From 8b68e6b7379186ed643256f6b9435acaee3f874d Mon Sep 17 00:00:00 2001
From: Claudio <122022571+claudio-tw@users.noreply.github.com>
Date: Fri, 8 Sep 2023 09:20:58 +0100
Subject: [PATCH] Feature selection - Recommended workflow (#50)

* Feature selection - Recommendend workflow

* print
---
 .../recommended_feature_selection_workflow.py | 70 +++++++++++++++++++
 1 file changed, 70 insertions(+)
 create mode 100644 examples/recommended_feature_selection_workflow.py

diff --git a/examples/recommended_feature_selection_workflow.py b/examples/recommended_feature_selection_workflow.py
new file mode 100644
index 0000000..6cfc22f
--- /dev/null
+++ b/examples/recommended_feature_selection_workflow.py
@@ -0,0 +1,70 @@
+import pandas as pd
+import hisel
+from arfs.feature_selection.allrelevant import Leshy
+from xgboost import XGBRegressor
+
+
+def print_results(selection, method):
+    print('\n\n##########################################################')
+    print(
+        f'The following features have been selected using {method}:')
+    print(f'{selection}')
+
+
+def main():
+    df = pd.read_csv('mydata.csv')
+    # df is expected with categorical variables already encoded as integers
+    # and possibly with continuous variables normalised.
+    # The one-dimensional target is expected to be in hte last column of the dataframe.
+    xdf = df.iloc[:, :-1]
+    yser = df.iloc[:, -1]
+
+    # Plain KSG-based selection
+    selection, scores = hisel.select.ksgmi(xdf, yser, threshold=.01)
+    print_results(selection, 'KSG')
+
+    # HISEL with parameters specification
+    categorical_search_parameters = hisel.feature_selection.SearchParameters(
+        num_permutations=1,
+        im_ratio=.03,
+        max_iter=2,
+        parallel=True,
+        random_state=None,
+    )
+    hsiclasso_parameters = hisel.feature_selection.HSICLassoParameters(
+        mi_threshold=.00001,
+        hsic_threshold=0.005,
+        batch_size=5000,
+        minibatch_size=500,
+        number_of_epochs=3,
+        use_preselection=True,
+        device=hisel.kernels.Device.CPU  # if cuda is available you can pass GPU
+    )
+    results = hisel.feature_selection.select_features(
+        xdf, yser, hsiclasso_parameters, categorical_search_parameters)
+    print_results(results.selected_features, 'HISEL')
+
+    # Selection with Boruta
+    n_estimators = 'auto'
+    importance = "native"
+    max_iter = 100
+    random_state = None
+    verbose = 0
+    keep_weak = False
+    regressor = XGBRegressor(random_state=42)
+    leshy = Leshy(
+        regressor,
+        n_estimators=n_estimators,
+        importance=importance,
+        max_iter=max_iter,
+        random_state=random_state,
+        verbose=verbose,
+        keep_weak=keep_weak,
+    )
+    leshy.fit(xdf, yser)
+    leshy_selection = leshy.selected_features_
+    print_results(leshy_selection, 'BORUTA')
+
+
+if __name__ == '__main__':
+    main()