2)3fJG&|BqRo3jCyL#0S{v%}nr(C?&Y%OpTnz$KEsmiE5CC-SJ&HXPXV%_B|19yc z#Ai#Wv?c+Azp<9Eq1k#&Pud)leyF$#3#fh_24qCFc)tbaZTkw&X)x9Hv4X zE($I->qSw|3n>L18^m$wMOn6$eZWwumxY#mtcK1YaK#?*tZqti;Q3-9c6nRo7Wb<; zN64!&x_K^tnJ2mGcf)F}@g@raa~;`1HO*%|%I&{)gh(UhJVN~LL$*Djz^>E@F97z8rLN&?_`LYLI=zW8Yns}O?l=b~*m zh6p}!%W$R438Qz+j4%c>QJXpl*ke(ssY#%J>f#8rmn_Om1<0gJux&vMXpa#K%2;7j z7eY{Ji?od?gYN&SORvxGU2dOFV+=Ar`nPNz~vafahq*_ zx?%I3dyp87H7Hw~)%G)`5#yZ-ak@((#$z4KoW@Y)obz0w3~VVxj7c~pFbEjq5prjA z(12WNcoG2NUj{&9B~@|fg3&A}?jR(#Y9gJXifE1!DHWtHqj6VC7BMK{slmog4Uk$7 zFBtP!xnxQT>UWg5Tx@%JnRK1OQWI@|t7Jk+WR#5(lXrbXnF?9MOXZnzF;u`aWiJZu zlIu2@#zj~8J56uNg_DjR)1uVmu4T5yftREb5C1vgL`YF=v(lZ6$r3dNWBT4lfqR@# zjq@AlI0ol|aYEOUL86B74!b1-oK6DsYz|do)5t@iwwfU92`Rt7*h~>p0Mbl<;TXr_ zziCHDG(L+mq#%_pD&nPdohBmdF%4IsBe!qJV7~8Nu+|-KzjbqtI1YC?pwh8Rab#;B z#(%DrNyBteuK0uQfh=!d--7Wh&y>K;cERbM)^$FAM>YN%Cq1JW171b<<|s>{D&$qy z2twtaO3?27$n7|>Nd9hF&V~1X>pkpm-gg$p=FI)p-17GnFMA1vHB)KwRnZd$?b1l2 z4752Hwc09a9LS?1tjI;MnU{shm|#b41b7aI2EQlx+ZBb78C3J)9(2ut^bK(+8P*lI zi7HJ^hs_UVw={;LKxIEExc)vy7iD=nRQjHWLj-0y9-#tCx4PF=qn!nRH82b5bg&L` z3vSY=n(u`)HpE}P{KJ#8*IXL(WS*+L;?J@;5&u@hAnx2_Q7cfsJJ%l8!&L>As8Vi; z@j-A>nzaPOZiZGGz%S^w3Vs0BtvEzs2^82mzap)zPr64GlUT6)g2R{xF#hBGGQ!d9 z>(#nx*~TLr0uHxjL;SYnJHl;%({u)L`JV5KnS4IC@d=wd*5-!IlQ)TUlfsOyG2ss~ z#}uD1dM9fl{?yHX!3&~44hgE?ciQ#gbMkG+r`R^-<>$ax2s`)I8_JOG1%>DcD6Fgg z-eBOTf6eW*`%K@x12qd&EHv||Q>J~pYDkvFJ>lRu5&jLe&f1qSkzt%sDo(r%)E&jY zVaAXOCH{xbXndk({GDqufm(YFGxMw=K!qb277GE@dym3@9PWn4NnxOak=(9WlI@BK=K5$M7wBHstC!oDh5W1vt zJ=$>WtYfHNm$qZT!u0obargK<1&vplw{1=q>A!w`$c{(n|K{TLQ5f*%&BSSJ)=J+z z8VgQvw0t>#^mySA^;J`e5r~+0f7~=6_Qf0*Ok)#}5Tup_FaqL#=)lG&3KXBUm!IuP zaVJ~0SToZv4_Wf;Y1=&AQ(zps-28pnQRF`g_YdQJ-ci3t$4yU9BN{K^%>&`X`zpUG zDowuZg#oQ4o%chVU%meoNYtl~gz1~d+}UQYot=n(DEhLmipSX)$+w1`5l4*D#@TBn zsvqk_yJd5?EeYY0S=K1@{4f#Lz0;x8742=&32**w(F(^=Uao}C>}!$bW7+;gXD7U^ znwmHp`ofVKX^t+lK>E%RK^Ub&Yw@H?c~w{Wd|8Ubl!^z$Yo6F{yEpRchI!}9e!1#w zXM4GST7ksMMFz%g*%43T&zfryV6Jaoq~&PdH#;a#{Wmf%$+4uGI`Ej)$K#g7!ARHU zGqA&#H_Qo2PYDy3SlbLYOYx>)FF4qUJQsr5MXN=zlS*aLZE;FUIL?z;RZt8xUTNVYx zEt?+ZmOUg-k>US&8qC^_q+_1;5qp`PEUSPp8G*EN_5%;c!OlDvy>`OoUkyAGeL8u6 z0t)gcvPaV3@D!A7x%E49BffBFWUBzDkEWLTQKxZ?^IU@Mc`hUm)u(asz4Kgx{mDmg z1idOnOKj$8BRox2q!xa~yyJx(=hQfV#U5$inGrsaVgJRYYrQCn#f6xL@-Iq;LFPp` zF?bAERl}SI1OgEp0RnC+{)$Rb{VRii2f%wPGY~$Jp-FhnVw&(mibI;?Ne9e(w}_Zv z=T^XmD;ovLWUF|*SY+#{47*MNDPcge5*^JjXjk;kNX8!<9zcabKb3;gv#|tU2Wmp6IQ_njd#Q z7x}Br)-0uw)bE_3JRIo!R){2z=P$4?bs~~EEyO2Pu=2BzuAb{O@XlI2d@={Uv1n;6 z0K8)rW|24s25|Ts*bz~(*q3*I+C-x4Eu3}3d|njJsJ)&&B7rT?NS%;GT}gbC$t3?r z?LAIu%KFPIu%)M=U2gwN~qzI0Rc1^g)s$ zmZp15AJTH=>6)~}OF%cd)E;xIY}m`-hlv-YntJi%`Ku>CJwY_@0-d2K@gh_DQFcdk8lz?q zYmZ?s^;1hto?emfRx1P8HugQ?%-B|iX6yg=sB(;mJ4Sm*fznQYG1Mzc_fOW?Yf0u* z0j&rz)+DrA1DH&^w(Aot{3V&1d!a8a96%-lTiJ2IQhbuHp9CqO3)-uABC!~S0S|y! zKuT7?7WEVJkj$B}ha*JJfYdf@RpJrd1xNtT$LUCkw}v&6)eo!!*>fW{V@v|VK5>|u zc7dvW7G+hk&VrVIMm7$~YAI>5z*P5mY051r8QwT;8KAZhh4(_-bSvVTX}-L;8hD6r zWf!gVqAsbLysNpQZDs|N9J+E|_F4v->FKcYFe5nSBsh2l)2I?)~tI) zmyOECyd4=NdAZW}MvPNHaJwx-2SRH9fan8^yvrpgSA;$pod$n^hrdCCYjs7gq|8@QSQD zH_h--DU-P$r6a3^;?zhewn3wW*T^4%;zvv~CRO^XZG+JHjdDg+p=K)b$;>Pk$)f1M^t>9=F0t}1^jv+5%qYWNlR90|#HAm9G+>Cuc7teEbF1vXiWlaSa}x#? zulsdyk&Ztt%zB59Tv?9An7BuRaEjAeZD|a)1j{1cCRvg$I}=pZ{?Q;}Ei0J8D#4oP zhqEIHQmodj0DzgP3r9u2sCgwW1!onmD2Wv&v=Eo3S0qAStoqX2v^m?f%QH(VdZf*& z?}fX6FSQOAg6^=+QY=r8DzlcZ9?CDW7_`pNrj9-@D^2eaXheEOaI|F(fZ?fg<}F@{ zrQv9_oDpM6VD^3vctkKsR{nARQ0ypCF){FaAEPTx6*AltKtxY3^afDPdSDT$wsOYT z7Q+?S7Z+y5?o;5Q8AU=Y*pJh^*z&@Yqh)lTD;uTn$wCd=-}tN01#h!E2)_%*RVrwx=51iSrAnf9JxGy z%45oMxR>R&HUGcBZcSpTrWR#qJG|}AITM>H+E(|M=AFB^6D;C{xUMi5NP*PBINBw? zVXcR>tQRamEW#~xLeeUwaQXq5KSBG&@0>lku$LgDxv_q9I~7t@8T!NC<*w#bDJZ1L z*y%L#l;AWy!(Jqo{ew7M?XxKlN}x=CwT-(gl~(TZqs!p|fFMVLd{8uTyw&FsAlPzd z`8-VEW(lwqQlRKZkryV(4iHC&SA~xE7a02+Asz6fjJN_x?}X_X?4`dECcWx#d^1M&lm9oL!Ex_QkgM9pUqP zNI$Y-2m1RPR;PY&^-l9x)jJ=5UpWlSj%MERB{6V+LqHl*+XuWi+^E+!Cc9idh}i8W9C*312WqAaL@Bi(T=@;XIOd9CJF&m zJw>9`Q>3GM3TIbd;brWnvJ++Zu+H*Fbp5c{sH}>cHeWAIFS7zYa1DUJi~FA z%vxux{F#V8llW$%^1!8Y<-!$|$x;uicBM!3-x857)|;0X1exw*)paYyGvUmR0G;iU z4W}>ZTJO57)vRFAvul)p?lNv5WVyT2%nO6`7Yml_zUQ*F&RApt6ThXEltGfffe{xX zX5#EXZ8WMi(=@w(B4O=uiLWj zT`tIj`tZ;`4pdry5ylLb`Mg;7?327&FY_;49t^reV^r~p0w3|#)c10Kj8UErNI*|} z_#FlPvgUhydoS3>r-c3Qo8tPqB;|hu^lX<`1+hrTs}BqE2Z7zugKQ3a)j@u6%Tx0J zc)wGLz4V3_N}*h>(i~@}ZFqW;qyOw%zFyEb&uQ~zbIH?xXCIBuKGM(r*B-vU&(VK& zUnS(%E-q>4w{$}^G{VtIGGykaJ~>i6f4{EsvSt8fO>mhLf`Xgq)?cvKh;zPby67)qgYHNQq_(3`fbN|3ch!@)= zjN$fhh#1(o$vGah5hlz^(jXqNe`oSg%cdYbdH>E|G-1Izb;<%dbb#pSgoR{}9-f~P z7Cxj7o{zS#N6bcGmzxo%vemPby}vHG#{2>2C%oE!=i;vzd5OCCi{$;2aOkzOTTSF) z)m?$ypKdqyl*K$o?U?iOv&f6$;IP20RXDD-O`~^7Rqm_n$W4`5WZg(5{UOde$izzn z)rq7ms)hS z*IP#+(2rIYoY%xM;nyY6&|BcoT4%#V?4dY+*}*W_27u;}edQhVZqW<@GNE2QEwYg= zi~{er?B2;j?6#cCS|)nWbio@@G#b(A9pywX5?*oNyL+xpBoVPh3%X2N#;9_V;IltMZ+e zfylc{TkKcMX63Shusg=O)lKr`26PyCX{+F0WAA6@N#+CU$rY&Wu@y!B~)f)IeUS7n_Gy z^`~P!P%&@R4F6U3B@TFACQaslhIiLPFCXvNC86XEmYpeFO@%qhQR>7TAF;2sR>5f0 znt^`L^;>w&?-cO0zC?73>Fi5(t1NuQFC7O1`l{lic5~JL7oMPm1pkQ(ihj?sVaYen zM$$4t+328G7dKkwHW4SHtw%n-l-BN}s8seypf;5$(SC;nr+E{J{{v+TGN)Z zCsut>@`|1*%ilMc5(R%Zx?@)MRue4A(LT@817lFNm6^4F=ZNtwrIJoBL@Pk-g;{E+ z_jfIv3d?(zoKS9|tk;9+cT(iEXV`^PI^2l6!*i#N>H8o<(_IW;uP!m6LH|0QvL;J( z749x@@Vg}1zf-?xIvt09p6@3n+`e|69=ya!49Zu66HPr~YPJ0AhrQM6do+cNy$qC9 z{V5wQHH>wmr780)?U^+3R5uZn?w%TcHr9tn7anMBzFFzX$fgJNOEvxPv>f({ON%4Y zA}e`d7;=jkjEE$%Tbge%W!ph7!MKogW>+LBgmytK4XHM93OX`>yDgQ6*tM64w8kcV zj4@bS80}aMGmxhHM{9a<^qBT5Nf)MP}7syzbFNT#J+LEq^HD7!_YxczSWip z0lI=pDoxryyUa)CCxfF$Li(Mu-L=`jm9%Rjj<3|i5q{7%)-e!1=>UtP-KHx3w>MRT zon5Sb&=iQ4oPjBS8p95@_*)X+i(<>+Vq#_rI2Q%9LlOeYKp!A>1`my$IHTzKAvan7 z-pBFY7`lt$O(PzP&=wyFh+g1^AXPH7)kQ*X0wJoIY7Kq2bXLSEm&DX1E5ebDcWdV; z^Y!M|`!^lia#NNNvF1!5{9vYlbyg42(I=GYjPL%}Gb90F4FUlvCkWPeq$5@imSWFO z@g6rc?DE5&_=*Pn%Usc5!imIQ65S7b5Mv0p3WY2N z+0JF}fC=->b}+eZ;5d{99=a5fO0^nsk!U^HDPF6xPecntbETpn{+wQv03QNMY-?S3 z$yQ}t(P%e45I=-7*6{b8uzzSBXpeqKM3)b*aF!h^e9M)^A1XL6ahAB*!-GO~!aD&RKXrMkCo?#rt?r2!WF*Qc z-+vaqCCxwKK9Vp1CKRwE$#xt`e5O6MmpAFXo6Gva*llBH$fl8`bJ``To49F{#dt}S zP-J=-f^X~n=uzgyDtr08J1V0VtKa715Szf=Fmvv?T11(~BM&EwYaN6wH;6lx?o^CM z^k5T_-erC(L#84%v8mZrN<1veQVH++<$s0CaO?(o{Y*EyYD7=180uaGUh~_c&BbjX zU0xNI0iDZq3=zl{4aV4G^47E5xuSJ$-LJ&1z+qb#!0zX+mQ;F&vhmGN_Vu?~!l-_| zOs{USQ60C8r>$Frw%^OFU(>@3Sb6o8t6iM#^wbQ;6Z7%juSn^M5-+q+w+NBkt>#+tY2EKvR5RYF3X$c zCC{R5uf=`h^;~3+6?wP0?3+uOkfGX}|6l;Nv%-{x zu5Jl0RT$W4ct>0bOGa(3i|S-*4u9w*QjaIE=Q+)^_i48Y`bp0J6Hv(K#WgYHgOpwB zCx__sD@umsH6a`NpLj*c3vqvQ;Paks_5$SnC@z23=nRT}j5GDT8Wyq~AYaKOjk&X< z8ysM{DCC8glFR_1FI2BPXD`TZU1`}b*u1~Yx~i*I2xVRnPLlgZ*YA)_eSh=*O3Sav zZnNhX%~SNvrE`rStLp}C=apNC)L}@~RuSgN4;Ff{%8a`(cgGutg6TFGDTBh%IdAH^ z(AbnHD}Xbp@~&eS#yM`YW7u-f4ttj&W{MH=!YDaKw>byfdXz4E>%}kGUhZqbjSxrq z&FR~%X#-njc$Ab*(e~)?2Y(gMen*-Otcp>^9Pdl<`k+=Ke`=QXXH&JN`_~^otLyv( z!@$n|OmaL=Fw{3C5X5MlHwO>&^qYp0KboBef+p%c=44Lj>W`Lxj<&Mosw5KPFhyTU{m` z*||)x`FmE~eg(^ULF(~{1x(g9QQgaVt$2#;fUx$wQ(ZX(Yq$TCNRaQ5@<=ac3)4S% z`G@IW^8e3YJ^6okc1LQ1VLpdpJ^_beJ_3hfJ_CngJ_NU6J_R+R1~51@Ff%ikoZtl= z1~51@Ff%i^zu*Ob3_~z6GB`FdI5ageK{hr;F*QL$G(HyiNF*v6B>Uu)N7QtdzA|mPQP6lMcAjkq?bGE3w_Lfj7 zq*Ww0mI;4ILMkEJSqKF%4su~EjDVZimkEO*LnWCPM;QcJY?(kc6tY#aN~1dja$pz? zhmnx0GW7M>k_T~)DO96j9E?)QTNYc!z*s1NLWr{!LB7i9WbB&=lVCi@i)u2&k>eJo zLa|E0(fCFQl(Kmu6hj^igUK)yM!^^u2{|wvMniubWhP8hS^PvCc{!`X4x*?H7?pu*Wz=($}Jasu5cfwlO z24p$(c%!w;IW_h$?19AImt`;bdl4W z4||=@{&qg+vs1m!7cV(h4m|tQwa%G&&X@b(59h0AoUl){qN1ShYwdcB=^GScW z=3JGbb(F1I^BaJC;Ku*P_>EMG<%jnQJ+z_hR&l`HF_!Bnwa zV0hSYm0ur=2E)Tf0CU3#j``fUdOW&}42x+to+|QN1k+eDL99UK&uy_ewlG?(7$&Ox zwK5uv5o6LAIk(9w18+xzsbWkOBkX@xs`B5B`{12hoP(iYvw*}Kqrk|yRRIyV8LFvC z@wK@?&W#yi^Pv(L6h?lnT6IEqY+eX8+{|={C&?5rrhqXXY&j4TTM4AcYJq4N(J(S& ztASJ)sW9SV^*|tuKp07}^*|PkEEplNO{!Ds;({7fmu`=$QMKdwsG3v{Y>0oVS@qR{ zs9ID%xbgg2Rr~*qs?GnJR!6m26U8M_ZPCP{_}{XvnrK`bRl6p-K8&hE6W0z2cP^Ul zZx&6~9*L!Daz|ZM2~ECuEUKg?dv-^a(&R5YqDpJ3B0s7OP1SWqm8q$&D^U&7RPWNL zvNSbN64hW$AL))N8)m>vm<1$dP!1JP39}(4`W%=G^I$$KfNEF>HJa}8ft>#Bs{9Wq VtTrZ>*6{@(2{SbcB_%~qMhYTOZh8O! diff --git a/doc/pub/How2ReadData/pdf/How2ReadData-minted.pdf b/doc/pub/How2ReadData/pdf/How2ReadData-minted.pdf index ee8c1fabd314d43fe31073c5104b5b6b7841d9a6..e8e88a825f816c1cc80d19b388e5ae502a1a5cbf 100644 GIT binary patch delta 6943 zcmai%MNpg#7p!p&?oN;dcN^SYgF6IwcfyN1!3OuBK{Gf6cXubahGBvQhj73D*4^L5 zQ`L)J^y1X1?!gqYnG~`g2CN8zTlgOR7ZYFgw!|A) z*Z?(B%8lcm>HFXrnhgW%pItpCP8YeE#KS*>fhvsTAMOAC94CJbLw{Auum4Pdo`L`B z_gGPRW2w3Fpewm1(9&I62|J#l`!5{&eRvVQ9l5Oh9SLA>fj5Y<2J1CS%b;&AxAZ8@^h8-H4LLt;2DE%!Agoq4O( zv>TSQa#AzJ`J^8{?(%q8y1>nZ0>_%KSzTW*OibjLTz(Gc#$>O+(=^k*t-rOoPAP zLSSL~s4CWHCt0rZBT|VxmFxf>=nhrt_r}M;RSuu?jF&;rc}LqYsL5_}s7HOw31Hq^ z(@h=5GGJtlMeUZZA->weC*a9r1*;UN(!|+RI5SKC&iJUtf{%U_WklN09b-CAOu~m9 z&?h^C-TgeL=D_x;tAOg;DV#ZzSS@^B7Q`aoH{SVTTI z`bNb*5dvlio}F)>&RpPSP*>5D7);b80wel7{z|7~hh&rI0;cD4ZE#a~0cvfon&=!0 z_yUJ`giAy`uEB%HoC)X?2K{hMtx*jN8nXS^3G=W*b{2HoicY6?5h~pJ22A6b*jPjz z33PsIQ(kbm^sx#E{83Wl5B=8)*NlKeWQQD6a{9FQx5+r@*xOgh!*jG~{i6fcF?Y|t ztSov>c}GjkcBZ1Kp6h~Raiv_=y^X?SfA<8F)EL+&PHx#C?Y!m=t=iOEK zcs}r?ao4y(n$pn`_$~;A0qe->?%PAV3}^XSO!$z*xGb0L6R;3-k%d(C=;zsuMgeWZ z-{ic~h7)PKb(KUS3`q2KSo>qgYd*+`C)6bec(g9)vCk+D-K9w{p;&7QJrH|8q>D5# z`WG|#K1~()!5f?ASSzJqm)v^uSTbcy=9q^guY7+E7-c&hyBz)qAhV9Iwf;H0)4kSm zSkwWl5GDFYuY<~(cg|vZ6T(a5Oup_0FIu!keR)>IKBRSjTK?*DrHHFQyHD)r5B3lF zL|t~Y!X)uiviK0q*m!SEXD#mc%&c20k~4v|mfMTbEx7tpW4lFq2-B(RqTu?&g`9iK zjXz&uCG1K!{@yIjz$S9<39i-mF=)vUVqXPYRk$<~v z@4A0bIaaH5Ynv;wiZa|W&nBBiHs6C&G8{wMRhY$FT=88%$i^xwfI>la=0$Da-<7~w{Va2P8xR$$q z-pk%(+hQl-_!L0|qL6f8x|}Xkx6#Q}c5lliVchgIU!tO}`34=*{9O4*pKwwW9L`-J*y>4S-^>}FtRb%!aBDTToyt?_%rn0m4_x9myXk}D#&(J znG>sqJjQjmwy=pdIwbXVUVsdZvh$E7p&3os=+t)AKAL;uY8@ctl` zJW@f*MElT_m%(A`voshH6%cs6S#nF=Smxqhv}pFPN`&H2jN0k8-bNHmk=yZJ9Q-=3 zqN?dX+Y`v=EyF1cnL_&nV@qj!r7rT2KPBcn1UyzIN`F#hK~eYm z$>KvsS)iF~l6i?|h7URqKUhc6@jOMa!288(G}X~(Xsn)se##@OJyZJu>6=_Sejsdr zzbfp^W7S=}Ee8Smlb-1ULg!_l4)TRK{5Y4qkCl~O#*fx3j7&?}{dtgUEhsm8Olqvm zAfniH@W(K|59MX40FAb#o$=oQ!}$%Oly*?2uzx^N7XJ%eTvp^7Ws&Z?QEkG&Y<2(3 z(}njTetaKVMfmpn9Z9c5_w`t0kZpLrz04i|kaUXR$SD!I+8bMUI>aK0X9(kK!Xvx; z=!1gYi=9C^Dl%_vO^xNXb0WIddb3juqO9pE_eeu;9W~B;1~T3LyY2On?VqePHY^!* zGl?2PAw!HGeL9MSRZ0;G>$WaUrqoM{ZqVkp4tl4VC-fbs9*|~ivajg{&S-Hxzlz_g zTI36V4Pt;K)kIV;u89k^ut&L<8DzWq3e@L*d|huoJISHwK$~WPw2*wnxQDG8K{y#^ zS(S+3?Ixbuf%W-(QyWT3LL52*31n~d3?qE(@PEd?FR2{J<(u9o+BW@-j(kh=b*W&3 zDI8~B+2zx&)=8LX&u8NwT0^a}4lYT{HVMC*!fU!$f~H)a zAwf)Mh6gE0JNUl@0JiK91DZdFZZgBZ`Q#Gjtsh)Y0Ll$x?QIQ(jK%lug&4uAYk}6B zSD|v7PkcpwD#vjmEif}(7P^8!)(jOrM_K#!SsTp1_YP!)D&_&acxTH<|ArT0%#_Tf zW5y5e6C&tj5iVrkp4Q@)&I^&4_M_hK4IHWrOidkfYzaZ2cT;$dvSWBJ^Y)nXg=!oe-~(a zL(7(HYr)XQ8^Bz-Q_-2?ZF6hThN%Uluck{Ai|cO}tc~5&*Q(ngzUF`O;(aDX4ec$| zG+OGxJIPEHD?7eWxJx%a65mkUyt~=8RJp3>1Uy%vrcrc}zO;Wbj9cz&nhM-{8sD%P zZds5QA$y#JQfQe4_f&QfG6KK=(!p%$iOi|I@j>=5_@4pAuFX*VH zA1U93_{@w{uE}1mYu86cO%_EinMdy}Q}>ci=E=u8h_CVxIph{wN)_*SNeB|_NW39> z@|a>Y6$_Ooz;plKwd6WPOR6e9hwoC8DwiZH=nK4xVSW-|G}brzy4X}PvCnaJ5u6Ln>`EAp3ciqR259SsMh_`L#fYNTl?$Eia#e)1B!Vib>%q}kER4r z?%go<43gWA7fzkZqtl|`42J!WUPYxTDQ`Mdz&amdoSTT(YQ1<>)|9%PMX$7O~D_u z18Fc2#8Fd>WTL`Gp2`qDWG_z+%7N-59eM+d1=L#dF}KA3xzBfrI!^!<;nnR{4CZ}~ z;q0&Bsq?AbaCSUU0jt0*TS8x#(G;V$lIOXW`Qo(}$YG07+xpg#z)*%;*6qWO@B-R{ z8xrZB298`8W4Ar!7#9nNHk!uf4wB{{#hA00JlO-5okls~yX{OH19Mv`EeOpYZvAz^ zH=ZW5A8oknXImx9V@d(#9TZmOs3^whIim&)a(PEOA~PI%MstSqcQW=Ze#N)u`0HO} z|44W28SILmHa91!r{BrivBo%%eFw46huq%17b*a>N!d%O>h4 z>1W@BTxM+Vad`kqrmTsdpKlQ!o4qkyF;QoNl?7cWDS1}s>JhPSR zely?n=9@;`MMtsNmCvPqv1T!IpWxW{`#Dc>8oq3C?TkU1Ox>PyOW&S_lq=h)%C(Y^N1~n+#jJy1on}Bz#3-k9kHZ(6`?Rrz?7Ob)L?HKdqFTrJw>HE6H zwt?=_4R`wL+re)e7OnX$z2=9GghUg&5A~qWa(p(s)FlWmM0D%^etzvt9&1K0w3^WL z;(rUb>*lUs#?BFB+_2o>ic!-zyBG$O7|G5}i|4glC@2G5(tK0(i^F*hBkaB0)g$cd z2Cue9ESsx9?1GUiyGT~RzNArd)dEiXssS6;LG=7*G7k=4v^sSiB%7G|Xw+$y+pNj% zFUFmFo1Zsu`WznS-+$kPvy4BT<}z}__?e*`Q2cj{1AsszU^UMa9bJcinj5-IR!$P~ zmoIYm*Be>Zwc-N(DQ>5HebaoRDDMdXBNyf;*OWg1cfw4>3F8nUmeHc> z$3v~r>g_VPX3VF`)0v*UtG}f!?`l_1mL0qDIUa&~(#8 zAkLZEQ@T~s%mPWBu9gcLle!sLEmwO$Mf$OK1}BII>Uu(8xQx&SDN&r>&q( zVH~Ju)kHf)Iz|Ehxy$U>iv2IBnL-eX$G|DnvCF|=6OD=+f>bS#n0UPfTI(9BKVHW~ zaTCTfbEn<1$6{;dV)`OVY-(jz!lL8XrreiOqm{-&{l-U=145=`%YXpU*X}gds=ffj zD?v~30)0CBM+}j~V0vk=L|oOOajp6zYbd=+z)*A+fgvz%)m$r@p}i8FKh6uvYC+-} znthh)D<(2W$IQ&(E0F$r_}W_!TG7eETt2A^Z%2 z`sx+d7MG>I9O)KP_?a2+lQXif=Ak`8y2e^Z&f0dbYfVgU0=B?SzaPe5to5!A()C8l z2ojRqF#*c|yAi86)_y@Na*>!YuY-M^=fd9}e_fpjcwSQmnoY7t zJKKY0Cc|YTrPmz3giFXvp5S4F-ch#;Atk#Dn~46(pA<|FhGA}?!Gj%~LXu6F3Hl#T zr+6((FCnl4VU1{HC78AUClgUoCIFg$<{wIY02~Vla8iBd;3zYxwD6J$>dqWN??Ly; zy(tpz^{S!=5BYFxm@*w*h zo1vlX3Beb&M&D!WE!^|gO^PChmHunan}kypRljojBllOltT>j=gnOZ+?$z$5q-g9Y zfNN$gl#utslTA{b&$z*FE{Lf?{jU+f*EElkH4Z8iS7zojVv<9F333AH!sn8~A-(Zd zv)&7$ETbCXZP0qutqG4-mJ5`USBnItpBM8R$LL6pA+ApzdyuQ|ci3V^tdvnh;2 za)}Ckt%$&ff+i^Fn`_O)Ou@y(wKjALQ1rv^mTn?A#>y>Vk|mL-ocQ3f;0H)C=eLm(~5 z>G2M}gc~+`h~yptVdp;JVjm2P(DGLM)q7mc-o?RKD(CV6E+h5CQLYpSBxS$G#K_r} z`B{)uWTBwVj_01g`)4Zhd!WsZJo4A2=om5V&Ahyg;9@NDmvW+7D;o1V`c^HXyG#kZ zU&>R!VhJBOTXj#BT)vh--N~5wE2<|4#pT(ct3vc2^6=+@nH?dCbQ<5#ioEulLBSaI z4T=7(x}l%$eiV=>EH!1sxCmxKvF!dBP_ZfK7gfM!%oIek9X-umkUOPzk=RkK)@5UGonB%*CdHh3>A8|{Pgw~?xO+nm_k z-Go*96pZggPJ~icU7 zGDb+#N&$QvNU<2X(Bt3JjBs&OkYmbeDjRboM)4ggk}hAcCWk|5aq43z`T#2)UGc<- zD|dR;aBlqTog3W`eX>jzxe~;mh(4)>(J@7jC$Mg3-5n{oRO(r6}Bwx$UUil-GhcW`ZY+lYf%Ah9UpZ9O#az zyYC6ISX=8{UDI=RcH@4+w1|NmRC`Fch)D2I2lub(A~sO`B%#1rrU>#S54uVGf|?YA zng}TT`i(_O&onrLcAbHbxM+{Ng8B(L!wgf+-rbO&HXx3B8^@R0e+G6)4SiPm;HdzGfidN(?s{2LW2f@(z_PJi%Pr1Su+7~ zPDK8mQ|_K$cEQW&Zsg`omOFF4nOM(vwH~e%FP85-n|H)4@`26vzAKZ5AyOmj;NDTv zW4I7fZjK(Cmu!A{q2wapL85~bJ`-M!`77w?$dUF)21(M+L(5LnzdI8QVRFwpmo;EB zwi`c)@lN?XV~y2gzBv+@c@S!Bur^1p^o(F43B~tdyi#V!$Y;$8_q1V()5mWpbS;v) zYe8yH1=pVO;dPu!$i`+^IDv5O>W~_%t#!xC_2LfAuF1;ul(P#HFntm|j2Lf7vj!NGI&qH^bb1fEBel^h@$Fm)#ve?-8D`$YOr(#3 zI>8h7;AHd1hm2%%{kgs8uC$5!5TX<~MtEfeFcK5GCVnx4q*5q8B`j1u^;qE*t5a6} zZ(xhrH}uKwv&^bPq6PuQ=5xjd0p%s1sTwu>B+(PyWCrDifMwP=)06JAbYQ5+3E_f- zAp@fFVg0sRBAy}n-;plEc)EVKvf_1;R6|-C<>@6$roUs7K6Bi&ao0qu|6xFbXUV&X z5`>qELU>(GEZ!0n+kWS&D3B(qSO07qJo$S>wYhgU#d4pLeD3TT^9J9DN=a-mY+S9q jygh9!UH+%sK09J~d0TpBaVer;pbPS0Ffz(%$YcB;3#E*K delta 6984 zcmai%F4*@-?()gL|EV74SKlq7SJyOnA3C_nhi#3@_KZcnt4ZiT_h$tDuw@= zKeit5BMk5_$Sv+PvoAo+d+db$YOZe3B@_8IY@P+Hz!%+94XhV3!nqZ}YLN`o=Ii*`*U zT>TzP_m2d2YZp7Lp4mCc)KGuh&#{!NvkOrBS}b%~k~^b<$|L3FUCV!afhwj#4vKb^ zT0G*WWl)nPo1DN+c9Yc8WTP=DDVv+v1(nNhR-p}JzIAFx@Nwv3gwDnSSbsH;MVYJ2CC;#M1QyRXgwXrJg ztV~DGDqzbuH(xs4(_B7&Ze~UBXf@gWS@hBTD9pKpW;Q^4d|^z!2ILjUPp&n zgrJW8KP~4a*b1te%leN^I>2s5nIOhs^v4wcxtmf-VRI;`#S{pEI4X4g!MgsVdSTF=gIi8rsUo|1uO_(|ogWgh!4S?w?jcZkw2VTyvs zQ?J2up@gvl+JJPuZ!UmO*If{r7Qm{M4kk zzFq7`Ms2Z8MLYpf(Xa7BQ8ckX*YfK$_-w?_w6oDy{eV76L={^X>etAo=qM1P zbuJeH$UTK#xJ}~&$fG6n>qPYC0+OMgYRwJWp;QjqTVm2q`BS1QUHDs$-<8_6t)p49 zEIbiNpG8CcBx5zCO3b2VhhX)+U60%Q$5A!?i7FYap_oH`^Otrl1AaHAovYjBo*xo= zFAgQ{YNAtrH~}|zH3OR7Jatr}-PePNIlV(N0GN^6iFjF+oo46O4|vrC8IEXMWmDm6!LZb?7b9%YJcu%rJE@h783O%_)9knxjrwU>$)_A@4 z`b_Nxurl)1FWfb3x#)ffO=66(0<}d=lYyB7^}a7j2(jg4(izpq`be3;k>`J#`^Nl+a7dO$bpE{c?=zQs##coA+XDQtVQzV4cjBap% zhYaaqAwl*M-pDFIpGA`FL|)7;H9QiQH?>0qq!s(=K;y(LdFQpsxBIH(7t}v&#w{_9 z(TOlRiRe4dYt%u1T_Y=p8kexSwCx(O_;#JI9N4jLbj=(GW!t1Y2MO-;D|bvq#4UW5 zRU1_z0|o3T>!g#!D``KR=|M{^jp1?CBRlmFSjlOjzkl?Cg#iK}r`XN;I=pyXNq+}; zhYuF$Ub(0)d;b%X_On6ArwrUX3n6#l!Q45#TfW~S`ECpw70kgz{@{Jrk8_YKTR@{` z!&b#PWv8qPiMMXFw{7ZB*=ubdM$7ow*~rNzdyDE>bF1^I2YOBId=(fX_}yTZ?zb2~ z_a;ZQ@YkJi`r{R^TE%A!Ibk7dtW7Q8v!)C1vaY(1>@L2$hv_-HYj9L99w?hd*A_@QNYn43@Z4sE@t8?1dBDQ%V^tQ7+7 z2N0p&z>xN=JW?H45U=V>s$&2o(@5{sqldg=x)@ zz1>8k)RSXECPnX&F?XG*JbJ=}FHxlY6=N`bV&2V*_EeDJ5*DTrrVYF8uROfp?{-#O z#~>A!S}ZnrmTu|eA4L&(*Dtw3^ww^nn*BVToSjO~?BRZJrYdyZUpbFNfIJ@JbMQPo zLXkTFV>N2s>8Q04wN!*{id6u8&$S=HyP_1fp7mBD>EPTBYO(Y`^U5lUPS;I=mXEm| zYws+eHMxvpob2@$`sc2N!{FdfTvU)mTHPGa2cF1HDz_OI9R`|-AFt9qq14Gg@ zMKOv%l$4X^!r^Ob!)1a@w^635p5cUDc|!EV_MfaTYbP8@s(Z-a_QLevD)TU0M8kRW zlHE6U)j;jX-6mK^GBH2er@uli*6RYePw)urRL-Xj9x6T6rp)QK=PtOCL^c$u5)_$j zcNuk-SBOl_;E*$r0aR1xEzt*-0>3GoW0Sor0qHG(}>KFa{418$WET(Bn&nvi^vpB$Ik`!Uhf<kdcOPN8o6F*pAHSM3J(arVG1e0*>>)nm<6^ z1@>I>`MIliAZ-AH5N2-o9I%qxTHp?GR`nI2(9Jdf2b^hnab8fHpM>=NKqgY%fd;H| z$u zW9gPv=O<5xbA$Gos&AV9S1BSf{q=(7@$2@r0&C=#d@Xm;2Mv_qfkF)MQ zG@K25gA_JVQ!S~=GrbI;Jv4LLU zmU_>;ESD9`(gDlp*mh~Ic^yE*?txEmZ#sj$Rr`SBn=|ll3G~OLaNyh?b|9*WhS3cK z?;%+>nvj-UQat}thc7vU{)#zK?UwGZBG+l6SQ#b#`-HYY;$Qc=Q<-y$xzL-9!_Cu~ zB!wSM8e1Gv<8SziMi5l?w@Ja0WM*FEwe0#D`pw;48*Ik(eRSHlPEDR*VgO{qN;>QQd)mQD3mTJNBI7Ht6 zJxM-4s7T_)=kOeBQe+ck1iXM3Q4B49`YC-QwO=ibKeiA7g@9L4v!vap8d>^X*i|dp zgw0KF)&8^+-8b-r_hNOwL1GP|o*w+ZDOY8x;->t**hlD#YJ>TB1^P>h2X9{9q!IWLMV8Fj0KZ- ziII?Nu-1%%W73sgWsXneL&K~cQ_)Bk54>B6-+bi zAcdare{Hn?7`f1m^(U%6;mt^W0)4|`_cWZNgStMTr*M42Q~_I)f6lByUTAJO>qfRsia%#y|RkR?FP@($ORcoJ(2>78UnxuZvJ zE2FTnkV?;-N)0WPgA-pI!>tclt0~DgE%zeCY~n0Detntl2U`3F(3L}Vl4P|4M?B{KN;cF_s`qUooqzX>12%7e4#I56WWvFV&T_RI5K=4osGboJgM zoDhRen5>nYg3QD0t1U)sz-><(pYxkyZ`3N3O+oJ$1hs$s$#HyvYnEiQG{X6&iDc6F{QqLXC@6JoS&;L&M@49ymSPnQ0Yk&6- zEAR%l*Sz{;?Oex#Q`~9QoeOLA&2Z`J=l1+Iz+l&ofN*m2zW%@B@mg(Clpr}0(yaIZ z-knUY%SL~2^$-~)U&>Ef2L3)KcVS@mm3JTpQz(dCr($+wS~lK=$f++nx$L$nuF9H) zfMzVQN~Z^fuu9vo$Na%j^FZl->l%(x7=C;VvaM*TlDLYQMH6p4ST4@$N7piStFR_+ z`g=KWKg6thz=Yxy_KSUk2pWW-Hk^d5c!m_tTq-aO>0>_CW7)$fmxo^wgP)-N-2)j@ z%5bx%$tkPArg6iULgqsDn$$lp?uCaPvjfmrHs;t}tfTvPK%Tte=EO*+QLM9b z=$EGK{OBFG=cZhB3eVbORggqaF%Szs?IBecH=6qH7id;DI&~B}aKIYFcA>r}v~bb; zSoi#}-I(8CG)XySEi_^h!IJ(3TA4@0@X~6cyP^T6{AVL5^PQ%bMO&L$qo)2vh7-z8&BG^YUk4%K+Uq*+m4()`IfIb zroLa%5=#jbwQ@5Q7&wF*8;}B4A6%G;1l!GGqAl74D3ndm`$?h8LYVnr;)>Nfm87rU z#(Bv{Q3T)ihD4S|_P|C&bTkA*YA5*5ACQETT*l2WzK>A*ghRKsxAsl?b9U`N&B>t^ zs~BJVpH>&5&B)snqJtxn94UYMay`T)O*6H zc~2`RZMGvgc_3ra~yCe6JevG^c7M=`?Mw*Z!G(n8nAiL~l_({KLz<%MHA7qdXe%rN@J0SyO9)onzcF z{Xe^WL@Fw#F4eIwXoHobP0zE}y&2=A%d(UMwj^X-%7GbbLnLFGm=M@Pb_2XBIP-}tBbmR`T59AJQ)9z-}j_#4x z0%ZFZ-Sfrb(1b|UF^{ixR|6_;_(rXNAj>gJSu)wJXL7^+mOb+v53e)@hrgfFivj-A z?nGB?-=zCM;rjKy0O4(n$2(;Qq7|l{nwx*A!(W4#gq}ek9=7CbJ;wTq-;ZWC$Sfm$ zdTG6w8b}0oT9JUO$9$Ricj(p=qi2xJSJdX;nxRX~!HAiwrKTIw5-)1+XuTRUK*JEB z4-~@^qoY!>4Q{qgIZdYj|{&d^Vb-<(F0 z=YXq4BIodxNRfu*X1923S&5cnbNox6nhp3C_->|@XbIYtDG z3jGdwLUmMY6|(v5n!sw|Hqn zd1*~{_+gW?QIoTX|AF)$$p3-TF*%ERZOzOj$jimY!_LLa$i>CT4B}LEwv;jVu%cp; z5$5LlAjrkXE%?8Yj(z?&JD6I$+|qnJLOk4(vb-`wAh>6;*rfi&{ zw**3CT2;eHC*v#qf^LWCI%tYgv@2`DtNAFiKxTotpG~ll(T&4Ys@s@=$cl&mDZcRh zaxd+A0QP*a{(2sXGc=Hb<3V2=x+o{2aNo?Ofi6R(E_FY^4r)?Ju8Yt>Pz0DFwU|OD z;^VYcL-3fsvLk92<6P0J9A9TJ4tlzXt|177zK)Vt(`197pM?pAEzxBe12agu@U1^g zQ`%@xf6K+7*8sWGrN`g5`eM3ThjNNjrs8qeiIv4CR&fzq&$lB?((__ICg**o){#(> zg4vxL{J{LwK=BKYKgQeg^9Mj@2+sgCRSc>wd1F_iCUqVt)`8xPC;aQ*GD1Ql`7>J{ zy+@LtGkEPy-S?a^p=+PG@VDYEJR5wcJ5G_tn|cP+TP0yjPcFpyd=x7dr-e=;Mpl0b zzax(pqFZn}CKd1&3V!*UH5}a3FtJ?3kA2ErNU*bsxKky$b5t|ha=i(}x5+fWsnm*Ub3 zuLpPa6`ND#Oc8-;nw+HGEEkq7o$cB>nw4zt)}~(P12UEkXq!%dMZojqSVO>bcOmx1 z0hUhx9XJGq30(uPm`*}5I2C~&Q3)x16Y|z0mKT04Fdjob>$^ywLdx2AoT+dn+R(0O z`3Gj^n56*6YVcC9JZ;n$#bfW;UsQ{^ExN4MF9q7(t^=G3W=rq{19%`n5c59of`z6j zM7%Fn$>HK-!xy}#fEIn%m|1n-493fTqlPPbF|UOh6t-XjHk5%XRw^t~1I&y@l*}RS z%22yXMDY#b$|yEo30k%&hKZWlwE - + -Data Analysis and Machine Learning: Representing data +Introduction to Applied Data Analysis and Machine Learning @@ -43,13 +43,11 @@ Automatically generated HTML file from DocOnce source 'sections': [('Introduction', 2, None, '___sec0'), ('Learning outcomes', 2, None, '___sec1'), ('Types of Machine Learning', 2, None, '___sec2'), - ('Why this text?', 2, None, '___sec3'), - ('Choice of programming language', 2, None, '___sec4'), + ('Choice of programming language', 2, None, '___sec3'), ('Data handling, machine learning and ethical aspects', 2, None, - '___sec5'), - ('Acknowledgements', 2, None, '___sec6')]} + '___sec4')]} end of tocinfo --> @@ -63,7 +61,7 @@ end of tocinfo --> - Data Analysis and Machine Learning: Representing data + Introduction to Applied Data Analysis and Machine Learning

@@ -74,10 +72,8 @@ end of tocinfo -->
  • Introduction
  • Learning outcomes
  • Types of Machine Learning
    • -
  • Why this text?
    • -
  • Choice of programming language
    • -
  • Data handling, machine learning and ethical aspects
    • -
  • Acknowledgements
    • +
  • Choice of programming language
    • +
  • Data handling, machine learning and ethical aspects
    • @@ -95,7 +91,7 @@ end of tocinfo -->
    -

    Data Analysis and Machine Learning: Representing data

    +

    Introduction to Applied Data Analysis and Machine Learning

    @@ -111,14 +107,11 @@ end of tocinfo -->

    [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University

    -

    May 22, 2018

    +

    May 28, 2018


    -

    - -

    Introduction

    @@ -126,57 +119,68 @@ 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. +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. -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. +processing and speech recognition. Many of the methods we will study are also +strongly rooted in basic mathematics and physics research.

    -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 +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. +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.

    @@ -184,7 +188,7 @@ algorithms and methods we will discuss.

    Learning outcomes

    -These lectures aim at giving you an overview of central aspects of +These setsof lectures aim at giving you an overview of central aspects of statistical data analysis as well as some of the central algorithms used in machine learning. We will introduce a variety of central algorithms and methods essential for studies of data analysis and @@ -235,23 +239,22 @@ and jackknife methods. The second part of the material covers several algorithms used in machine learning. -

    - -

    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, +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: +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.
      • @@ -261,10 +264,14 @@ Some of the most common tasks are: 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, the second is a model which is -normally a function of some parameters. The last ingredient is a -so-called cost function which allows us to present an estimate on -how good our model is in reproducing the data it is supposed to train. +ingredient is normally our data set (which can be subdivided into +training and test data), 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 +function which allows us to present an estimate on how good our model +is in reproducing the data it is supposed to train.

      Here we will build our machine learning approach on elements of the @@ -279,28 +286,161 @@ machine learning algorithms

    • Bayesian statistics and regression
    • Support vector machines and finally various variants of
    • Artifical neural networks and deep learning
      • +
    • Networks for unsupervised learning using for example reduced Boltzmann machines.
      • - - -

      Why this text?

      +

      Choice of programming language

      - - -

      Choice of programming language

      +Python plays nowadays a central role in the development of machine +learning techniques and tools for data analysis. In particular, seen +the wealth of machine learning and data analysis packages written in +Python, easy to use libraries with immediate visualization(and not the +least impressive galleries of existing example), the popularity of the +Jupyter notebook framework with the possibility to run R codes or +compiled programs written in C++, and much more made our choice of +programming language for this series of lectures of easy. However, +since the focus here is not only on using existing Python tools such +as scikit-learn or tensorflow, but also on developing your own +algorithms and codes, we will as far as possible present many of these +algorithms eithers a Python codes or C++ codes. Finally, we will, as +far as possible keep parallel versions of the data analysis and +machine larning programming aspects in R as +well. R is a language and environment +for statistical computing and graphics which is widely used in +statistics and mathematics applications.

      - - -

      Data handling, machine learning and ethical aspects

      +The reason we also focus on compiled languages like C++ (or +Fortran), is that Python is still notoriously slow when we do not +utilize highly streamlined computational libraries like +Lapack or other numerical libraries +written in compiled languages (many of these libraries are written in +Fortran). Although a project like Numba +holds great promise for speeding up the unrolling of lengthy loops, C+ +and Fortran are presently still the performance winners. Numba gives +you potentially the power to speed up your applications with high +performance functions written directly in Python. In particular, +array-oriented and math-heavy Python code can achieve similar +performance to C, C++ and Fortran. However, even with these speed-ups, +for codes involving heavy Markov Chain Monte Carlo analyses and +optimizations of cost functions, C++/C or Fortran codes tend to +outperform Python codes.

      - - -

      Acknowledgements

      +Presently thus, the community tends to let +code written in C++/C or Fortran do the heavy duty numerical +number crunching and leave the post-analysis of the data to the above +mentioned Python modules or software packages. However, with the developments taking place in for example the Python community, and seen +the changes during the last decade, the above situation may change swiftly in the not too distant future.

      +Many of the examples we discuss in this series of lectures come with +existing data files or provide code examples which produce the data to +be analyzed. Most of the applications we will discuss deal with +small data sets (less than a terabyte of information) and can easily +be analyzed and tested on standard off the shelf laptops you find in general +grocery stores. + +

      Data handling, machine learning and ethical aspects

      + +

      +In most of the cases we will study, we will either generate the data +to analyze ourselves (both for supervised learning and unsupervised +learning) or we will recur again and again to data present in say +scikit-learn or tensorflow. Many of the examples we end up +dealing with are from a privacy and data protection point of view, +rather inoccuous and boring results of numerical +calculations. However, this does not hinder us from developing a sound +ethical attitude to the data we use, how we analyze the data and how +we handle the data. + +

      +The most immediate and simplest possible ethical aspects deal with our +approach to the scientific process. Nowadays, with version control +software like Git and various online +repositories like Github, +Gitlab etc, we can easily make our codes +and data sets we have used, freely and easily accessible to a wider +community. This helps us almost automagically in making our science +reproducible. The large open-source development communities involved +in say Scikit-learn, +Tensorflow, +PyTorch and Keras, are +all excellent examples of this. The codes can be tested and improved +upon continuosly, helping thereby our scientific community at large in +developing data analysis and machine learning tools. It is much +easier today to gain traction and acceptance for making your science +reproducible. From a societal stand, this is an important element +since many of the developers are employees of large public institutions like +universities and research labs. Our taxpayer do deserve to get +something back for their bucks. + +

      +However, this more mechanical aspect of the ethics of science (in +particular the reproducibility of scientific results) is something +which is obvious and everybody should do as part of the dialectics of +science. The fact that many scientists are not willing to share their codes or +data is detrimental to the scientific discourse. + +

      +Before we proceed, we should add a disclaimer. Even though +we may dream of computers developing some kind of higher learning +capabilities, at the end (even if the artificial intelligence +community keeps touting our ears full of fancy futuristic avenues), it is we +who end up constructing and instructing, via various algorithms, the +computers. Self-driving cars for example, rely on sofisticated +programs which take into account all possible situations a car can +encounter. In addition, extensive usage of training datas from GPS +information, maps etc, are typically fed into the software for +self-driving cars. Adding to this various sensors and cameras that +feed information to the programs, there are zillions of ethical issues +which arise from this. + +

      +For self-driving cars, where basically many of the standard machine +learning algorithms discussed here enter into the codes, at a certain +stage we have to make choices. Yes, we , the lads and lasses who wrote +a program for a specific brand of a self-driving car. As an example, +a most carmakers have as their utmost priority the security of the +driver and the accompanying passengers. A famous carmaker, which is +one of the leaders in the market of self-driving cars, had if +statements of the following type: suppose there are two obstacles in +front of you and you cannot avoid to collide with one of them. One of +the obstacles is a monstertruck while the other one is a kindergarten +class trying to cross the road. The self-driving car algo would then +opt for the hitting the small folks instead of the monstertruck, since +the likelihood of surving a collision with our future citizens, is +much higher. + +

      +This brings us leads then to serious ethical aspects. Why should we +opt for such an option? Who decides and who is entitled to make such +choices? Keep in mind that many of the algorithms you will about in +this series of lectures or hear about later, are indeed based on +simple programming instructions. And you are very likely to be one of +the people who may end up writing such a code. Thus, developing a +sound ethical attitude to what we do, an approach well beyond the +simple mechanistic one of making our science available and +reproducible, is much needed. The example of the self-driving cars is +just one of infinitely many cases where we have to make choices. When +you analyze data on economic inequalities, who guarantees that you are +not weighting some data in a particular way, perhaps because you dearly want a +specific conclusion which may support your political views? + +

      +We do not have the answers here, but we want you think over these +topics in a more overarching way. A statistical data analysis with +its dry numbers and graphs meant to guide the eye, do not necessarily +reflect the truth, whatever that is. As a scientist, and after a +university education, you are supposedly a better citizen, with an +improved critical view and understanding of the scientific method, and +perhaps some deeper understandings of the ethics of science at +large. Use these insights. Be a critical citizen. You owe it to our +societies. + +

      +To do: Add references and acknowledgements diff --git a/doc/pub/Introduction/html/Introduction-reveal.html b/doc/pub/Introduction/html/Introduction-reveal.html index a34392da4..8565b2273 100644 --- a/doc/pub/Introduction/html/Introduction-reveal.html +++ b/doc/pub/Introduction/html/Introduction-reveal.html @@ -3,9 +3,9 @@ - + -Data Analysis and Machine Learning: Representing data +Introduction to Applied Data Analysis and Machine Learning @@ -116,7 +116,7 @@ td.padding { -

      Data Analysis and Machine Learning: Representing data

      +

      Introduction to Applied Data Analysis and Machine Learning

      @@ -132,8 +132,79 @@ td.padding {

      [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University

       
      -

      May 22, 2018

      +

      May 28, 2018


      + +

      Introduction

      + +

      +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. +

      @@ -142,73 +213,11 @@ td.padding { -
      -

      Introduction

      - -

      -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. -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. - -

      -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. -

      - -

      Learning outcomes

      -These lectures aim at giving you an overview of central aspects of +These setsof lectures aim at giving you an overview of central aspects of statistical data analysis as well as some of the central algorithms used in machine learning. We will introduce a variety of central algorithms and methods essential for studies of data analysis and @@ -259,24 +268,23 @@ and jackknife methods.

      The second part of the material covers several algorithms used in machine learning. -

      - -

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

        @@ -290,10 +298,14 @@ Some of the most common tasks are: 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, the second is a model which is -normally a function of some parameters. The last ingredient is a -so-called cost function which allows us to present an estimate on -how good our model is in reproducing the data it is supposed to train. +ingredient is normally our data set (which can be subdivided into +training and test data), 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 +function which allows us to present an estimate on how good our model +is in reproducing the data it is supposed to train.

        Here we will build our machine learning approach on elements of the @@ -308,27 +320,161 @@ machine learning algorithms

      • Bayesian statistics and regression

      • Support vector machines and finally various variants of

      • Artifical neural networks and deep learning
        • +

      • Networks for unsupervised learning using for example reduced Boltzmann machines.
        • -
      +

      Choice of programming language

      -
      -

      Why this text?

      -
      +

      +Python plays nowadays a central role in the development of machine +learning techniques and tools for data analysis. In particular, seen +the wealth of machine learning and data analysis packages written in +Python, easy to use libraries with immediate visualization(and not the +least impressive galleries of existing example), the popularity of the +Jupyter notebook framework with the possibility to run R codes or +compiled programs written in C++, and much more made our choice of +programming language for this series of lectures of easy. However, +since the focus here is not only on using existing Python tools such +as scikit-learn or tensorflow, but also on developing your own +algorithms and codes, we will as far as possible present many of these +algorithms eithers a Python codes or C++ codes. Finally, we will, as +far as possible keep parallel versions of the data analysis and +machine larning programming aspects in R as +well. R is a language and environment +for statistical computing and graphics which is widely used in +statistics and mathematics applications. +

      +The reason we also focus on compiled languages like C++ (or +Fortran), is that Python is still notoriously slow when we do not +utilize highly streamlined computational libraries like +Lapack or other numerical libraries +written in compiled languages (many of these libraries are written in +Fortran). Although a project like Numba +holds great promise for speeding up the unrolling of lengthy loops, C+ +and Fortran are presently still the performance winners. Numba gives +you potentially the power to speed up your applications with high +performance functions written directly in Python. In particular, +array-oriented and math-heavy Python code can achieve similar +performance to C, C++ and Fortran. However, even with these speed-ups, +for codes involving heavy Markov Chain Monte Carlo analyses and +optimizations of cost functions, C++/C or Fortran codes tend to +outperform Python codes. -

      -

      Choice of programming language

      -
      +

      +Presently thus, the community tends to let +code written in C++/C or Fortran do the heavy duty numerical +number crunching and leave the post-analysis of the data to the above +mentioned Python modules or software packages. However, with the developments taking place in for example the Python community, and seen +the changes during the last decade, the above situation may change swiftly in the not too distant future. +

      +Many of the examples we discuss in this series of lectures come with +existing data files or provide code examples which produce the data to +be analyzed. Most of the applications we will discuss deal with +small data sets (less than a terabyte of information) and can easily +be analyzed and tested on standard off the shelf laptops you find in general +grocery stores. -

      -

      Data handling, machine learning and ethical aspects

      -
      +

      Data handling, machine learning and ethical aspects

      +

      +In most of the cases we will study, we will either generate the data +to analyze ourselves (both for supervised learning and unsupervised +learning) or we will recur again and again to data present in say +scikit-learn or tensorflow. Many of the examples we end up +dealing with are from a privacy and data protection point of view, +rather inoccuous and boring results of numerical +calculations. However, this does not hinder us from developing a sound +ethical attitude to the data we use, how we analyze the data and how +we handle the data. -

      -

      Acknowledgements

      +

      +The most immediate and simplest possible ethical aspects deal with our +approach to the scientific process. Nowadays, with version control +software like Git and various online +repositories like Github, +Gitlab etc, we can easily make our codes +and data sets we have used, freely and easily accessible to a wider +community. This helps us almost automagically in making our science +reproducible. The large open-source development communities involved +in say Scikit-learn, +Tensorflow, +PyTorch and Keras, are +all excellent examples of this. The codes can be tested and improved +upon continuosly, helping thereby our scientific community at large in +developing data analysis and machine learning tools. It is much +easier today to gain traction and acceptance for making your science +reproducible. From a societal stand, this is an important element +since many of the developers are employees of large public institutions like +universities and research labs. Our taxpayer do deserve to get +something back for their bucks. + +

      +However, this more mechanical aspect of the ethics of science (in +particular the reproducibility of scientific results) is something +which is obvious and everybody should do as part of the dialectics of +science. The fact that many scientists are not willing to share their codes or +data is detrimental to the scientific discourse. + +

      +Before we proceed, we should add a disclaimer. Even though +we may dream of computers developing some kind of higher learning +capabilities, at the end (even if the artificial intelligence +community keeps touting our ears full of fancy futuristic avenues), it is we +who end up constructing and instructing, via various algorithms, the +computers. Self-driving cars for example, rely on sofisticated +programs which take into account all possible situations a car can +encounter. In addition, extensive usage of training datas from GPS +information, maps etc, are typically fed into the software for +self-driving cars. Adding to this various sensors and cameras that +feed information to the programs, there are zillions of ethical issues +which arise from this. + +

      +For self-driving cars, where basically many of the standard machine +learning algorithms discussed here enter into the codes, at a certain +stage we have to make choices. Yes, we , the lads and lasses who wrote +a program for a specific brand of a self-driving car. As an example, +a most carmakers have as their utmost priority the security of the +driver and the accompanying passengers. A famous carmaker, which is +one of the leaders in the market of self-driving cars, had if +statements of the following type: suppose there are two obstacles in +front of you and you cannot avoid to collide with one of them. One of +the obstacles is a monstertruck while the other one is a kindergarten +class trying to cross the road. The self-driving car algo would then +opt for the hitting the small folks instead of the monstertruck, since +the likelihood of surving a collision with our future citizens, is +much higher. + +

      +This brings us leads then to serious ethical aspects. Why should we +opt for such an option? Who decides and who is entitled to make such +choices? Keep in mind that many of the algorithms you will about in +this series of lectures or hear about later, are indeed based on +simple programming instructions. And you are very likely to be one of +the people who may end up writing such a code. Thus, developing a +sound ethical attitude to what we do, an approach well beyond the +simple mechanistic one of making our science available and +reproducible, is much needed. The example of the self-driving cars is +just one of infinitely many cases where we have to make choices. When +you analyze data on economic inequalities, who guarantees that you are +not weighting some data in a particular way, perhaps because you dearly want a +specific conclusion which may support your political views? + +

      +We do not have the answers here, but we want you think over these +topics in a more overarching way. A statistical data analysis with +its dry numbers and graphs meant to guide the eye, do not necessarily +reflect the truth, whatever that is. As a scientist, and after a +university education, you are supposedly a better citizen, with an +improved critical view and understanding of the scientific method, and +perhaps some deeper understandings of the ethics of science at +large. Use these insights. Be a critical citizen. You owe it to our +societies. + +

      +To do: Add references and acknowledgements

      diff --git a/doc/pub/Introduction/html/Introduction-solarized.html b/doc/pub/Introduction/html/Introduction-solarized.html index a8ed8a3ff..8953c7afc 100644 --- a/doc/pub/Introduction/html/Introduction-solarized.html +++ b/doc/pub/Introduction/html/Introduction-solarized.html @@ -6,9 +6,9 @@ Automatically generated HTML file from DocOnce source - + -Data Analysis and Machine Learning: Representing data +Introduction to Applied Data Analysis and Machine Learning @@ -37,13 +37,11 @@ div { text-align: justify; text-justify: inter-word; } 'sections': [('Introduction', 2, None, '___sec0'), ('Learning outcomes', 2, None, '___sec1'), ('Types of Machine Learning', 2, None, '___sec2'), - ('Why this text?', 2, None, '___sec3'), - ('Choice of programming language', 2, None, '___sec4'), + ('Choice of programming language', 2, None, '___sec3'), ('Data handling, machine learning and ethical aspects', 2, None, - '___sec5'), - ('Acknowledgements', 2, None, '___sec6')]} + '___sec4')]} end of tocinfo --> @@ -53,7 +51,7 @@ end of tocinfo --> -

      Data Analysis and Machine Learning: Representing data

      +

      Introduction to Applied Data Analysis and Machine Learning

      @@ -69,10 +67,8 @@ end of tocinfo -->

      [2] Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University

      -

      May 22, 2018

      +

      May 28, 2018


      -

      -









      Introduction

      @@ -81,57 +77,68 @@ 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. +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. -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. +processing and speech recognition. Many of the methods we will study are also +strongly rooted in basic mathematics and physics research.

      -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 +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. +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.

      @@ -139,7 +146,7 @@ algorithms and methods we will discuss.

      Learning outcomes

      -These lectures aim at giving you an overview of central aspects of +These setsof lectures aim at giving you an overview of central aspects of statistical data analysis as well as some of the central algorithms used in machine learning. We will introduce a variety of central algorithms and methods essential for studies of data analysis and @@ -190,23 +197,22 @@ and jackknife methods. The second part of the material covers several algorithms used in machine learning. -

      -









      -

      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, +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: +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.
        • @@ -216,10 +222,14 @@ Some of the most common tasks are: 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, the second is a model which is -normally a function of some parameters. The last ingredient is a -so-called cost function which allows us to present an estimate on -how good our model is in reproducing the data it is supposed to train. +ingredient is normally our data set (which can be subdivided into +training and test data), 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 +function which allows us to present an estimate on how good our model +is in reproducing the data it is supposed to train.

        Here we will build our machine learning approach on elements of the @@ -234,28 +244,161 @@ machine learning algorithms

      • Bayesian statistics and regression
      • Support vector machines and finally various variants of
      • Artifical neural networks and deep learning
        • +
      • Networks for unsupervised learning using for example reduced Boltzmann machines.
        • -









        - -

        Why this text?

        +

        Choice of programming language

        -









        - -

        Choice of programming language

        +Python plays nowadays a central role in the development of machine +learning techniques and tools for data analysis. In particular, seen +the wealth of machine learning and data analysis packages written in +Python, easy to use libraries with immediate visualization(and not the +least impressive galleries of existing example), the popularity of the +Jupyter notebook framework with the possibility to run R codes or +compiled programs written in C++, and much more made our choice of +programming language for this series of lectures of easy. However, +since the focus here is not only on using existing Python tools such +as scikit-learn or tensorflow, but also on developing your own +algorithms and codes, we will as far as possible present many of these +algorithms eithers a Python codes or C++ codes. Finally, we will, as +far as possible keep parallel versions of the data analysis and +machine larning programming aspects in R as +well. R is a language and environment +for statistical computing and graphics which is widely used in +statistics and mathematics applications.

        -









        - -

        Data handling, machine learning and ethical aspects

        +The reason we also focus on compiled languages like C++ (or +Fortran), is that Python is still notoriously slow when we do not +utilize highly streamlined computational libraries like +Lapack or other numerical libraries +written in compiled languages (many of these libraries are written in +Fortran). Although a project like Numba +holds great promise for speeding up the unrolling of lengthy loops, C+ +and Fortran are presently still the performance winners. Numba gives +you potentially the power to speed up your applications with high +performance functions written directly in Python. In particular, +array-oriented and math-heavy Python code can achieve similar +performance to C, C++ and Fortran. However, even with these speed-ups, +for codes involving heavy Markov Chain Monte Carlo analyses and +optimizations of cost functions, C++/C or Fortran codes tend to +outperform Python codes.

        -









        - -

        Acknowledgements

        +Presently thus, the community tends to let +code written in C++/C or Fortran do the heavy duty numerical +number crunching and leave the post-analysis of the data to the above +mentioned Python modules or software packages. However, with the developments taking place in for example the Python community, and seen +the changes during the last decade, the above situation may change swiftly in the not too distant future.

        +Many of the examples we discuss in this series of lectures come with +existing data files or provide code examples which produce the data to +be analyzed. Most of the applications we will discuss deal with +small data sets (less than a terabyte of information) and can easily +be analyzed and tested on standard off the shelf laptops you find in general +grocery stores. + +

        Data handling, machine learning and ethical aspects

        + +

        +In most of the cases we will study, we will either generate the data +to analyze ourselves (both for supervised learning and unsupervised +learning) or we will recur again and again to data present in say +scikit-learn or tensorflow. Many of the examples we end up +dealing with are from a privacy and data protection point of view, +rather inoccuous and boring results of numerical +calculations. However, this does not hinder us from developing a sound +ethical attitude to the data we use, how we analyze the data and how +we handle the data. + +

        +The most immediate and simplest possible ethical aspects deal with our +approach to the scientific process. Nowadays, with version control +software like Git and various online +repositories like Github, +Gitlab etc, we can easily make our codes +and data sets we have used, freely and easily accessible to a wider +community. This helps us almost automagically in making our science +reproducible. The large open-source development communities involved +in say Scikit-learn, +Tensorflow, +PyTorch and Keras, are +all excellent examples of this. The codes can be tested and improved +upon continuosly, helping thereby our scientific community at large in +developing data analysis and machine learning tools. It is much +easier today to gain traction and acceptance for making your science +reproducible. From a societal stand, this is an important element +since many of the developers are employees of large public institutions like +universities and research labs. Our taxpayer do deserve to get +something back for their bucks. + +

        +However, this more mechanical aspect of the ethics of science (in +particular the reproducibility of scientific results) is something +which is obvious and everybody should do as part of the dialectics of +science. The fact that many scientists are not willing to share their codes or +data is detrimental to the scientific discourse. + +

        +Before we proceed, we should add a disclaimer. Even though +we may dream of computers developing some kind of higher learning +capabilities, at the end (even if the artificial intelligence +community keeps touting our ears full of fancy futuristic avenues), it is we +who end up constructing and instructing, via various algorithms, the +computers. Self-driving cars for example, rely on sofisticated +programs which take into account all possible situations a car can +encounter. In addition, extensive usage of training datas from GPS +information, maps etc, are typically fed into the software for +self-driving cars. Adding to this various sensors and cameras that +feed information to the programs, there are zillions of ethical issues +which arise from this. + +

        +For self-driving cars, where basically many of the standard machine +learning algorithms discussed here enter into the codes, at a certain +stage we have to make choices. Yes, we , the lads and lasses who wrote +a program for a specific brand of a self-driving car. As an example, +a most carmakers have as their utmost priority the security of the +driver and the accompanying passengers. A famous carmaker, which is +one of the leaders in the market of self-driving cars, had if +statements of the following type: suppose there are two obstacles in +front of you and you cannot avoid to collide with one of them. One of +the obstacles is a monstertruck while the other one is a kindergarten +class trying to cross the road. The self-driving car algo would then +opt for the hitting the small folks instead of the monstertruck, since +the likelihood of surving a collision with our future citizens, is +much higher. + +

        +This brings us leads then to serious ethical aspects. Why should we +opt for such an option? Who decides and who is entitled to make such +choices? Keep in mind that many of the algorithms you will about in +this series of lectures or hear about later, are indeed based on +simple programming instructions. And you are very likely to be one of +the people who may end up writing such a code. Thus, developing a +sound ethical attitude to what we do, an approach well beyond the +simple mechanistic one of making our science available and +reproducible, is much needed. The example of the self-driving cars is +just one of infinitely many cases where we have to make choices. When +you analyze data on economic inequalities, who guarantees that you are +not weighting some data in a particular way, perhaps because you dearly want a +specific conclusion which may support your political views? + +

        +We do not have the answers here, but we want you think over these +topics in a more overarching way. A statistical data analysis with +its dry numbers and graphs meant to guide the eye, do not necessarily +reflect the truth, whatever that is. As a scientist, and after a +university education, you are supposedly a better citizen, with an +improved critical view and understanding of the scientific method, and +perhaps some deeper understandings of the ethics of science at +large. Use these insights. Be a critical citizen. You owe it to our +societies. + +

        +To do: Add references and acknowledgements diff --git a/doc/pub/Introduction/html/Introduction.html b/doc/pub/Introduction/html/Introduction.html index 971aa236b..478112798 100644 --- a/doc/pub/Introduction/html/Introduction.html +++ b/doc/pub/Introduction/html/Introduction.html @@ -6,9 +6,9 @@ Automatically generated HTML file from DocOnce source - + -Data Analysis and Machine Learning: Representing data +Introduction to Applied Data Analysis and Machine Learning