Keep decision_function minimal

scikit-learn · Dec 11, 2024 · b64baaa · b64baaa
1 parent 84535ef
commit b64baaa
Showing 1 changed file with 3 additions and 18 deletions.
diff --git a/sklearn/linear_model/_base.py b/sklearn/linear_model/_base.py
@@ -27,11 +27,9 @@
 from ..utils._array_api import (
     _asarray_with_order,
     _average,
-    _convert_to_numpy,
     get_namespace,
     get_namespace_and_device,
     indexing_dtype,
-    make_converter,
     supported_float_dtypes,
 )
 from ..utils._seq_dataset import (
@@ -348,16 +346,11 @@ def decision_function(self, X):
             this class would be predicted.
         """
         check_is_fitted(self)
-        # X must be in the same namespace as self.coef_; self.intercept_ is
-        # either a python float or in the same namespace as self.coef_
         xp, _ = get_namespace(X)
-        follow_X = make_converter(X)
 
         X = validate_data(self, X, accept_sparse="csr", reset=False)
-        coef = follow_X(self.coef_)
-        if coef.ndim == 2:
-            coef = coef.T
-        scores = safe_sparse_dot(X, coef, dense_output=True) + follow_X(self.intercept_)
+        coef = self.coef_.T if self.coef_.ndim == 2 else self.coef_
+        scores = safe_sparse_dot(X, coef, dense_output=True) + self.intercept_
         return (
             xp.reshape(scores, (-1,))
             if (scores.ndim > 1 and scores.shape[1] == 1)
@@ -385,15 +378,7 @@ def predict(self, X):
         else:
             indices = xp.argmax(scores, axis=1)
 
-        classes_xp, _ = get_namespace(self.classes_)
-        if classes_xp == xp:
-            return xp.take(self.classes_, indices, axis=0)
-
-        return np.take(
-            _convert_to_numpy(self.classes_, classes_xp),
-            _convert_to_numpy(indices, xp),
-            axis=0,
-        )
+        return xp.take(self.classes_, indices, axis=0)
 
     def _predict_proba_lr(self, X):
         """Probability estimation for OvR logistic regression.