few more examples

examples/run_ranking_3MR.sh

@@ -0,0 +1,19 @@
+##########################################################################################################
+# Performing non-myopic ranking with 3MR
+##########################################################################################################
+
+# This run computes 3MR-based rankings, see repo's papers for more details.
+# hint - if unsure what parameters do, you can always run "outrank --help"
+
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-3mr \
+    --target_ranking_only True \
+    --combination_number_upper_bound 2048 \
+    --num_threads 12 \
+    --interaction_order 1 \
+    --transformers fw-transformers \
+    --output_folder ./some_output_folder \
+    --subsampling 30

examples/run_ranking_combinations.sh

@@ -0,0 +1,35 @@
+##########################################################################################################
+# Ranking of feature combinations
+##########################################################################################################
+
+# This run demonstrates how to perform "supervised combination ranking" - the process of figuring out
+# which feature combinations are potentially promising.
+# Note that this process' time is directly correlated with interaction order (higher=longer runs)
+
+# hint - if unsure what parameters do, you can always run "outrank --help"
+# Example for feature pairs
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-randomized \
+    --target_ranking_only True \
+    --interaction_order 2 \
+    --combination_number_upper_bound 2048 \
+    --num_threads 50 \
+    --output_folder ./some_output_folder \
+    --subsampling 100
+
+
+# And feature triplets. The combination_number_upper_bound bounds the number of sampled combinations (RAM controller)
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-randomized \
+    --target_ranking_only True \
+    --interaction_order 2 \
+    --combination_number_upper_bound 2048 \
+    --num_threads 50 \
+    --output_folder ./some_output_folder_triplets \
+    --subsampling 100

examples/run_ranking_pairwise.sh

@@ -0,0 +1,18 @@
+##########################################################################################################
+# Pairwise feature ranking (feature redundancy calculation)
+##########################################################################################################
+
+# This run demonstrates how to obtain "feature heatmaps" - pairwise summaries of mutual redundancy
+# Note that pairwise calculations take more time - increasing thread count is a possible mitigation
+
+# hint - if unsure what parameters do, you can always run "outrank --help"
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-randomized \
+    --target_ranking_only False \
+    --combination_number_upper_bound 2048 \
+    --num_threads 50 \
+    --output_folder ./some_output_folder \
+    --subsampling 100

examples/run_ranking_singles.sh

@@ -0,0 +1,19 @@
+##########################################################################################################
+# A very generic OutRank invocation (default). It includes visualizations and other relevant statistics. #
+##########################################################################################################
+
+# This run compares features "one-at-a-time" and summarizes, visualizes the outputs.
+# hint - if unsure what parameters do, you can always run "outrank --help"
+
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-randomized \
+    --target_ranking_only True \
+    --combination_number_upper_bound 2048 \
+    --num_threads 12 \
+    --interaction_order 1 \
+    --transformers fw-transformers \
+    --output_folder ./some_output_folder \
+    --subsampling 30

examples/run_ranking_transformations.sh

@@ -0,0 +1,20 @@
+##########################################################################################################
+# Ranking of feature transformations
+##########################################################################################################
+
+# A common and very important task is figuring out which transformations of a feature are promising.
+
+# hint - if unsure what parameters do, you can always run "outrank --help"
+# Example considering some generic transformations of features. Note that OutRank is type aware, if using formats such as ob-vw or ob-csv,
+# type-aware transformations can be produced. See e.g., https://improved-dollop-9k1wgvm.pages.github.io/outrank.html?search=ob-vw for more details on the format.
+outrank \
+    --task all \
+    --data_path $PATH_TO_YOUR_DATA \
+    --data_source csv-raw \
+    --heuristic MI-numba-randomized \
+    --target_ranking_only True \
+    --combination_number_upper_bound 2048 \
+    --num_threads 50 \
+    --transformers default \
+    --output_folder ./some_output_folder \
+    --subsampling 100