rename values for .skb.apply(how=...)

CHANGES.rst

@@ -20,9 +20,15 @@ Changes
   :func:`datasets.get_ken_table_aliases`, and :func:`datasets.get_ken_types` will be
   removed in the next release of skrub.
   :pr:`1546` by :user:`Vincent Maladiere <Vincent-Maladiere>`.
-
 - Improved error messages when a DataOp is being sent to dispatched functions.
   :pr:`1607` by :user:`Riccardo Cappuzzo<rcap107>`.
+- The accepted values for the parameter ``how`` of :meth:`DataOp.skb.apply` have
+  changed. The new values are ``"auto"`` (unchanged), ``"cols"`` to wrap the
+  transformer in :class:`ApplyToCols`, ``"frame"`` to wrap the transformer in
+  :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>`.
+
 
 Bugfixes
 --------

skrub/_data_ops/_data_ops.py

@@ -722,8 +722,8 @@ def _check_wrap_params(cols, how, allow_reject, reason):
     msg = None
     if not isinstance(cols, type(s.all())):
         msg = f"`cols` must be `all()` (the default) when {reason}"
-    elif how not in ["auto", "full_frame"]:
-        msg = f"`how` must be 'auto' (the default) or 'full_frame' when {reason}"
+    elif how not in ["auto", "no_wrap"]:
+        msg = f"`how` must be 'auto' (the default) or 'no_wrap' when {reason}"
     elif allow_reject:
         msg = f"`allow_reject` must be False (the default) when {reason}"
     if msg is not None:
@@ -754,11 +754,34 @@ def _check_estimator_type(estimator):
     )
 
 
+def _check_apply_how(how):
+    valid = ["auto", "cols", "frame", "no_wrap"]
+    if how in valid:
+        return how
+
+    # TODO remove when the old names are completely dropped in 0.7.0
+    translate = {"columnwise": "cols", "sub_frame": "frame", "full_frame": "no_wrap"}
+    if how in translate:
+        new = translate[how]
+        warnings.warn(
+            (
+                f"{how!r} has been renamed to {new!r}: use .skb.apply(how={new!r})"
+                " instead."
+            ),
+            FutureWarning,
+        )
+        return new
+
+    raise ValueError(f"`how` must be one of {valid}. Got: {how!r}")
+
+
 def _wrap_estimator(estimator, cols, how, allow_reject, X):
     """
     Wrap the estimator passed to .skb.apply in ApplyToCols or ApplyToFrame if
     needed.
     """
+    how = _check_apply_how(how)
+
     if estimator in [None, "passthrough"]:
         estimator = PassThrough()
 
@@ -767,16 +790,16 @@ def _wrap_estimator(estimator, cols, how, allow_reject, X):
     def _check(reason):
         _check_wrap_params(cols, how, allow_reject, reason)
 
-    if how == "full_frame":
-        _check("`how` is 'full_frame'")
+    if how == "no_wrap":
+        _check("`how` is 'no_wrap'")
         return estimator
     if hasattr(estimator, "predict") or not hasattr(estimator, "transform"):
         _check("`estimator` is a predictor (not a transformer)")
         return estimator
     if not sbd.is_dataframe(X):
         _check("the input is not a DataFrame")
         return estimator
-    columnwise = {"auto": "auto", "columnwise": True, "sub_frame": False}[how]
+    columnwise = {"auto": "auto", "cols": True, "frame": False}[how]
     return wrap_transformer(
         estimator, cols, allow_reject=allow_reject, columnwise=columnwise
     )
@@ -1186,7 +1209,7 @@ def eval(self, *, mode, environment):
 
 
 def _check_column_names(X):
-    # NOTE: could allow int column names when how='full_frame', prob. not worth
+    # NOTE: could allow int column names when how='no_wrap', prob. not worth
     # the added complexity.
     #
     # TODO: maybe also forbid duplicates? use a reduced version of

skrub/_data_ops/_estimator.py

@@ -249,15 +249,15 @@ def find_fitted_estimator(self, name):
         >>> learner.find_fitted_estimator("classifier")
         DummyClassifier()
 
-        Depending on the parameters passed to ``skb.apply()``, the estimator we provide
-        can be wrapped in a skrub transformer that applies it to several columns in the
-        input, or to a subset of the columns in a dataframe. In other cases it may be
-        applied without any wrapping. We provide examples for those 3 different cases
-        below.
+        Depending on the parameters passed to :meth:`DataOp.skb.apply`, the
+        estimator we provide can be wrapped in a skrub transformer that applies
+        it to several columns in the input, or to a subset of the columns in a
+        dataframe. In other cases it may be applied without any wrapping. We
+        provide examples for those 3 different cases below.
 
         Case 1: the ``StringEncoder`` is a skrub single-column transformer: it
         transforms a single column. In the learner it gets wrapped in a
-        ``skrub.ApplyToCols`` which independently fits a separate instance of the
+        :class:`ApplyToCols` which independently fits a separate instance of the
         ``StringEncoder`` to each of the columns it transforms (in this case there is
         only one column, ``'product'``). The individual transformers can be found in the
         fitted attribute ``transformers_`` which maps column names to the corresponding
@@ -269,13 +269,13 @@ def find_fitted_estimator(self, name):
         >>> encoder.transformers_['product'].vectorizer_.vocabulary_
         {' pe': 2, 'pen': 12, 'en ': 8, ' pen': 3, 'pen ': 13, ' cu': 0, 'cup': 6, 'up ': 18, ' cup': 1, 'cup ': 7, ' sp': 4, 'spo': 16, 'poo': 14, 'oon': 10, 'on ': 9, ' spo': 5, 'spoo': 17, 'poon': 15, 'oon ': 11}
 
-        This case (wrapping in ``ApplyToCols``) happens when the estimator is a skrub
+        This case (wrapping in :class:`ApplyToCols`) happens when the estimator is a skrub
         single-column transformer (it has a ``__single_column_transformer__``
-        attribute), we pass ``.skb.apply(how='columnwise')`` or we pass
+        attribute), we pass ``.skb.apply(how='cols')`` or we pass
         ``.skb.apply(allow_reject=True)``.
 
         Case 2: the ``PCA`` is a regular scikit-learn transformer. In the learner it
-        gets wrapped in a ``skrub.ApplyToFrame`` which applies it to the subset of columns
+        gets wrapped in a :class:`ApplyToFrame` which applies it to the subset of columns
         in the dataframe selected by the ``cols`` argument passed to ``.skb.apply()``.
         The fitted ``PCA`` can be found in the fitted attribute ``transformer_``.
 
@@ -287,8 +287,9 @@ def find_fitted_estimator(self, name):
         >>> pca.transformer_.mean_
         array([2020.,    4.,    4.], dtype=float32)
 
-        This case (wrapping in ``ApplyToFrame``) happens when the estimator is a
-        scikit-learn transformer but not a single-column transformer.
+        This case (wrapping in :class:`ApplyToFrame`) happens when the estimator is a
+        scikit-learn transformer but not a single-column transformer, or we
+        pass ``.skb.apply(how='frame')``.
 
         The ``DummyRegressor`` is a scikit-learn predictor. In the learner it gets
         applied directly to the input dataframe without any wrapping.
@@ -301,7 +302,7 @@ def find_fitted_estimator(self, name):
 
         This case (no wrapping) happens when the estimator is a scikit-learn predictor
         (not a transformer), the input is not a dataframe (e.g. it is a numpy array), or
-        we pass ``.skb.apply(how='full_frame')``.
+        we pass ``.skb.apply(how='no_wrap')``.
         """  # noqa: E501
         node = find_node_by_name(self.data_op, name)
         if node is None:

skrub/_data_ops/_skrub_namespace.py

@@ -143,13 +143,24 @@ def apply(
             _not_ be applied. The columns that are matched by ``cols`` AND not
             matched by ``exclude_cols`` are transformed.
 
-        how : "auto", "columnwise", "subframe" or "full_frame", optional
-            The mode in which it is applied. In the vast majority of cases the
-            default "auto" is appropriate. "columnwise" means a separate clone
-            of the transformer is applied to each column. "subframe" means it
-            is applied to a subset of the columns, passed as a single
-            dataframe. "full_frame" means the whole input dataframe is passed
-            directly to the provided ``estimator``.
+        how : "auto", "cols", "frame" or "no_wrap", optional
+            How the estimator is applied. In most cases the default "auto"
+            is appropriate.
+            - "cols" means `estimator` is wrapped in a :class:`ApplyToCols`
+              transformer, which fits a separate clone of `estimator` each
+              column in `cols`. `estimator` must be a transformer (have a
+              ``fit_transform`` method).
+            - "frame" means `estimator` is wrapped in a :class:`ApplyToFrame`
+              transformer, which fits a single clone of `estimator` to the
+              selected part of the input dataframe. `estimator` must be a
+              transformer.
+            - "no_wrap" means no wrapping, `estimator` is applied directly to
+              the unmodified input.
+            - "auto" chooses the wrapping depending on the input and estimator.
+              If the input is not a dataframe or the estimator is not a
+              transformer, the "no_wrap" strategy is chosen. Otherwise if the
+              estimator has a ``__single_column_transformer__`` attribute,
+              "cols" is chosen. Otherwise "frame" is chosen.
 
         allow_reject : bool, optional
             Whether the transformer can refuse to transform columns for which
@@ -181,6 +192,12 @@ def apply(
         --------
         skrub.DataOp.skb.make_learner :
             Get a skrub learner for this DataOp.
+        skrub.ApplyToCols :
+            Transformer that applies a given transformer separately to each
+            selected column.
+        skrub.ApplyToFrame:
+            Transformer that applies a given transformer to part of a
+            dataframe.
 
         Examples
         --------
@@ -554,7 +571,7 @@ def select(self, cols):
         2     cup  2020-04-04
         3   spoon  2020-04-05
         """
-        return self._apply(SelectCols(cols), how="full_frame")
+        return self._apply(SelectCols(cols), how="no_wrap")
 
     @check_data_op
     def drop(self, cols):
@@ -607,7 +624,7 @@ def drop(self, cols):
         2   3         5
         3   4         1
         """
-        return self._apply(DropCols(cols), how="full_frame")
+        return self._apply(DropCols(cols), how="no_wrap")
 
     @check_data_op
     def concat(self, others, axis=0):

skrub/_data_ops/tests/test_data_ops.py

@@ -12,6 +12,7 @@
 from sklearn.model_selection import train_test_split
 
 import skrub
+from skrub import ApplyToCols
 from skrub import selectors as s
 from skrub._data_ops import _data_ops
 from skrub._utils import PassThrough
@@ -329,45 +330,65 @@ class A(_data_ops.DataOpImpl):
         a.skb.eval()
 
 
-@pytest.mark.parametrize("why_full_frame", ["numpy", "predictor", "how"])
+@pytest.mark.parametrize("why_no_wrap", ["numpy", "predictor", "how"])
 @pytest.mark.parametrize("bad_param", ["cols", "how", "allow_reject"])
-def test_apply_bad_params(why_full_frame, bad_param):
+def test_apply_bad_params(why_no_wrap, bad_param):
     # When the estimator is a predictor or the input is a numpy array (not a
-    # dataframe) (or how='full_frame') the estimator can only be applied to the
+    # dataframe) (or how='no_wrap') the estimator can only be applied to the
     # full input without wrapping in ApplyToCols or ApplyToFrame. In this case
     # if the user passed a parameter that would require wrapping, such as
     # passing a value for `cols` that is not `all()`, or passing
-    # how='columnwise' or allow_reject=True, we get an error.
+    # how='cols' or allow_reject=True, we get an error.
 
-    if why_full_frame == bad_param == "how":
+    if why_no_wrap == bad_param == "how":
         return
     X_a, y_a = make_classification(random_state=0)
     X_df = pd.DataFrame(X_a, columns=[f"col_{i}" for i in range(X_a.shape[1])])
 
-    X = skrub.X(X_a) if why_full_frame == "numpy" else skrub.X(X_df)
-    if why_full_frame == "predictor":
+    X = skrub.X(X_a) if why_no_wrap == "numpy" else skrub.X(X_df)
+    if why_no_wrap == "predictor":
         estimator = LogisticRegression()
         y = skrub.y(y_a)
     else:
         estimator = PassThrough()
         y = None
-    how = "full_frame" if why_full_frame == "how" else "auto"
-    # X is a numpy array: how must be full_frame and selecting columns is not
+    how = "no_wrap" if why_no_wrap == "how" else "auto"
+    # X is a numpy array: how must be no_wrap and selecting columns is not
     # allowed.
     if bad_param == "cols":
-        if why_full_frame == "numpy":
+        if why_no_wrap == "numpy":
             cols = [0, 1]
         else:
             cols = ["col_0", "col_1"]
     else:
         cols = s.all()
-    how = "columnwise" if bad_param == "how" else how
+    how = "cols" if bad_param == "how" else how
     allow_reject = True if bad_param == "allow_reject" else False
 
-    with pytest.raises((ValueError, RuntimeError), match=""):
+    with pytest.raises(
+        (ValueError, RuntimeError),
+        match=(
+            r"(`cols` must be `all\(\)`|`how` must be 'auto'|`allow_reject` must be"
+            r" False)"
+        ),
+    ):
         X.skb.apply(estimator, y=y, how=how, allow_reject=allow_reject, cols=cols)
 
 
+def test_apply_invalid_how():
+    df = pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
+    X = skrub.var("X", df)
+    t = PassThrough()
+    for how in ["auto", "cols", "frame", "no_wrap"]:
+        assert list(X.skb.apply(t, how=how).skb.eval().columns) == ["a", "b"]
+    with pytest.raises(RuntimeError, match="`how` must be one of"):
+        X.skb.apply(t, how="bad value")
+    # TODO: remove when old names are dropped in 0.7.0
+    with pytest.warns(FutureWarning, match="'columnwise' has been renamed to 'cols'"):
+        wrapper = X.skb.apply(t, how="columnwise").skb.applied_estimator.skb.eval()
+        assert isinstance(wrapper, ApplyToCols)
+
+
 class Mul(BaseEstimator):
     def __init__(self, factor=1):
         self.factor = factor