Swap TransformerMixin inheritance by sklearn’s SelectorMixin (#145)

* Adapt BorutaPy to SelectorMixin interface

* Test full SelectorMixin API

Author: nworb-cire

boruta/boruta_py.py

@@ -12,11 +12,13 @@
 import numpy as np
 import scipy as sp
 from sklearn.utils import check_random_state, check_X_y
-from sklearn.base import TransformerMixin, BaseEstimator
+from sklearn.base import BaseEstimator
+from sklearn.feature_selection import SelectorMixin
+from sklearn.utils.validation import check_is_fitted
 import warnings
 
 
-class BorutaPy(BaseEstimator, TransformerMixin):
+class BorutaPy(BaseEstimator, SelectorMixin):
     """
     Improved Python implementation of the Boruta R package.
 
@@ -287,11 +289,19 @@ def _fit(self, X, y):
         # check input params
         self._check_params(X, y)
 
+        feature_names = getattr(X, "columns", None)
+        if feature_names is not None:
+            self.feature_names_in_ = np.asarray(feature_names, dtype=object)
+        else:
+            self.feature_names_in_ = None
+
         if not isinstance(X, np.ndarray):
             X = self._validate_pandas_input(X) 
         if not isinstance(y, np.ndarray):
             y = self._validate_pandas_input(y)
 
+        self.n_features_in_ = X.shape[1]
+
         self.random_state = check_random_state(self.random_state)
 
         early_stopping = False
@@ -465,6 +475,10 @@ def _set_n_estimators(self, n_estimators):
             )
         return self
 
+    def _get_support_mask(self):
+        check_is_fitted(self, 'support_')
+        return self.support_
+
     def _get_tree_num(self, n_feat):
         depth = None
         try:

boruta/test/test_boruta.py

@@ -2,6 +2,7 @@
 import pandas as pd
 import pytest
 from sklearn.ensemble import RandomForestClassifier
+from sklearn.exceptions import NotFittedError
 from sklearn.tree import DecisionTreeClassifier, ExtraTreeClassifier
 
 from boruta import BorutaPy
@@ -68,6 +69,62 @@ def test_dataframe_is_returned(Xy):
     assert isinstance(bt.transform(X_df, return_df=True), pd.DataFrame)
 
 
+def test_selector_mixin_get_support_requires_fit():
+    bt = BorutaPy(RandomForestClassifier())
+    with pytest.raises(NotFittedError):
+        bt.get_support()
+
+
+def test_selector_mixin_get_support_matches_mask(Xy):
+    X, y = Xy
+    bt = BorutaPy(RandomForestClassifier())
+    bt.fit(X, y)
+
+    assert np.array_equal(bt.get_support(), bt.support_)
+    assert np.array_equal(bt.get_support(indices=True),
+                          np.where(bt.support_)[0])
+
+
+def test_selector_mixin_inverse_transform_restores_selected_features(Xy):
+    X, y = Xy
+    bt = BorutaPy(RandomForestClassifier())
+    bt.fit(X, y)
+
+    X_selected = bt.transform(X)
+    X_reconstructed = bt.inverse_transform(X_selected)
+
+    assert X_reconstructed.shape == X.shape
+    assert np.allclose(X_reconstructed[:, bt.support_], X[:, bt.support_])
+
+    if (~bt.support_).any():
+        assert np.allclose(X_reconstructed[:, ~bt.support_], 0)
+
+
+def test_selector_mixin_get_feature_names_out_requires_fit():
+    bt = BorutaPy(RandomForestClassifier())
+    with pytest.raises(NotFittedError):
+        bt.get_feature_names_out()
+
+
+def test_selector_mixin_get_feature_names_out_returns_selected_names(Xy):
+    X, y = Xy
+    bt = BorutaPy(RandomForestClassifier())
+    bt.fit(X, y)
+
+    expected_default = np.array([f"x{i}" for i in np.where(bt.support_)[0]])
+    assert np.array_equal(bt.get_feature_names_out(), expected_default)
+
+    custom_names = np.array([f"feature_{i}" for i in range(X.shape[1])])
+    selected_names = bt.get_feature_names_out(custom_names)
+    assert np.array_equal(selected_names, custom_names[bt.support_])
+
+    columns = [f"col_{i}" for i in range(X.shape[1])]
+    X_df = pd.DataFrame(X, columns=columns)
+    bt_df = BorutaPy(RandomForestClassifier())
+    bt_df.fit(X_df, y)
+    assert np.array_equal(bt_df.get_feature_names_out(), np.array(columns)[bt_df.support_])
+
+
 @pytest.mark.parametrize("tree", [ExtraTreeClassifier(), DecisionTreeClassifier()])
 def test_boruta_with_decision_trees(tree, Xy):
     msg = (
@@ -80,4 +137,4 @@ def test_boruta_with_decision_trees(tree, Xy):
     with pytest.raises(ValueError) as record:
         bt.fit(X, y)
 
-    assert str(record.value) == msg
+    assert str(record.value) == msg