scikit-learn
diff --git a/‎doc/whats_new/upcoming_changes/sklearn.tree/33354.major-feature.rst‎
Lines changed: 3 additions & 2 deletions b/‎doc/whats_new/upcoming_changes/sklearn.tree/33354.major-feature.rst‎
Lines changed: 3 additions & 2 deletions
diff --git a/‎sklearn/tree/_classes.py‎
Lines changed: 7 additions & 0 deletions b/‎sklearn/tree/_classes.py‎
Lines changed: 7 additions & 0 deletions
diff --git a/‎sklearn/tree/tests/test_split.py‎
Lines changed: 5 additions & 0 deletions b/‎sklearn/tree/tests/test_split.py‎
Lines changed: 5 additions & 0 deletions
diff --git a/‎sklearn/tree/tests/test_tree.py‎
Lines changed: 21 additions & 0 deletions b/‎sklearn/tree/tests/test_tree.py‎
Lines changed: 21 additions & 0 deletions
@@ -1,6 +1,7 @@
 - :class:`tree.DecisionTreeRegressor`, :class:`tree.DecisionTreeClassifier`
   now have native support for categorical features for binary classification and
-  single-output regression. Categorical features can be specified with the
-  `categorical_features` parameter. Up to 256 categories per features are supported.
+  single-output regression. All criteria are supported except for `'absolute_error'`.
+  Categorical features can be specified with the `categorical_features` parameter.
+  Up to 256 categories per features are supported.
   By `:user:`Adam Li <adam2392>`, `:user:`Arthur Lacote <cakedev0>`,
   :user:`Adrin Jalali <adrinjalali>` and :user:`Christian Lorentzen <lorentzenchr>`
@@ -510,6 +510,11 @@ def _fit(
                 "Categorical features are only supported for binary classification. "
                 f"Found {self.n_classes_.max()} classes."
             )
+        if has_categorical and self.criterion == "absolute_error":
+            raise ValueError(
+                "Categorical features are not supported with "
+                "criterion='absolute_error'."
+            )
 
         SPLITTERS = SPARSE_SPLITTERS if issparse(X) else DENSE_SPLITTERS
         splitter = SPLITTERS[self.splitter](
@@ -1456,6 +1461,8 @@ class DecisionTreeRegressor(RegressorMixin, BaseDecisionTree):
         represented by ``np.nan``; unknown categories at prediction time are
         also treated as missing values.
 
+        Categorical features are not supported with `criterion="absolute_error"`.
+
         .. versionadded:: 1.10
 
     Attributes
 
@@ -247,6 +247,11 @@ def test_split_impurity(
 
     if categorical and "Extra" in Tree.__name__:
         pytest.skip("Categorical features not implemented for the random splitter")
+    if categorical and criterion == "absolute_error":
+        pytest.skip(
+            "absolute_error is not supported with categorical features in "
+            "DecisionTreeRegressor"
+        )
     rng = np.random.default_rng(global_random_seed)
 
     ns = [5] * 5 + [10] * 5 + [20, 30, 50, 100]
 
@@ -2469,6 +2469,11 @@ def test_missing_values_best_splitter_on_equal_nodes_no_missing(
     criterion, categorical_features
 ):
     """Check missing values goes to correct node during predictions."""
+    if categorical_features is not None and criterion == "absolute_error":
+        pytest.skip(
+            "absolute_error is not supported with categorical features in "
+            "DecisionTreeRegressor"
+        )
     X = np.array([[0, 1, 2, 3, 8, 9, 11, 12, 15]]).T
     y = np.array([0.1, 0.2, 0.3, 0.2, 1.4, 1.4, 1.5, 1.6, 2.6])
     node_value_func = np.median if criterion == "absolute_error" else np.mean
@@ -3136,6 +3141,22 @@ def test_fit_categorical_with_monotonic_constraint(Tree):
         Tree(categorical_features=[0], monotonic_cst=[1], random_state=0).fit(X, y)
 
 
+def test_fit_categorical_with_absolute_error():
+    # Non-regression test: the categorical split-finding algorithm is not
+    # valid for criterion="absolute_error" (see gh-34578), so it should be
+    # rejected rather than silently return a possibly suboptimal split.
+    X = np.array([[0.0], [1.0], [0.0], [1.0]], dtype=np.float64)
+    y = np.array([0.0, 1.0, 0.0, 1.0])
+
+    with pytest.raises(
+        ValueError,
+        match="Categorical features are not supported with criterion='absolute_error'",
+    ):
+        DecisionTreeRegressor(categorical_features=[0], criterion="absolute_error").fit(
+            X, y
+        )
+
+
 def test_predict_sparse_int64_indices_raises():
     X = np.array([[0.0], [1.0]], dtype=np.float64)
     y = np.array([0, 1])