MAINT Deprecate scoring='max_error' and replace it by scoring='neg_max_error' (scikit-learn#29462)

Co-authored-by: Loïc Estève <loic.esteve@ymail.com>

Author: artificialfintelligence

doc/modules/model_evaluation.rst

@@ -92,7 +92,7 @@ Scoring                                Function
 
 **Regression**
 'explained_variance'                   :func:`metrics.explained_variance_score`
-'max_error'                            :func:`metrics.max_error`
+'neg_max_error'                        :func:`metrics.max_error`
 'neg_mean_absolute_error'              :func:`metrics.mean_absolute_error`
 'neg_mean_squared_error'               :func:`metrics.mean_squared_error`
 'neg_root_mean_squared_error'          :func:`metrics.root_mean_squared_error`

doc/whats_new/v1.6.rst

@@ -219,6 +219,10 @@ Changelog
   :pr:`29210` by :user:`Marc Torrellas Socastro <marctorsoc>` and
   :user:`Stefanie Senger <StefanieSenger>`.
 
+- |API| scoring="neg_max_error" should be used instead of
+  scoring="max_error" which is now deprecated.
+  :pr:`29462` by :user:`Farid "Freddie" Taba <artificialfintelligence>`.
+
 :mod:`sklearn.model_selection`
 ..............................
 

sklearn/metrics/_scorer.py

@@ -219,6 +219,8 @@ def __init__(self, score_func, sign, kwargs, response_method="predict"):
         self._sign = sign
         self._kwargs = kwargs
         self._response_method = response_method
+        # TODO (1.8): remove in 1.8 (scoring="max_error" has been deprecated in 1.6)
+        self._deprecation_msg = None
 
     def _get_pos_label(self):
         if "pos_label" in self._kwargs:
@@ -270,6 +272,12 @@ def __call__(self, estimator, X, y_true, sample_weight=None, **kwargs):
         score : float
             Score function applied to prediction of estimator on X.
         """
+        # TODO (1.8): remove in 1.8 (scoring="max_error" has been deprecated in 1.6)
+        if self._deprecation_msg is not None:
+            warnings.warn(
+                self._deprecation_msg, category=DeprecationWarning, stacklevel=2
+            )
+
         _raise_for_params(kwargs, self, None)
 
         _kwargs = copy.deepcopy(kwargs)
@@ -420,7 +428,12 @@ def get_scorer(scoring):
     """
     if isinstance(scoring, str):
         try:
-            scorer = copy.deepcopy(_SCORERS[scoring])
+            if scoring == "max_error":
+                # TODO (1.8): scoring="max_error" has been deprecated in 1.6,
+                # remove in 1.8
+                scorer = max_error_scorer
+            else:
+                scorer = copy.deepcopy(_SCORERS[scoring])
         except KeyError:
             raise ValueError(
                 "%r is not a valid scoring value. "
@@ -758,7 +771,15 @@ def make_scorer(
 # Standard regression scores
 explained_variance_scorer = make_scorer(explained_variance_score)
 r2_scorer = make_scorer(r2_score)
+neg_max_error_scorer = make_scorer(max_error, greater_is_better=False)
 max_error_scorer = make_scorer(max_error, greater_is_better=False)
+# TODO (1.8): remove in 1.8 (scoring="max_error" has been deprecated in 1.6)
+deprecation_msg = (
+    "Scoring method max_error was renamed to "
+    "neg_max_error in version 1.6 and will "
+    "be removed in 1.8."
+)
+max_error_scorer._deprecation_msg = deprecation_msg
 neg_mean_squared_error_scorer = make_scorer(mean_squared_error, greater_is_better=False)
 neg_mean_squared_log_error_scorer = make_scorer(
     mean_squared_log_error, greater_is_better=False
@@ -867,7 +888,7 @@ def negative_likelihood_ratio(y_true, y_pred):
 _SCORERS = dict(
     explained_variance=explained_variance_scorer,
     r2=r2_scorer,
-    max_error=max_error_scorer,
+    neg_max_error=neg_max_error_scorer,
     matthews_corrcoef=matthews_corrcoef_scorer,
     neg_median_absolute_error=neg_median_absolute_error_scorer,
     neg_mean_absolute_error=neg_mean_absolute_error_scorer,

sklearn/metrics/tests/test_score_objects.py

@@ -78,7 +78,7 @@
     "mean_absolute_percentage_error",
     "mean_squared_error",
     "median_absolute_error",
-    "max_error",
+    "neg_max_error",
     "neg_mean_poisson_deviance",
     "neg_mean_gamma_deviance",
 ]
@@ -706,6 +706,16 @@ def test_scoring_is_not_metric():
         check_scoring(KMeans(), scoring=cluster_module.rand_score)
 
 
+def test_deprecated_scorer():
+    X, y = make_regression(n_samples=10, n_features=1, random_state=0)
+    X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
+    reg = DecisionTreeRegressor()
+    reg.fit(X_train, y_train)
+    deprecated_scorer = get_scorer("max_error")
+    with pytest.warns(DeprecationWarning):
+        deprecated_scorer(reg, X_test, y_test)
+
+
 @pytest.mark.parametrize(
     (
         "scorers,expected_predict_count,"