lars/FYS-STK4155
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

small update

Browse Source
This commit is contained in:
Morten Hjorth-Jensen
2022-10-31 12:57:57 +01:00
parent 52c686aec2
commit f0023dec32
9 changed files with 1195 additions and 897 deletions
Download Patch File Download Diff File Expand all files Collapse all files
doc/Articles/cs231n_2017_lecture10.pdf
BIN
View File
Binary file not shown.
doc/Articles/cs231n_2017_lecture5.pdf
BIN
View File
Binary file not shown.
doc/pub/week39/html/week39-bs.html
+1 -1
View File
@@ -418,7 +418,7 @@ MathJax.Hub.Config({
</center>
<br>
<center>
<h4>Oct 4, 2022</h4>
<h4>Oct 31, 2022</h4>
</center> <!-- date -->
<br>
doc/pub/week39/html/week39-reveal.html
+4 -4
View File
@@ -184,7 +184,7 @@ MathJax.Hub.Config({
</center>
<br>
<center>
<h4>Oct 4, 2022</h4>
<h4>Oct 31, 2022</h4>
</center> <!-- date -->
<br>
@@ -3642,11 +3642,11 @@ delta = <span style="color: #B452CD">1e-8</span>
<span style="color: #228B22"># Previous value for the outer product of gradients</span>
Previous = Giter
<span style="color: #228B22"># Accumulated gradient</span>
Giter +=gradients @ gradients.T
<span style="color: #228B22"># Giter +=gradients @ gradients.T</span>
<span style="color: #228B22"># Scaling with rho the new and the previous results</span>
Gnew = (rho*Previous+(<span style="color: #B452CD">1</span>-rho)*Giter)
Giter = (rho*Giter+(<span style="color: #B452CD">1</span>-rho)*gradients*gradients)
<span style="color: #228B22"># Taking the diagonal only and inverting</span>
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Gnew)))]
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Giter)))]
<span style="color: #228B22"># Hadamard product</span>
update = np.multiply(Ginverse,gradients)
theta -= update
doc/pub/week39/html/week39-solarized.html
+4 -4
View File
@@ -332,7 +332,7 @@ MathJax.Hub.Config({
</center>
<br>
<center>
<h4>Oct 4, 2022</h4>
<h4>Oct 31, 2022</h4>
</center> <!-- date -->
<br>
@@ -3575,11 +3575,11 @@ delta = <span style="color: #B452CD">1e-8</span>
<span style="color: #228B22"># Previous value for the outer product of gradients</span>
Previous = Giter
<span style="color: #228B22"># Accumulated gradient</span>
Giter +=gradients @ gradients.T
<span style="color: #228B22"># Giter +=gradients @ gradients.T</span>
<span style="color: #228B22"># Scaling with rho the new and the previous results</span>
Gnew = (rho*Previous+(<span style="color: #B452CD">1</span>-rho)*Giter)
Giter = (rho*Giter+(<span style="color: #B452CD">1</span>-rho)*gradients*gradients)
<span style="color: #228B22"># Taking the diagonal only and inverting</span>
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Gnew)))]
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Giter)))]
<span style="color: #228B22"># Hadamard product</span>
update = np.multiply(Ginverse,gradients)
theta -= update
doc/pub/week39/html/week39.html
+4 -4
View File
@@ -409,7 +409,7 @@ MathJax.Hub.Config({
</center>
<br>
<center>
<h4>Oct 4, 2022</h4>
<h4>Oct 31, 2022</h4>
</center> <!-- date -->
<br>
@@ -3652,11 +3652,11 @@ delta <span style="color: #666666">=</span> <span style="color: #666666">1e-8</
<span style="color: #408080; font-style: italic"># Previous value for the outer product of gradients</span>
Previous <span style="color: #666666">=</span> Giter
<span style="color: #408080; font-style: italic"># Accumulated gradient</span>
Giter <span style="color: #666666">+=</span>gradients <span style="color: #666666">@</span> gradients<span style="color: #666666">.</span>T
<span style="color: #408080; font-style: italic"># Giter +=gradients @ gradients.T</span>
<span style="color: #408080; font-style: italic"># Scaling with rho the new and the previous results</span>
Gnew <span style="color: #666666">=</span> (rho<span style="color: #666666">*</span>Previous<span style="color: #666666">+</span>(<span style="color: #666666">1-</span>rho)<span style="color: #666666">*</span>Giter)
Giter <span style="color: #666666">=</span> (rho<span style="color: #666666">*</span>Giter<span style="color: #666666">+</span>(<span style="color: #666666">1-</span>rho)<span style="color: #666666">*</span>gradients<span style="color: #666666">*</span>gradients)
<span style="color: #408080; font-style: italic"># Taking the diagonal only and inverting</span>
Ginverse <span style="color: #666666">=</span> np<span style="color: #666666">.</span>c_[eta<span style="color: #666666">/</span>(delta<span style="color: #666666">+</span>np<span style="color: #666666">.</span>sqrt(np<span style="color: #666666">.</span>diagonal(Gnew)))]
Ginverse <span style="color: #666666">=</span> np<span style="color: #666666">.</span>c_[eta<span style="color: #666666">/</span>(delta<span style="color: #666666">+</span>np<span style="color: #666666">.</span>sqrt(np<span style="color: #666666">.</span>diagonal(Giter)))]
<span style="color: #408080; font-style: italic"># Hadamard product</span>
update <span style="color: #666666">=</span> np<span style="color: #666666">.</span>multiply(Ginverse,gradients)
theta <span style="color: #666666">-=</span> update
doc/pub/week39/ipynb/ipynb-week39-src.tar.gz
BIN
View File
Binary file not shown.
doc/pub/week39/ipynb/week39.ipynb
+1179 -881
View File
File diff suppressed because one or more lines are too long
doc/src/week39/week39.do.txt
+3 -3
View File
@@ -2553,11 +2553,11 @@ for epoch in range(n_epochs):
# Previous value for the outer product of gradients
Previous = Giter
# Accumulated gradient
Giter +=gradients @ gradients.T
# Giter +=gradients @ gradients.T
# Scaling with rho the new and the previous results
Gnew = (rho*Previous+(1-rho)*Giter)
Giter = (rho*Giter+(1-rho)*gradients*gradients)
# Taking the diagonal only and inverting
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Gnew)))]
Ginverse = np.c_[eta/(delta+np.sqrt(np.diagonal(Giter)))]
# Hadamard product
update = np.multiply(Ginverse,gradients)
theta -= update