Fix ColumnTransformer in parallel with joblib's auto memmapping

doc/whats_new/v1.5.rst

@@ -145,6 +145,10 @@ Changelog
   being explicitly set as well.
   :pr:`28483` by :user:`Stefanie Senger <StefanieSenger>`.
 
+- |Fix| Fixed an bug in :class:`compose.ColumnTransformer` with `n_jobs > 1`, where the
+  intermediate selected columns were passed to the transformers as read-only arrays.
+  :pr:`28822` by :user:`Jérémie du Boisberranger <jeremiedbb>`.
+
 :mod:`sklearn.cross_decomposition`
 ..................................
 

sklearn/compose/_column_transformer.py

@@ -18,7 +18,7 @@
 from ..base import TransformerMixin, _fit_context, clone
 from ..pipeline import _fit_transform_one, _name_estimators, _transform_one
 from ..preprocessing import FunctionTransformer
-from ..utils import Bunch, _safe_indexing
+from ..utils import Bunch
 from ..utils._estimator_html_repr import _VisualBlock
 from ..utils._indexing import _get_column_indices
 from ..utils._metadata_requests import METHODS
@@ -389,7 +389,7 @@ def set_params(self, **kwargs):
 
     def _iter(self, fitted, column_as_labels, skip_drop, skip_empty_columns):
         """
-        Generate (name, trans, column, weight) tuples.
+        Generate (name, trans, columns, weight) tuples.
 
 
         Parameters
@@ -794,7 +794,7 @@ def _call_func_on_transformers(self, X, y, func, column_as_labels, routed_params
         )
         try:
             jobs = []
-            for idx, (name, trans, column, weight) in enumerate(transformers, start=1):
+            for idx, (name, trans, columns, weight) in enumerate(transformers, start=1):
                 if func is _fit_transform_one:
                     if trans == "passthrough":
                         output_config = _get_output_config("transform", self)
@@ -813,9 +813,10 @@ def _call_func_on_transformers(self, X, y, func, column_as_labels, routed_params
                 jobs.append(
                     delayed(func)(
                         transformer=clone(trans) if not fitted else trans,
-                        X=_safe_indexing(X, column, axis=1),
+                        X=X,
                         y=y,
                         weight=weight,
+                        columns=columns,
                         **extra_args,
                         params=routed_params[name],
                     )

sklearn/compose/tests/test_column_transformer.py

@@ -6,6 +6,7 @@
 import re
 import warnings
 
+import joblib
 import numpy as np
 import pytest
 from numpy.testing import assert_allclose
@@ -36,7 +37,7 @@
     assert_almost_equal,
     assert_array_equal,
 )
-from sklearn.utils.fixes import CSR_CONTAINERS
+from sklearn.utils.fixes import CSR_CONTAINERS, parse_version
 
 
 class Trans(TransformerMixin, BaseEstimator):
@@ -2447,6 +2448,30 @@ def test_column_transformer_error_with_duplicated_columns(dataframe_lib):
         transformer.fit_transform(df)
 
 
+@pytest.mark.skipif(
+    parse_version(joblib.__version__) < parse_version("1.3"),
+    reason="requires joblib >= 1.3",
+)
+def test_column_transformer_auto_memmap():
+    """Check that ColumnTransformer works in parallel with joblib's auto-memmapping.
+
+    non-regression test for issue #28781
+    """
+    X = np.random.RandomState(0).uniform(size=(3, 4))
+
+    scaler = StandardScaler(copy=False)
+
+    transformer = ColumnTransformer(
+        transformers=[("scaler", scaler, [0])],
+        n_jobs=2,
+    )
+
+    with joblib.parallel_backend("loky", max_nbytes=1):
+        Xt = transformer.fit_transform(X)
+
+    assert_allclose(Xt, StandardScaler().fit_transform(X[:, [0]]))
+
+
 # Metadata Routing Tests
 # ======================
 

sklearn/pipeline.py

@@ -19,7 +19,7 @@
 from .base import TransformerMixin, _fit_context, clone
 from .exceptions import NotFittedError
 from .preprocessing import FunctionTransformer
-from .utils import Bunch
+from .utils import Bunch, _safe_indexing
 from .utils._estimator_html_repr import _VisualBlock
 from .utils._metadata_requests import METHODS
 from .utils._param_validation import HasMethods, Hidden
@@ -1258,7 +1258,7 @@ def make_pipeline(*steps, memory=None, verbose=False):
     return Pipeline(_name_estimators(steps), memory=memory, verbose=verbose)
 
 
-def _transform_one(transformer, X, y, weight, params):
+def _transform_one(transformer, X, y, weight, columns=None, params=None):
     """Call transform and apply weight to output.
 
     Parameters
@@ -1275,11 +1275,17 @@ def _transform_one(transformer, X, y, weight, params):
     weight : float
         Weight to be applied to the output of the transformation.
 
+    columns : str, array-like of str, int, array-like of int, array-like of bool, slice
+        Columns to select before transforming.
+
     params : dict
         Parameters to be passed to the transformer's ``transform`` method.
 
         This should be of the form ``process_routing()["step_name"]``.
     """
+    if columns is not None:
+        X = _safe_indexing(X, columns, axis=1)
+
     res = transformer.transform(X, **params.transform)
     # if we have a weight for this transformer, multiply output
     if weight is None:
@@ -1288,7 +1294,14 @@ def _transform_one(transformer, X, y, weight, params):
 
 
 def _fit_transform_one(
-    transformer, X, y, weight, message_clsname="", message=None, params=None
+    transformer,
+    X,
+    y,
+    weight,
+    columns=None,
+    message_clsname="",
+    message=None,
+    params=None,
 ):
     """
     Fits ``transformer`` to ``X`` and ``y``. The transformed result is returned
@@ -1297,6 +1310,9 @@ def _fit_transform_one(
 
     ``params`` needs to be of the form ``process_routing()["step_name"]``.
     """
+    if columns is not None:
+        X = _safe_indexing(X, columns, axis=1)
+
     params = params or {}
     with _print_elapsed_time(message_clsname, message):
         if hasattr(transformer, "fit_transform"):
@@ -1792,7 +1808,7 @@ def transform(self, X, **params):
                 routed_params[name] = Bunch(transform={})
 
         Xs = Parallel(n_jobs=self.n_jobs)(
-            delayed(_transform_one)(trans, X, None, weight, routed_params[name])
+            delayed(_transform_one)(trans, X, None, weight, params=routed_params[name])
             for name, trans, weight in self._iter()
         )
         if not Xs: