FIX RidgeCV works with multioutput and sample-weight non-default score (#29877)

glemaitre · web-flow · commit 25bc29ab9703 · 2024-09-18T17:49:03.000+02:00
diff --git a/sklearn/linear_model/_ridge.py b/sklearn/linear_model/_ridge.py
@@ -2270,12 +2270,7 @@ def _score(self, *, predictions, y, n_y, scorer, score_params):
                     ]
                 )
             else:
-                _score = scorer(
-                    identity_estimator,
-                    predictions.ravel(),
-                    y.ravel(),
-                    **score_params,
-                )
+                _score = scorer(identity_estimator, predictions, y, **score_params)
 
         return _score
 
diff --git a/sklearn/linear_model/tests/test_ridge.py b/sklearn/linear_model/tests/test_ridge.py
@@ -2252,6 +2252,32 @@ def test_ridge_cv_values_deprecated():
         ridge.cv_values_
 
 
+def test_ridge_cv_multioutput_sample_weight(global_random_seed):
+    """Check that `RidgeCV` works properly with multioutput and sample_weight
+    when `scoring != None`.
+
+    We check the error reported by the RidgeCV is close to a naive LOO-CV using a
+    Ridge estimator.
+    """
+    X, y = make_regression(n_targets=2, random_state=global_random_seed)
+    sample_weight = np.ones(shape=(X.shape[0],))
+
+    ridge_cv = RidgeCV(scoring="neg_mean_squared_error", store_cv_results=True)
+    ridge_cv.fit(X, y, sample_weight=sample_weight)
+
+    cv = LeaveOneOut()
+    ridge = Ridge(alpha=ridge_cv.alpha_)
+    y_pred_loo = np.squeeze(
+        [
+            ridge.fit(X[train], y[train], sample_weight=sample_weight[train]).predict(
+                X[test]
+            )
+            for train, test in cv.split(X)
+        ]
+    )
+    assert_allclose(ridge_cv.best_score_, -mean_squared_error(y, y_pred_loo))
+
+
 # Metadata Routing Tests
 # ======================
 

Original file line number	Diff line number	Diff line change
`@@ -2270,12 +2270,7 @@ def _score(self, *, predictions, y, n_y, scorer, score_params):`
`2270`	`2270`	`]`
`2271`	`2271`	`)`
`2272`	`2272`	`else:`
`2273`		`- _score = scorer(`
`2274`		`- identity_estimator,`
`2275`		`- predictions.ravel(),`
`2276`		`- y.ravel(),`
`2277`		`- **score_params,`
`2278`		`- )`
	`2273`	`+ _score = scorer(identity_estimator, predictions, y, **score_params)`
`2279`	`2274`
`2280`	`2275`	`return _score`
`2281`	`2276`