Merge pull request #1269 from slds-lmu/classif-cleanup

classif cleanup
slds-lmu · Nov 20, 2024 · f79e08b · f79e08b
2 parents c0e9036 + 78fbf24
commit f79e08b
Show file tree

Hide file tree

Showing 112 changed files with 174 additions and 4,338 deletions.
diff --git a/exercises-pdf/ex_testexam.pdf b/exercises-pdf/ex_testexam.pdf
diff --git a/exercises-pdf/ic_cart.pdf b/exercises-pdf/ic_cart.pdf
diff --git a/exercises-pdf/ic_classification_1.pdf b/exercises-pdf/ic_classification_1.pdf
diff --git a/exercises-pdf/ic_classification_2.pdf b/exercises-pdf/ic_classification_2.pdf
diff --git a/exercises-pdf/ic_classification_extra.pdf b/exercises-pdf/ic_classification_extra.pdf
diff --git a/exercises-pdf/ic_concepts.pdf b/exercises-pdf/ic_concepts.pdf
diff --git a/exercises-pdf/ic_forests.pdf b/exercises-pdf/ic_forests.pdf
diff --git a/exercises-pdf/ic_ml-basics-concepts.pdf b/exercises-pdf/ic_ml-basics-concepts.pdf
diff --git a/exercises-pdf/ic_ml-basics.pdf b/exercises-pdf/ic_ml-basics.pdf
diff --git a/exercises-pdf/ic_nested_resampling.pdf b/exercises-pdf/ic_nested_resampling.pdf
diff --git a/exercises-pdf/ic_nn.pdf b/exercises-pdf/ic_nn.pdf
diff --git a/exercises-pdf/ic_questions.pdf b/exercises-pdf/ic_questions.pdf
diff --git a/exercises-pdf/ic_regression.pdf b/exercises-pdf/ic_regression.pdf
diff --git a/exercises-pdf/ic_roc-curve.pdf b/exercises-pdf/ic_roc-curve.pdf
diff --git a/exercises-pdf/ic_sol_cart.pdf b/exercises-pdf/ic_sol_cart.pdf
diff --git a/exercises-pdf/ic_sol_classification_1.pdf b/exercises-pdf/ic_sol_classification_1.pdf
diff --git a/exercises-pdf/ic_sol_classification_2.pdf b/exercises-pdf/ic_sol_classification_2.pdf
diff --git a/exercises-pdf/ic_sol_forests.pdf b/exercises-pdf/ic_sol_forests.pdf
diff --git a/exercises-pdf/ic_sol_nested_resampling.pdf b/exercises-pdf/ic_sol_nested_resampling.pdf
diff --git a/exercises-pdf/ic_sol_nn.pdf b/exercises-pdf/ic_sol_nn.pdf
diff --git a/exercises-pdf/ic_sol_regression.pdf b/exercises-pdf/ic_sol_regression.pdf
diff --git a/exercises-pdf/ic_sol_train-test.pdf b/exercises-pdf/ic_sol_train-test.pdf
diff --git a/exercises-pdf/ic_sol_tuning-resampling.pdf b/exercises-pdf/ic_sol_tuning-resampling.pdf
diff --git a/exercises-pdf/ic_train-test.pdf b/exercises-pdf/ic_train-test.pdf
diff --git a/exercises-pdf/ic_tuning-resampling.pdf b/exercises-pdf/ic_tuning-resampling.pdf
diff --git a/exercises-pdf/ic_tuning_r.pdf b/exercises-pdf/ic_tuning_r.pdf
diff --git a/exercises-pdf/ic_usecase.pdf b/exercises-pdf/ic_usecase.pdf
diff --git a/exercises-pdf/sol_testexam.pdf b/exercises-pdf/sol_testexam.pdf
diff --git a/exercises/evaluation-tex/ic_concepts.Rnw b/exercises/evaluation-tex/ic_concepts.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{6}{Evaluation}
 

diff --git a/exercises/evaluation-tex/ic_questions.Rnw b/exercises/evaluation-tex/ic_questions.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{7}{Quiz Tournament}
 

diff --git a/exercises/evaluation-tex/ic_roc-curve.Rnw b/exercises/evaluation-tex/ic_roc-curve.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{7}{Evaluation -- ROC curve}
 

diff --git a/exercises/evaluation-tex/ic_sol_train-test.Rnw b/exercises/evaluation-tex/ic_sol_train-test.Rnw
@@ -5,9 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-eval.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{5}{Evaluation}
 

diff --git a/exercises/evaluation-tex/ic_train-test.Rnw b/exercises/evaluation-tex/ic_train-test.Rnw
@@ -5,9 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-eval.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{5}{Evaluation}
 

diff --git a/exercises/forests-tex/ic_forests.Rnw b/exercises/forests-tex/ic_forests.Rnw
@@ -5,10 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/ml-ensembles.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{9}{Random Forests}
 

diff --git a/exercises/forests-tex/ic_sol_forests.Rnw b/exercises/forests-tex/ic_sol_forests.Rnw
@@ -5,9 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{9}{Random Forests}
 

diff --git a/exercises/ml-basics-tex/ic_ml-basics-concepts.Rnw b/exercises/ml-basics-tex/ic_ml-basics-concepts.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{1}{ML Basics}
 

diff --git a/exercises/ml-basics-tex/ic_ml-basics.Rnw b/exercises/ml-basics-tex/ic_ml-basics.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{0}{Introduction}
 

diff --git a/exercises/nested-resampling-tex/ic_nested_resampling.Rnw b/exercises/nested-resampling-tex/ic_nested_resampling.Rnw
@@ -5,11 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
-\input{../../latex-math/ml-ensembles.tex}
-\input{../../latex-math/ml-hpo.tex}
-\input{../../latex-math/ml-eval.tex}
 
 
 \kopfic{12}{Nested Resampling}

diff --git a/exercises/nested-resampling-tex/ic_sol_nested_resampling.Rnw b/exercises/nested-resampling-tex/ic_sol_nested_resampling.Rnw
@@ -5,11 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
-\input{../../latex-math/ml-ensembles.tex}
-\input{../../latex-math/ml-hpo.tex}
-\input{../../latex-math/ml-eval.tex}
 
 
 \kopfic{12}{Nested Resampling}

diff --git a/exercises/neural-networks-tex/ic_nn.Rnw b/exercises/neural-networks-tex/ic_nn.Rnw
@@ -5,9 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{10}{Neural Networks}
 

diff --git a/exercises/neural-networks-tex/ic_sol_nn.Rnw b/exercises/neural-networks-tex/ic_sol_nn.Rnw
@@ -5,9 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{10}{Neural Networks}
 

diff --git a/exercises/supervised-classification-tex/ic_classification_1.Rnw b/exercises/supervised-classification-tex/ic_classification_1.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{3}{Supervised Classification}
 

diff --git a/exercises/supervised-classification-tex/ic_classification_2.Rnw b/exercises/supervised-classification-tex/ic_classification_2.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{4}{Supervised Classification}
 

diff --git a/exercises/supervised-classification-tex/ic_classification_extra.Rnw b/exercises/supervised-classification-tex/ic_classification_extra.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{4}
 

diff --git a/exercises/supervised-classification-tex/ic_sol_classification_1.Rnw b/exercises/supervised-classification-tex/ic_sol_classification_1.Rnw
@@ -5,8 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{3}{Supervised Classification}
 

diff --git a/exercises/supervised-classification-tex/ic_sol_classification_2.Rnw b/exercises/supervised-classification-tex/ic_sol_classification_2.Rnw
@@ -5,8 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{4}{Supervised Classification}
 

diff --git a/exercises/supervised-regression-tex/ic_regression.Rnw b/exercises/supervised-regression-tex/ic_regression.Rnw
@@ -5,8 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{2}{Supervised Regression}
 

diff --git a/exercises/supervised-regression-tex/ic_sol_regression.Rnw b/exercises/supervised-regression-tex/ic_sol_regression.Rnw
@@ -5,8 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{2}{Supervised Regression}
 

diff --git a/exercises/testexam-tex/ex_testexam.Rnw b/exercises/testexam-tex/ex_testexam.Rnw
@@ -5,9 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
-\input{../../latex-math/ml-eval.tex}
 
 \section*{I2ML -- Test exam -- WS2022/23}
 

diff --git a/exercises/testexam-tex/sol_testexam.Rnw b/exercises/testexam-tex/sol_testexam.Rnw
@@ -4,9 +4,6 @@
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/basic-ml.tex}
-\input{../../latex-math/ml-trees.tex}
 
 \section*{I2ML -- Test exam -- WS2022/23}
 

diff --git a/exercises/trees-tex/ic_cart.Rnw b/exercises/trees-tex/ic_cart.Rnw
@@ -5,9 +5,6 @@ library('knitr')
 knitr::set_parent("../../style/preamble_ueb.Rnw")
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{8}{CART}
 

diff --git a/exercises/trees-tex/ic_sol_cart.Rnw b/exercises/trees-tex/ic_sol_cart.Rnw
@@ -5,9 +5,6 @@ knitr::set_parent("../../style/preamble_ueb.Rnw")
 options(warn = -1)
 @
 
-\input{../../latex-math/basic-math.tex}
-\input{../../latex-math/ml-trees.tex}
-\input{../../latex-math/basic-ml.tex}
 
 \kopfic{8}{CART}
 

diff --git a/exercises/tuning-tex/ex_rnw/ex_tuning_Rcode.Rnw b/exercises/tuning-tex/ex_rnw/ex_tuning_Rcode.Rnw
@@ -0,0 +1,160 @@
+We want to predict the scored home runs
+(variable \texttt{runs\_scored}) from a number of characteristics of US
+baseball games.
+You can find the corresponding data on Moodle.
+
+The corresponding task is created (adapt your path if necessary) as follows:
+
+<<Start, eval=FALSE>>=
+library(data.table)
+data_baseball <- fread("baseball.csv")
+data_baseball$team <- as.factor(data_baseball$team)
+data_baseball$league <- as.factor(data_baseball$league)
+@
+
+We want to use a $k$-NN algorithm.
+However, we are not sure what number of neighbors $k$ yields the best result.
+This is why we want to use tuning to determine the best value for $k$.
+We assume that it might be somewhere in the range from 1 to 100.
+
+Furthermore, we want to use \textbf{random search} to search our defined search
+space.
+In addition, we only want to carry out the tuning a maximum of 80 times.
+
+\begin{itemize}
+  \item[1)] Think about what random search actually means with regard to the
+  search space and a suitable stopping criterion.
+\end{itemize}
+
+We still need to define which resampling method is supposed to be used
+during tuning.
+We want to use 5-fold CV and compute the MSE in each iteration to
+estimate the generalization error of the respective candidate.
+
+\begin{itemize}
+  \item[2)] Implement the resampling procedure for $n$-fold CV as an auxiliary
+  function that returns the train and test indices for each fold.
+  The user should be able to feed in the data indices (\texttt{idx}), the number
+  of \texttt{folds}, and a random \texttt{seed}.
+  <<eval=FALSE>>=
+    resample_cv <- function(idx, folds, seed = 123) {
+        # shuffle indices
+        ...
+        # prepare objects needed to store indices
+        ...
+        # CV iterations
+        for (i in seq(folds)) {
+            ...
+        }
+        # return
+        ...
+    }
+    @
+    In order to make this easier, let's break the code down to smaller building 
+    blocks:
+    \begin{itemize}
+        \item First, we shuffle the input indices (\texttt{idx}) to 
+        avoid any unintended sorting:
+        <<eval=FALSE>>=
+        # shuffle indices
+        ...
+        @
+        \item We will take this randomly sorted list of indices and cut it into 
+        non-overlapping slices representing the test sets -- the  
+        rest makes up the training set for the respective fold.
+        For this, we will move along the indices from a start index for one 
+        interval length (= number of test indices in the fold).
+        Take the largest integer number possible as \texttt{interval\_length} 
+        (for the sake of simplicity, you can ignore a potential rest resulting
+        from division here).
+        Moreover, we create an empty list object of length \texttt{folds} in 
+        which to store the indices of each fold.
+        <<eval=FALSE>>=
+        # prepare objects needed to store indices
+        start_index <- 1
+        interval_length <- ...
+        idx_list <- ...
+        @
+        \item Now we can loop through the folds and find the train and test 
+        indices for each iteration.
+        Your code should find the test indices for this fold, set the remainder 
+        as training indices, and store everything in \texttt{idx\_list}.
+        At the end of each iteration, remember to move the \texttt{start\_index}.
+        <<eval=FALSE>>=
+        # CV iterations
+        for (i in seq(folds)) {
+            # Define test and train indices
+            test_idx <- ...
+            train_idx <- ...
+            idx_list[[i]]$test <- test_idx
+            idx_list[[i]]$train <- train_idx
+            # Move start index
+            start_index <- ...
+        }
+        @
+        \item In the last step, return the list object of indices:
+        <<eval=FALSE>>=
+        # return
+        ...
+        @
+    \end{itemize}
+  \item[3)] Perform the random search with the above specifications, storing the
+  results for each candidate configuration.
+  Use the following set-up:
+    <<eval=FALSE>>=
+    # Define task and learner
+    task <- TaskRegr$new("bb", backend = data_baseball, target = "runs_scored")
+    lrn_knn = lrn("regr.kknn")
+    
+    # Define tuning settings
+    search_space <- ...
+    max_evals <- ...
+    folds <- ...
+    resampling_idx <- resample_cv(task$row_ids, folds)
+
+    # Get configuration candidates
+    set.seed(123)
+    k_candidates <- ...
+    @
+  And perform the tuning procedure as follows:
+    <<eval=FALSE>>=
+    library(mlr3)
+    library(mlr3learners)
+    # Define archive to store results
+    tuning_archive <- data.table(
+        "iteration" = seq_along(k_candidates), "k" = k_candidates, "ge" = 0
+    )
+    @
+    <<eval=FALSE>>=
+    # Perform tuning
+    for (i in tuning_archive$iteration) {
+        # Set current configuration
+        lrn_knn$param_set$values <- list(k = tuning_archive[iteration == i, k])
+        # Estimate GE via 5-CV
+        ge_est <- 0
+        for (j in 1:folds) {
+            # Train on training data in j-th fold 
+            ...
+            # Predict on test data in j-th fold
+            predictions <- ...
+            # Accumulate estimated GE
+            ge_est <- ge_est + (1 / folds) * predictions$score()
+        }
+        tuning_archive[iteration == i]$ge <- ge_est
+        
+    }
+    @
+  \item[4)] In the last step, plot the estimated generalization error for different 
+  values of $k$:
+  <<eval=FALSE>>=
+    library(ggplot2)
+    ggplot(tuning_archive, aes(x = k, y = ge)) +
+        geom_line() +
+        geom_point(
+            tuning_archive[which.min(tuning_archive$ge)], 
+            mapping = aes(x = k, y = ge),
+            col = "blue",
+            size = 5
+    )
+    @
+\end{itemize}