Cleanup style folder

.gitignore

@@ -1,6 +1,14 @@
+#--------------------------------------------#
+#  Things that need to exist but not in git  #
+#--------------------------------------------#
+
 nospeakermargin.tex
 speakermargin.tex
 
+#-----------------------------------------------------------------------------#
+#  TeX intermediate stuff everybody loves to hate and hates to commit to git  #
+#-----------------------------------------------------------------------------#
+
 *.pdf
 *.aux
 *.fdb_latexmk
@@ -12,6 +20,11 @@ speakermargin.tex
 *.toc
 *.vrb
 *.synctex.gz
+
+#----------------------------------------------------------#
+#  Editor-specific stuff that should generally be ignored  #
+#----------------------------------------------------------#
+
 *.DS_Store
 *.Rproj
 *.Rhistory
@@ -59,7 +72,12 @@ slides/ml-philosophy/slides-*.pdf
 slides/*/speakermargin.tex
 
 # vim swap files
-*.swp
+# http://stratus3d.com/blog/2018/06/03/stop-excluding-editor-temp-files-in-gitignore/
+[._]*.s[a-v][a-z]
+[._]*.sw[a-p]
+[._]s[a-v][a-z]
+[._]sw[a-p]
+
 # used for atom editor
 .latexcfg
 # Xournal files
@@ -72,3 +90,15 @@ NAMESPACE
 .idea/*
 *.pkl
 
+# RStudio / R in general
+*.Rproj
+*.Rhistory
+.Rproj.user
+.RData
+.Rdata
+
+#-----------------------------------#
+#  OS-specific temp/preview files   #
+#-----------------------------------#
+*.DS_Store
+

.ignore

@@ -0,0 +1,5 @@
+# Complementary to .gitignore, this file also affects tools like ripgrep
+
+slides/attic/*
+slides/*/attic
+attic

slides/cart/slides-cart-computationalaspects.tex

@@ -27,7 +27,7 @@
 \begin{columns}[T]
 \column{0.49\textwidth}
 Original data
-\begin{knitrout}\scriptsize
+\scriptsize
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 \begin{tabular}{l|r|r|r|r|r}
 \hline
@@ -38,12 +38,12 @@
 \end{tabular}
 
 
-\end{knitrout}
+
 % FIGURE SOURCE: Use picture created in rsrc/monotone_trafo.R
 \includegraphics[width = \textwidth]{figure/cart_splitcomp_1}
 \column{0.49\textwidth}
 Data with log-transformed $x$
-\begin{knitrout}\scriptsize
+\scriptsize
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 \begin{tabular}{l|r|r|r|r|r}
 \hline
@@ -54,7 +54,7 @@
 \end{tabular}
 
 
-\end{knitrout}
+
 % FIGURE SOURCE: Use picture created in rsrc/monotone_trafo.R
 \includegraphics[width = \textwidth]{figure/cart_splitcomp_2}
 \end{columns}
@@ -68,10 +68,10 @@
     $$x_j \in \{a,b,c\} \leftarrow \Np \rightarrow x_j \in \{d,e\} $$
   \end{itemize}
   \begin{figure}
-   \includegraphics[width=0.8\textwidth]{figure/tree-categorical.pdf} 
+   \includegraphics[width=0.8\textwidth]{figure/tree-categorical.pdf}
   \end{figure}
   \end{vbframe}
-  
+
   \begin{vbframe}{Categorical Features}
   \begin{itemize}
   \item A split on a categorical feature partitions the feature levels:
@@ -83,7 +83,7 @@
   \end{itemize}
 
   \end{vbframe}
-  
+
   \begin{frame}{Categorical Features}
 
 For $0-1$ responses, in each node:
@@ -94,7 +94,7 @@
   \begin{columns}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
@@ -117,12 +117,12 @@
   \begin{columns}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary2.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary2.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
@@ -142,17 +142,17 @@
   \begin{columns}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary1.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary2.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary2.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary3.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-binary3.pdf}
   \end{figure}
   \end{column}
   \end{columns}
@@ -195,17 +195,17 @@
   \begin{columns}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont1.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont1.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont2.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont2.pdf}
   \end{figure}
   \end{column}
   \begin{column}{0.33\textwidth}
   \begin{figure}
-  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont3.pdf} 
+  \includegraphics[width=0.8\textwidth]{figure/categoryplot-cont3.pdf}
   \end{figure}
   \end{column}
   \end{columns}
@@ -215,11 +215,11 @@
 
 \begin{vbframe}{Missing feature values}
   \begin{itemize}
-    \item When splits are evaluated, only observations for which the used feature is not missing are used. (This can actually bias splits towards using features with lots of missing values.) 
+    \item When splits are evaluated, only observations for which the used feature is not missing are used. (This can actually bias splits towards using features with lots of missing values.)
   \item \textbf{Surrogate splits} can deal with missing values during prediction.
   \item Surrogate splits are created during training. They define replacement splitting rules, using a different feature, that result in almost the same child nodes as the original split.
-   \item When observations are passed down the tree, % (in training or prediction), 
-   and the feature value used in a split is missing, we use the surrogate split instead to decide to which child the data should be assigned. 
+   \item When observations are passed down the tree, % (in training or prediction),
+   and the feature value used in a split is missing, we use the surrogate split instead to decide to which child the data should be assigned.
   \end{itemize}
 \end{vbframe}
 
@@ -228,7 +228,7 @@
 \item Each surrogate split is a decision stump that tries to learn the actual splitting rule
 \item Consider this tree with the primary split w.r.t. \texttt{Sepal.Length} where we perform binary classification (\texttt{setosa} vs. \texttt{virginica}):
 \begin{figure}
-\includegraphics[width=0.75\textwidth]{figure/tree-binary.pdf} 
+\includegraphics[width=0.75\textwidth]{figure/tree-binary.pdf}
 \end{figure}
 \item Our surrogate split should optimize a splitting criterion w.r.t. \texttt{Sepal.Length < 5.8}
 \end{itemize}
@@ -245,25 +245,25 @@
 \centering
 \begin{tabular}{rrrrrll}
   \hline
- & Sepal.Length & ... & Petal.Width & Species & Sepal.Length $<$ 5.8 \\ 
+ & Sepal.Length & ... & Petal.Width & Species & Sepal.Length $<$ 5.8 \\
   \hline
-1 & 5.10 & ... & 0.20 & setosa & TRUE \\ 
-  4 & 4.60 & ... & 0.20 & setosa & TRUE \\ 
-  9 & 4.40 & ... & 0.20 & setosa & TRUE \\ 
-  15 & 5.80 & ... & 0.20 & setosa & FALSE \\ 
-  18 & 5.10 & ... & 0.30 & setosa & TRUE \\ 
-  52 & 5.80 & ... & 1.90 & virginica & FALSE \\ 
-  57 & 4.90 & ... & 1.70 & virginica & TRUE \\ 
-  62 & 6.40 & ... & 1.90 & virginica & FALSE \\ 
-  77 & 6.20 & ... & 1.80 & virginica & FALSE \\ 
-  99 & 6.20 & ... & 2.30 & virginica & FALSE \\ 
+1 & 5.10 & ... & 0.20 & setosa & TRUE \\
+  4 & 4.60 & ... & 0.20 & setosa & TRUE \\
+  9 & 4.40 & ... & 0.20 & setosa & TRUE \\
+  15 & 5.80 & ... & 0.20 & setosa & FALSE \\
+  18 & 5.10 & ... & 0.30 & setosa & TRUE \\
+  52 & 5.80 & ... & 1.90 & virginica & FALSE \\
+  57 & 4.90 & ... & 1.70 & virginica & TRUE \\
+  62 & 6.40 & ... & 1.90 & virginica & FALSE \\
+  77 & 6.20 & ... & 1.80 & virginica & FALSE \\
+  99 & 6.20 & ... & 2.30 & virginica & FALSE \\
    \hline
 \end{tabular}
 \end{table}
 \item Add column that indicates whether \texttt{Sepal.Length < 5.8}
 %\item As this splitting rule is very good, we will have many instances where \texttt{Sepal.Length < 5.8} is \texttt{TRUE} and \texttt{Species} is \texttt{setosa}
-\item Fit tree of depth 1 using all features but \texttt{Sepal.Length} %used 
-to derive a split that explains 
+\item Fit tree of depth 1 using all features but \texttt{Sepal.Length} %used
+to derive a split that explains
 \texttt{Sepal.Length < 5.8} best $\Rightarrow$ surrogate split
 \item Typically, software stores the best and a few more surrogate splits
 %\item A good surrogate tries to mimic the primary split this way

slides/cart/slides-cart-treegrowing.tex

@@ -7,7 +7,7 @@
 
 \newcommand{\titlefigure}{figure_man/tree_depth1_structure.png}
 \newcommand{\learninggoals}{
-\item Understand how a tree is grown by an exhaustive search 
+\item Understand how a tree is grown by an exhaustive search
 \item Know where and how the split point is set }
 
 \title{Introduction to Machine Learning}
@@ -28,15 +28,15 @@
 \item We start with an empty tree, a root node that contains all the data.\\
 Trees are then grown by recursively applying \textbf{greedy} optimization to each node $\Np$.
 
-\item Greedy means we do an \textbf{exhaustive search}: Ideally, all possible splits of $\Np$ on all possible points $t$ for all features $x_j$ are compared in terms of their empirical risk $\risk(\Np, j, t)$. 
+\item Greedy means we do an \textbf{exhaustive search}: Ideally, all possible splits of $\Np$ on all possible points $t$ for all features $x_j$ are compared in terms of their empirical risk $\risk(\Np, j, t)$.
 
 \item The training data is then distributed to child nodes according to the optimal split and the procedure is repeated in the child nodes.
 
 \end{itemize}
 
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 
-{\centering \includegraphics[width=0.65\textwidth]{figure/tree-classif-depth1-ann.pdf} 
+{\centering \includegraphics[width=0.65\textwidth]{figure/tree-classif-depth1-ann.pdf}
 
 }
 
@@ -52,7 +52,7 @@
 
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 
-{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth1.pdf} 
+{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth1.pdf}
 
 }
 
@@ -62,13 +62,13 @@
 
 \begin{enumerate}[3]
 \item Proceed recursively for each child node:
-%Iterate over all features, and for each feature over all possible split points. 
+%Iterate over all features, and for each feature over all possible split points.
 Select best split and divide data from parent node into left and right child nodes.
 \end{enumerate}
 
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 
-{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth2.pdf} 
+{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth2.pdf}
 
 }
 
@@ -81,7 +81,7 @@
 
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 
-{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth3.pdf} 
+{\centering \includegraphics[width=0.95\textwidth]{figure/tree-classif-depth3.pdf}
 
 }
 
@@ -90,7 +90,7 @@
 
 
 \begin{vbframe}{Split placement}
-\begin{knitrout}\scriptsize
+\scriptsize
 \definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}\color{fgcolor}
 
 {\centering \includegraphics[width=0.5\textwidth]{figure/split_point.pdf}
@@ -99,7 +99,7 @@
 
 
 
-\end{knitrout}
+
 \lz
 Splits are usually placed at the mid-point of the observations they split: the large margin to the next closest observations makes better generalization on new, unseen data more likely.
 \end{vbframe}
@@ -131,7 +131,7 @@
 \begin{itemize}
 \item We take the split with lowest MCE: \texttt{Sepal.Length} = $5.5$
 \item In real life, we actually search over many more splitting points.
-Common strategies involve: a) Searching over all possible split points (exhaustive search), b) searching quantile-wise 
+Common strategies involve: a) Searching over all possible split points (exhaustive search), b) searching quantile-wise
 \item MCE is rarely used, we will cover split criteria in detail later.
 %\item We will introduce additional (better) criteria soon
 \end{itemize}
@@ -161,19 +161,19 @@
 % rownames(ordered.design) = NULL
 % kable(ordered.design, digits = 3)
 % @
-% 
+%
 % \hspace{0.5cm}
 % \column{0.7\textwidth}
 % % FIGURE SOURCE: No source
 % \includegraphics[height = 0.55\textheight]{figure_man/regression_tree}
 % \end{columns}
 % \vspace{0.5cm}
 % Data points (red) were generated from the underlying function (black):
-% 
+%
 % $ sin(4x - 4) * (2x - 2)^2 * sin(20x -4) $
-% 
+%
 % % \framebreak
-% 
+%
 % % BB: doesnt seem too useful to show this, nothing really new in here
 % % <<fig.height=5>>=
 % % regr.task = makeRegrTask(data = design, target = "y")