FIX Fix ExtraTreeRegressor missing data handling (#30318)

scikit-learn · Nov 22, 2024 · 27a903b · 27a903b
1 parent 32a228d
commit 27a903b
Show file tree

Hide file tree

Showing 4 changed files with 23 additions and 6 deletions.
diff --git a/doc/whats_new/upcoming_changes/sklearn.tree/27966.feature.rst b/doc/whats_new/upcoming_changes/sklearn.tree/27966.feature.rst
@@ -2,4 +2,4 @@
   support missing-values in the data matrix ``X``. Missing-values are handled by
   randomly moving all of the samples to the left, or right child node as the tree is
   traversed.
-  By :user:`Adam Li <adam2392>`
+  By :user:`Adam Li <adam2392>` and :user:`Loïc Estève <lesteve>`
diff --git a/doc/whats_new/upcoming_changes/sklearn.tree/30318.feature.rst b/doc/whats_new/upcoming_changes/sklearn.tree/30318.feature.rst
@@ -0,0 +1,5 @@
+- :class:`tree.ExtraTreeClassifier` and :class:`tree.ExtraTreeRegressor` now
+  support missing-values in the data matrix ``X``. Missing-values are handled by
+  randomly moving all of the samples to the left, or right child node as the tree is
+  traversed.
+  By :user:`Adam Li <adam2392>` and :user:`Loïc Estève <lesteve>`
diff --git a/sklearn/tree/_partitioner.pyx b/sklearn/tree/_partitioner.pyx
@@ -194,7 +194,7 @@ cdef class DensePartitioner:
         """Partition samples for feature_values at the current_threshold."""
         cdef:
             intp_t p = self.start
-            intp_t partition_end = self.end
+            intp_t partition_end = self.end - self.n_missing
             intp_t[::1] samples = self.samples
             float32_t[::1] feature_values = self.feature_values
 

diff --git a/sklearn/tree/tests/test_tree.py b/sklearn/tree/tests/test_tree.py
@@ -2689,10 +2689,8 @@ def test_regression_tree_missing_values_toy(Tree, X, criterion):
     impurity = tree.tree_.impurity
     assert all(impurity >= 0), impurity.min()  # MSE should always be positive
 
-    # Note: the impurity matches after the first split only on greedy trees
-    if Tree is DecisionTreeRegressor:
-        # Check the impurity match after the first split
-        assert_allclose(tree.tree_.impurity[:2], tree_ref.tree_.impurity[:2])
+    # Check the impurity match after the first split
+    assert_allclose(tree.tree_.impurity[:2], tree_ref.tree_.impurity[:2])
 
     # Find the leaves with a single sample where the MSE should be 0
     leaves_idx = np.flatnonzero(
@@ -2701,6 +2699,20 @@ def test_regression_tree_missing_values_toy(Tree, X, criterion):
     assert_allclose(tree.tree_.impurity[leaves_idx], 0.0)
 
 
+def test_regression_extra_tree_missing_values_toy(global_random_seed):
+    rng = np.random.RandomState(global_random_seed)
+    n_samples = 100
+    X = np.arange(n_samples, dtype=np.float64).reshape(-1, 1)
+    X[-20:, :] = np.nan
+    rng.shuffle(X)
+    y = np.arange(n_samples)
+
+    tree = ExtraTreeRegressor(random_state=global_random_seed, max_depth=5).fit(X, y)
+
+    impurity = tree.tree_.impurity
+    assert all(impurity >= 0), impurity  # MSE should always be positive
+
+
 def test_classification_tree_missing_values_toy():
     """Check that we properly handle missing values in clasification trees using a toy
     dataset.