scikit-learn · AYUSH27112021 · Dec 27, 2024 · Dec 27, 2024 · Dec 28, 2024 · Dec 28, 2024
diff --git a/doc/whats_new/upcoming_changes/sklearn.compose/30543.fix.rst b/doc/whats_new/upcoming_changes/sklearn.compose/30543.fix.rst
@@ -0,0 +1,6 @@
+- :class:`compose.ColumnTransformer`
+  `ColumnTransformer` now validates sparse formats for input X and 
+  converts non-subscriptable sparse formats to csr format before 
+  processing.
+  added a non-regression unit test for verification.
+  By :user:`Ayush <AYUSH27112021>`
diff --git a/sklearn/compose/_column_transformer.py b/sklearn/compose/_column_transformer.py
@@ -940,6 +940,11 @@ def fit(self, X, y=None, **params):
         -------
         self : ColumnTransformer
             This estimator.
+
+        Notes
+        -----
+        If `X` is a sparse matrix and not indexable by default, it will be
+        internally converted to the CSR format to avoid errors during indexing.
         """
         _raise_for_params(params, self, "fit")
         # we use fit_transform to make sure to set sparse_output_ (for which we
@@ -980,6 +985,11 @@ def fit_transform(self, X, y=None, **params):
             sum of n_components (output dimension) over transformers. If
             any result is a sparse matrix, everything will be converted to
             sparse matrices.
+
+        Notes
+        -----
+        If `X` is a sparse matrix and not indexable by default, it will be
+        internally converted to the CSR format to avoid errors during indexing.
         """
         _raise_for_params(params, self, "fit_transform")
         _check_feature_names(self, X, reset=True)
@@ -1333,14 +1343,19 @@ def __sklearn_tags__(self):
 
 
 def _check_X(X):
-    """Use check_array only when necessary, e.g. on lists and other non-array-likes."""
-    if (
-        (hasattr(X, "__array__") and hasattr(X, "shape"))
-        or hasattr(X, "__dataframe__")
-        or sparse.issparse(X)
-    ):
+    """
+    Use `check_array` only when necessary, e.g., on lists and other non-array-likes,
+    Converts non-subscriptable sparse formats to CSR format.
+
+    """
+    if hasattr(X, "__array__") and hasattr(X, "shape") or hasattr(X, "__dataframe__"):
         return X
-    return check_array(X, ensure_all_finite="allow-nan", dtype=object)
+    elif sparse.issparse(X):
+        if X.format not in ("csr", "csc", "lil", "dok"):
+            return sparse.csr_matrix(X)
+        return X
+    else:
+        return check_array(X, ensure_all_finite="allow-nan", dtype=object)
 
 
 def _is_empty_column_selection(column):

diff --git a/sklearn/compose/tests/test_column_transformer.py b/sklearn/compose/tests/test_column_transformer.py
@@ -2665,6 +2665,34 @@ def test_column_transformer_auto_memmap():
     assert_allclose(Xt, StandardScaler().fit_transform(X[:, [0]]))
 
 
+def test_column_transformer_with_dia_coo_sparse():
+    # Non-regression test for:
+    # https://github.com/scikit-learn/scikit-learn/issues/30275
+    # Tests if ColumnTransformer can handle DIA and COO sparse matrices.
+
+    X_dense = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
+    X_dia = sparse.dia_matrix(X_dense)
+    X_coo = sparse.coo_matrix(X_dense)
+
+    transformer = ColumnTransformer(
+        transformers=[
+            ("scaler", StandardScaler(with_mean=False), [0]),
+            ("passthrough", "passthrough", [1, 2]),
+        ]
+    )
+
+    transformer.fit(X_dia)
+    transformed_dia = transformer.transform(X_dia)
+
+    transformer.fit(X_coo)
+    transformed_coo = transformer.transform(X_coo)
+
+    assert transformed_dia.shape[0] == 3  # number of rows should be preserved
+    assert np.count_nonzero(transformed_dia) >= 1  # one non-zero element should exist
+    assert transformed_dia.shape[0] == 3
+    assert np.count_nonzero(transformed_coo) >= 1
+
+
 # Metadata Routing Tests
 # ======================