rename splitter -> split_func in DataOp.skb.train_test_split (#1630)

Author: jeromedockes

CHANGES.rst

@@ -28,6 +28,8 @@ Changes
   :class:`ApplyToFrame`, or ``"no_wrap"`` for no wrapping. The old values are
   deprecated and will result in an error in a future release.
   :pr:`1628` by :user:`Jérôme Dockès <jeromedockes>`.
+- The parameter ``splitter`` of :meth:`DataOp.skb.train_test_split` has been
+  renamed ``split_func``. :pr:`1630` by :user:`Jérôme Dockès <jeromedockes>`.
 
 
 Bugfixes

doc/modules/data_ops/validation/tuning_validating_data_ops.rst

@@ -102,7 +102,7 @@ which we used :meth:`.skb.mark_as_X() <DataOp.skb.mark_as_X>` and
 :meth:`.skb.mark_as_y() <DataOp.skb.mark_as_y>`: the first few steps of the
 pipeline are executed until we have a value for ``X`` and for ``y``.
 Then, those
-dataframes are split using the provided splitter function (by default
+dataframes are split using the provided split function (by default
 scikit-learn's :func:`sklearn.model_selection.train_test_split`).
 
 >>> split = pred.skb.train_test_split(shuffle=False)
@@ -134,5 +134,5 @@ And we can obtain predictions on the test part:
 >>> r2_score(test_y_true, test_pred) # doctest: +SKIP
 0.440999149220359
 
-It is possible to define a custom splitter function to use instead of
+It is possible to define a custom split function to use instead of
 :func:`sklearn.model_selection.train_test_split`.

skrub/_data_ops/_estimator.py

@@ -644,8 +644,8 @@ def train_test_split(
     environment,
     *,
     keep_subsampling=False,
-    splitter=model_selection.train_test_split,
-    **splitter_kwargs,
+    split_func=model_selection.train_test_split,
+    **split_func_kwargs,
 ):
     """Split an environment into a training an testing environments.
 
@@ -656,9 +656,9 @@ def train_test_split(
     environment = env_with_subsampling(data_op, environment, keep_subsampling)
     X, y = _compute_Xy(data_op, environment)
     if y is None:
-        X_train, X_test = splitter(X, **splitter_kwargs)
+        X_train, X_test = split_func(X, **split_func_kwargs)
     else:
-        X_train, X_test, y_train, y_test = splitter(X, y, **splitter_kwargs)
+        X_train, X_test, y_train, y_test = split_func(X, y, **split_func_kwargs)
     train_env = {**environment, X_NAME: X_train}
     test_env = {**environment, X_NAME: X_test}
     result = {

skrub/_data_ops/_skrub_namespace.py

@@ -1,5 +1,6 @@
 import pickle
 import typing
+import warnings
 
 from sklearn import model_selection
 
@@ -1495,8 +1496,8 @@ def train_test_split(
         environment=None,
         *,
         keep_subsampling=False,
-        splitter=model_selection.train_test_split,
-        **splitter_kwargs,
+        split_func=model_selection.train_test_split,
+        **split_func_kwargs,
     ):
         """Split an environment into a training an testing environments.
 
@@ -1512,12 +1513,12 @@ def train_test_split(
             :meth:`DataOp.skb.subsample`), use a subsample of the data. By
             default subsampling is not applied and all the data is used.
 
-        splitter : function, optional
+        split_func : function, optional
             The function used to split X and y once they have been computed. By
             default, :func:`~sklearn.model_selection.train_test_split` is used.
 
-        splitter_kwargs
-            Additional named arguments to pass to the splitter.
+        split_func_kwargs
+            Additional named arguments to pass to the splitting function.
 
         Returns
         -------
@@ -1563,14 +1564,24 @@ def train_test_split(
         >>> accuracy_score(split["y_test"], predictions)
         0.0
         """
+        if (splitter := split_func_kwargs.pop("splitter", None)) is not None:
+            warnings.warn(
+                (
+                    "The `splitter` parameter of `.skb.train_test_split` has been"
+                    " renamed `split_func`. Using it will raise an error in a future"
+                    " release of skrub."
+                ),
+                category=FutureWarning,
+            )
+            split_func = splitter
         if environment is None:
             environment = self.get_data()
         return train_test_split(
             self._data_op,
             environment,
             keep_subsampling=keep_subsampling,
-            splitter=splitter,
-            **splitter_kwargs,
+            split_func=split_func,
+            **split_func_kwargs,
         )
 
     def make_grid_search(self, *, fitted=False, keep_subsampling=False, **kwargs):

skrub/_data_ops/tests/test_estimators.py

@@ -422,6 +422,19 @@ def test_train_test_split(with_y):
         assert e.skb.eval() == [7, 6, 5, 4, 3, 2, 1, 0]
 
 
+def test_train_test_split_splitter_renaming():
+    # TODO remove when `splitter` is removed in 0.7.0
+    X = skrub.X(list(range(10)))
+
+    def split(X, shuffle):
+        return train_test_split(X, shuffle=shuffle)
+
+    with pytest.warns(FutureWarning, match="`splitter`.*has been renamed"):
+        assert X.skb.train_test_split(splitter=split, shuffle=False)["X_train"] == list(
+            range(7)
+        )
+
+
 def test_iter_learners():
     e = skrub.choose_from([1, 2, 3], name="c").as_data_op()
     assert [p.describe_params() for p in e.skb.iter_learners_grid()] == [