scikit-learn
diff --git a/‎doc/whats_new/v0.24.rst
Lines changed: 5 additions & 0 deletions b/‎doc/whats_new/v0.24.rst
Lines changed: 5 additions & 0 deletions
diff --git a/‎sklearn/ensemble/_gb.py
Lines changed: 31 additions & 0 deletions b/‎sklearn/ensemble/_gb.py
Lines changed: 31 additions & 0 deletions
diff --git a/‎sklearn/ensemble/tests/test_gradient_boosting.py
Lines changed: 14 additions & 0 deletions b/‎sklearn/ensemble/tests/test_gradient_boosting.py
Lines changed: 14 additions & 0 deletions
@@ -199,6 +199,11 @@ Changelog
   :class:`ensemble.GradientBoostingRegressor` and returns `1`.
   :pr:`17702` by :user:`Simona Maggio <simonamaggio>`.
 
+- |API|: Mean absolute error ('mae') is now deprecated for the parameter
+  ``criterion`` in :class:`ensemble.GradientBoostingRegressor` and
+  :class:`ensemble.GradientBoostingClassifier`.
+  :pr:`18326` by :user:`Madhura Jayaratne <madhuracj>`.
+
 :mod:`sklearn.exceptions`
 .........................
 
 
@@ -358,6 +358,10 @@ def _check_initialized(self):
         """Check that the estimator is initialized, raising an error if not.&
8000
quot;""
         check_is_fitted(self)
 
+    @abstractmethod
+    def _warn_mae_for_criterion(self):
+        pass
+
     def fit(self, X, y, sample_weight=None, monitor=None):
         """Fit the gradient boosting model.
 
@@ -393,6 +397,10 @@ def fit(self, X, y, sample_weight=None, monitor=None):
         -------
         self : object
         """
+        if self.criterion == 'mae':
+            # TODO: This should raise an error from 0.26
+            self._warn_mae_for_criterion()
+
         # if not warmstart - clear the estimator state
         if not self.warm_start:
             self._clear_state()
@@ -802,6 +810,10 @@ class GradientBoostingClassifier(ClassifierMixin, BaseGradientBoosting):
         some cases.
 
         .. versionadded:: 0.18
+        .. deprecated:: 0.24
+            `criterion='mae'` is deprecated and will be removed in version
+            0.26. Use `criterion='friedman_mse'` or `'mse'` instead, as trees
+            should use a least-square criterion in Gradient Boosting.
 
     min_samples_split : int or float, default=2
         The minimum number of samples required to split an internal node:
@@ -1102,6 +1114,14 @@ def _validate_y(self, y, sample_weight):
         self.n_classes_ = self._n_classes
         return y
 
+    def _warn_mae_for_criterion(self):
+        # TODO: This should raise an error from 0.26
+        warnings.warn("criterion='mae' was deprecated in version 0.24 and "
+                      "will be removed in version 0.26. Use "
+                      "criterion='friedman_mse' or 'mse' instead, as trees "
+                      "should use a least-square criterion in Gradient "
+                      "Boosting.", FutureWarning)
+
     def decision_function(self, X):
         """Compute the decision function of ``X``.
 
@@ -1320,6 +1340,10 @@ class GradientBoostingRegressor(RegressorMixin, BaseGradientBoosting):
         some cases.
 
         .. versionadded:: 0.18
+        .. deprecated:: 0.24
+            `criterion='mae'` is deprecated and will be removed in version
+            0.26. The correct way of minimizing the absolute error is to use
+            `loss='lad'` instead.
 
     min_samples_split : int or float, default=2
         The minimum number of samples required to split an internal node:
@@ -1601,6 +1625,13 @@ def _validate_y(self, y, sample_weight=None):
             y = y.astype(DOUBLE)
         return y
 
+    def _warn_mae_for_criterion(self):
+        # TODO: This should raise an error from 0.26
+        warnings.warn("criterion='mae' was deprecated in version 0.24 and "
+                      "will be removed in version 0.26. The correct way of "
+                      "minimizing the absolute error is to use loss='lad' "
+                      "instead.", FutureWarning)
+
     def predict(self, X):
         """Predict regression target for X.
 
 
@@ -1333,3 +1333,17 @@ def test_attr_error_raised_if_not_fitted():
     )
     with pytest.raises(AttributeError, match=msg):
         gbr.n_classes_
+
+
+# TODO: Update in 0.26 to check for the error raised
+@pytest.mark.parametrize('estimator', [
+    GradientBoostingClassifier(criterion='mae'),
+    GradientBoostingRegressor(criterion='mae')
+])
+def test_criterion_mae_deprecation(estimator):
+    # checks whether a deprecation warning is issues when criterion='mae'
+    # is used.
+    msg = ("criterion='mae' was deprecated in version 0.24 and "
+           "will be removed in version 0.26.")
+    with pytest.warns(FutureWarning, match=msg):
+        estimator.fit(X, y)