scikit-learn · glemaitre · Jan 4, 2023 · Nov 9, 2022 · Nov 9, 2022 · Nov 14, 2022
diff --git a/doc/whats_new/v1.3.rst b/doc/whats_new/v1.3.rst
@@ -86,6 +86,12 @@ Changelog
   when the provided `sample_weight` reduces the problem to a single class in `fit`.
   :pr:`24140` by :user:`Jonathan Ohayon <Johayon>` and :user:`Chiara Marmo <cmarmo>`.
 
+:mod:`sklearn.neural_network`
+.............................
+- |Fix| :class:`neural_network.MLPClassifier` and :class:`neural_network.MLPRegressor`
+  no longer raise warnings when fitting data with feature names.
+  :pr:`24873` by :user:`Tim Head <betatim>`.
+
 :mod:`sklearn.pipeline`
 .......................
 - |Feature| :class:`pipeline.FeatureUnion` can now use indexing notation (e.g.

diff --git a/sklearn/neural_network/_multilayer_perceptron.py b/sklearn/neural_network/_multilayer_perceptron.py
@@ -23,6 +23,7 @@
 from ..base import is_classifier
 from ._base import ACTIVATIONS, DERIVATIVES, LOSS_FUNCTIONS
 from ._stochastic_optimizers import SGDOptimizer, AdamOptimizer
+from ..metrics import accuracy_score, r2_score
 from ..model_selection import train_test_split
 from ..preprocessing import LabelBinarizer
 from ..utils import gen_batches, check_random_state
@@ -178,7 +179,7 @@ def _forward_pass(self, activations):
 
         return activations
 
-    def _forward_pass_fast(self, X):
+    def _forward_pass_fast(self, X, check_input=True):
         """Predict using the trained model
 
         This is the same as _forward_pass but does not record the activations
@@ -189,12 +190,16 @@ def _forward_pass_fast(self, X):
         X : {array-like, sparse matrix} of shape (n_samples, n_features)
             The input data.
 
+        check_input : bool, default=True
+            Perform input data validation or not.
+
         Returns
         -------
         y_pred : ndarray of shape (n_samples,) or (n_samples, n_outputs)
             The decision function of the samples for each class in the model.
         """
-        X = self._validate_data(X, accept_sparse=["csr", "csc"], reset=False)
+        if check_input:
+            X = self._validate_data(X, accept_sparse=["csr", "csc"], reset=False)
 
         # Initialize first layer
         activation = X
@@ -694,7 +699,7 @@ def _fit_stochastic(
     def _update_no_improvement_count(self, early_stopping, X_val, y_val):
         if early_stopping:
             # compute validation score, use that for stopping
-            self.validation_scores_.append(self.score(X_val, y_val))
+            self.validation_scores_.append(self._score(X_val, y_val))
 
             if self.verbose:
                 print("Validation score: %f" % self.validation_scores_[-1])
@@ -1146,13 +1151,22 @@ def predict(self, X):
             The predicted classes.
         """
         check_is_fitted(self)
-        y_pred = self._forward_pass_fast(X)
+        return self._predict(X)
+
+    def _predict(self, X, check_input=True):
+        """Private predict method with optional input validation"""
+        y_pred = self._forward_pass_fast(X, check_input=check_input)
 
         if self.n_outputs_ == 1:
             y_pred = y_pred.ravel()
 
         return self._label_binarizer.inverse_transform(y_pred)
 
+    def _score(self, X, y):
+        """Private score method without input validation"""
+        # Input validation would remove feature names, so we disable it
+        return accuracy_score(y, self._predict(X, check_input=False))
+
     @available_if(lambda est: est._check_solver())
     def partial_fit(self, X, y, classes=None):
         """Update the model with a single iteration over the given data.
@@ -1574,11 +1588,21 @@ def predict(self, X):
             The predicted values.
         """
         check_is_fitted(self)
-        y_pred = self._forward_pass_fast(X)
+        return self._predict(X)
+
+    def _predict(self, X, check_input=True):
+        """Private predict method with optional input validation"""
+        y_pred = self._forward_pass_fast(X, check_input=check_input)
         if y_pred.shape[1] == 1:
             return y_pred.ravel()
         return y_pred
 
+    def _score(self, X, y):
+        """Private score method without input validation"""
+        # Input validation would remove feature names, so we disable it
+        y_pred = self._predict(X, check_input=False)
+        return r2_score(y, y_pred)
+
     def _validate_input(self, X, y, incremental, reset):
         X, y = self._validate_data(
             X,

diff --git a/sklearn/neural_network/tests/test_mlp.py b/sklearn/neural_network/tests/test_mlp.py
@@ -894,6 +894,27 @@ def test_mlp_loading_from_joblib_partial_fit(tmp_path):
     assert_allclose(predicted_value, fine_tune_target, rtol=1e-4)
 
 
+@pytest.mark.parametrize("Estimator", [MLPClassifier, MLPRegressor])
+def test_preserve_feature_names(Estimator):
+    """Check that feature names are preserved when early stopping is enabled.
+
+    Feature names are required for consistency checks during scoring.
+
+    Non-regression test for gh-24846
+    """
+    pd = pytest.importorskip("pandas")
+    rng = np.random.RandomState(0)
+
+    X = pd.DataFrame(data=rng.randn(10, 2), columns=["colname_a", "colname_b"])
+    y = pd.Series(data=np.full(10, 1), name="colname_y")
+
+    model = Estimator(early_stopping=True, validation_fraction=0.2)
+
+    with warnings.catch_warnings():
+        warnings.simplefilter("error", UserWarning)
+        model.fit(X, y)
+
+
 @pytest.mark.parametrize("MLPEstimator", [MLPClassifier, MLPRegressor])
 def test_mlp_warm_start_with_early_stopping(MLPEstimator):
     """Check that early stopping works with warm start."""