Skip to content

Commit

Permalink
Fix figure paths
Browse files Browse the repository at this point in the history
  • Loading branch information
@ludwigbothmann @github-actions
ludwigbothmann authored and github-actions[bot] committed Nov 21, 2024
1 parent 0323a37 commit 32b23ce
Showing 1 changed file with 34 additions and 34 deletions.
68 changes: 34 additions & 34 deletions slides/04-multivariate-first-order/slides-multivar-first-order-12-comparison.tex
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,8 +43,8 @@
\vspace{-0.4cm}
GD with medium $\alpha=2\cdot10^{-3}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_coef_med.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_coef_med.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -56,8 +56,8 @@
\vspace{-0.4cm}
GD with medium $\alpha=2\cdot10^{-3}$ and bad conditioning (corr. features):
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_coef_med_corr.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_coef_med_corr.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -71,8 +71,8 @@
\vspace{-0.4cm}
GD with (too small) $\alpha=3\cdot10^{-4}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_coef_small.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_coef_small.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -84,8 +84,8 @@
\vspace{-0.5cm}
GD with large $\alpha=1.5$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_large_lr_iters.pdf} \\
\includegraphics[width=0.7\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_reg_coef_large.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_reg_large_lr_iters.pdf} \\
\includegraphics[width=0.7\textwidth]{figure_man/simu_linmod/GD_reg_coef_large.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -100,8 +100,8 @@
\vspace{-0.4cm}
SGD with medium $\alpha=2\cdot10^{-3}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_coef_med.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_coef_med.pdf}\\
\begin{footnotesize}
%Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -114,8 +114,8 @@
\vspace{-0.4cm}
SGD with medium $\alpha=2\cdot10^{-3}$ and bad conditioning (corr. features):
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_coef_med_corr.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_coef_med_corr.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -129,8 +129,8 @@
\vspace{-0.4cm}
SGD with small $\alpha=3\cdot10^{-4}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_coef_small.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_coef_small.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand All @@ -144,8 +144,8 @@
\vspace{-0.4cm}
SGD with large $\alpha=1 \cdot 10^{-2}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_reg_coef_large.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_reg_coef_large.pdf}\\
\begin{footnotesize}
Irreducible error due to additive noise is $\sigma=1$
\end{footnotesize}
Expand Down Expand Up @@ -245,8 +245,8 @@
\vspace{-0.4cm}
GD with medium $\alpha=0.25$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_coef_med.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_coef_med.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -259,8 +259,8 @@
\vspace{-0.4cm}
GD with medium $\alpha=0.25$ and bad conditioning (corr. features):
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_coef_med_corr.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_coef_med_corr.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -273,8 +273,8 @@
\vspace{-0.4cm}
GD with small $\alpha=0.025$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_coef_small.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_coef_small.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -288,8 +288,8 @@
\vspace{-0.5cm}
GD with large $\alpha=10$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/GD_log_coef_large.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/GD_log_coef_large.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -305,8 +305,8 @@
\vspace{-0.4cm}
SGD with medium $\alpha=0.03$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_coef_med.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_med_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_coef_med.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -319,8 +319,8 @@
\vspace{-0.4cm}
SGD with medium $\alpha=5\cdot10^{-2}$ and bad conditioning (corr. features):
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_coef_med_corr.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_med_lr_corr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_coef_med_corr.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -334,8 +334,8 @@
\vspace{-0.4cm}
SGD with small $\alpha=3\cdot10^{-4}$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_coef_small.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_small_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_coef_small.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand All @@ -349,8 +349,8 @@
\vspace{-0.4cm}
SGD with large $\alpha=0.3$ and indep. features:
\begin{figure}
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{slides/04-multivariate-first-order/figure_man/simu_linmod/SGD_log_coef_large.pdf}\\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_large_lr_iters.pdf} \\
\includegraphics[width=0.8\textwidth]{figure_man/simu_linmod/SGD_log_coef_large.pdf}\\
\begin{footnotesize}
%Dashed line in test loss indicates irreducible error due to $\sigma=1$
\end{footnotesize}
Expand Down Expand Up @@ -379,7 +379,7 @@
\vspace{-0.3cm}
Mini-batch SGD with $\alpha \in \{0.01, 0.1\}$ for 100 epochs (5000 iterations):
\begin{figure}
\includegraphics[width=1.0\textwidth]{slides/04-multivariate-first-order/figure_man/simu_mnist/SGD_compar.pdf} \\
\includegraphics[width=1.0\textwidth]{figure_man/simu_mnist/SGD_compar.pdf} \\
%\begin{footnotesize}
% Irreducible error due to additive noise is $\sigma=1$
%\end{footnotesize}
Expand All @@ -400,7 +400,7 @@
\vspace{-0.2cm}
Why is it not a good idea to use GD in most DL applications? SGD is much faster. Compare runtime of mini-batch SGD (batch size=$100$) with GD (constant $\alpha=0.01$ without momentum for $t_{\text{max}}=5000$ iterations):
\begin{figure}
\includegraphics[width=1.0\textwidth]{slides/04-multivariate-first-order/figure_man/simu_mnist/SGD_GD_compar.pdf} \\
\includegraphics[width=1.0\textwidth]{figure_man/simu_mnist/SGD_GD_compar.pdf} \\
%\begin{footnotesize}
% Irreducible error due to additive noise is $\sigma=1$
%\end{footnotesize}
Expand Down

0 comments on commit 32b23ce

Please sign in to comment.