MAINT refactoring get_feature_names_out to define it in encoders (#1634)

Co-authored-by: Guillaume Lemaitre <guillaume@probabl.ai>

Author: rcap107

skrub/_apply_to_cols.py

@@ -123,30 +123,6 @@ def __init_subclass__(subclass, **kwargs):
                 wrapped = _wrap_add_check_single_column(getattr(subclass, method))
                 setattr(subclass, method, wrapped)
 
-    def get_feature_names_out(self, input_features=None):
-        """Return a list of features generated by the transformer.
-
-        Each feature has format ``{input_name}_{n_component}`` where ``input_name``
-        is the name of the input column, or a default name for the encoder, and
-        ``n_component`` is the idx of the specific feature.
-
-        Parameters
-        ----------
-        input_features : None
-            The input features. Ignored, only here for compatibility.
-
-        Returns
-        -------
-        list of str
-            The list of feature names.
-        """
-        check_is_fitted(self, "n_components_")
-        num_digits = len(str(self.n_components_ - 1))
-        return [
-            f"{self.input_name_}_{str(i).zfill(num_digits)}"
-            for i in range(self.n_components_)
-        ]
-
 
 def _wrap_add_check_single_column(f):
     # as we have only a few predefined functions to handle, using their exact

skrub/_datetime_encoder.py

@@ -511,7 +511,42 @@ def get_feature_names_out(self, input_features=None):
         return self.all_outputs_
 
 
-class _SplineEncoder(SingleColumnTransformer):
+class _BasePeriodicEncoder(SingleColumnTransformer):
+    """Base class for periodic encoders."""
+
+    def __init__(self, period=24):
+        self.period = period
+
+    def _post_process(self, X, new_features):
+        result = sbd.make_dataframe_like(X, dict(zip(self.all_outputs_, new_features)))
+        return sbd.copy_index(X, result)
+
+    def get_feature_names_out(self, input_features=None):
+        """Return a list of features generated by the transformer.
+
+        Each feature has format ``{input_name}_{n_component}`` where ``input_name``
+        is the name of the input column, or a default name for the encoder, and
+        ``n_component`` is the idx of the specific feature.
+
+        Parameters
+        ----------
+        input_features : None
+            The input features. Ignored, only here for compatibility.
+
+        Returns
+        -------
+        list of str
+            The list of feature names.
+        """
+        check_is_fitted(self, "n_components_")
+        num_digits = len(str(self.n_components_ - 1))
+        return [
+            f"{self.input_name_}_{str(i).zfill(num_digits)}"
+            for i in range(self.n_components_)
+        ]
+
+
+class _SplineEncoder(_BasePeriodicEncoder):
     """Generate univariate B-spline bases for features.
 
     This encoder will apply the scikit-learn SplineTransformer to the given
@@ -533,10 +568,21 @@ class _SplineEncoder(SingleColumnTransformer):
     """
 
     def __init__(self, period=24, n_splines=None, degree=3):
-        self.period = period
+        super().__init__(period=period)
         self.n_splines = n_splines
         self.degree = degree
 
+    def _periodic_spline_transformer(self):
+        n_splines = self.period if self.n_splines is None else self.n_splines
+        n_knots = n_splines + 1  # periodic and include_bias is True
+        return SplineTransformer(
+            degree=self.degree,
+            n_knots=n_knots,
+            knots=np.linspace(0, self.period, n_knots).reshape(n_knots, 1),
+            extrapolation="periodic",
+            include_bias=True,
+        )
+
     def fit_transform(self, X, y=None):
         """Fit the encoder and transform a column.
 
@@ -560,7 +606,6 @@ def fit_transform(self, X, y=None):
 
         X_out = self.transformer_.fit_transform(sbd.to_numpy(X).reshape(-1, 1))
 
-        self.is_fitted = True
         self.n_components_ = X_out.shape[1]
 
         # TODO: this will raise an error if X is None, but it should not happen
@@ -570,7 +615,7 @@ def fit_transform(self, X, y=None):
         self.input_name_ = sbd.name(X) + "_spline"
         self.all_outputs_ = self.get_feature_names_out()
 
-        return self._post_process(X, X_out)
+        return self._post_process(X, X_out.T)
 
     def transform(self, X):
         """Transform a column.
@@ -588,28 +633,10 @@ def transform(self, X):
 
         X_out = self.transformer_.transform(sbd.to_numpy(X).reshape(-1, 1))
 
-        return self._post_process(X, X_out)
-
-    def _post_process(self, X, result):
-        result = sbd.make_dataframe_like(X, dict(zip(self.all_outputs_, result.T)))
-        result = sbd.copy_index(X, result)
+        return self._post_process(X, X_out.T)
 
-        return result
 
-    def _periodic_spline_transformer(self):
-        if self.n_splines is None:
-            self.n_splines = self.period
-        n_knots = self.n_splines + 1  # periodic and include_bias is True
-        return SplineTransformer(
-            degree=self.degree,
-            n_knots=n_knots,
-            knots=np.linspace(0, self.period, n_knots).reshape(n_knots, 1),
-            extrapolation="periodic",
-            include_bias=True,
-        )
-
-
-class _CircularEncoder(SingleColumnTransformer):
+class _CircularEncoder(_BasePeriodicEncoder):
     """Generate trigonometric features for the given feature.
 
     This encoder will generate two features corresponding to the sine and cosine
@@ -621,9 +648,6 @@ class _CircularEncoder(SingleColumnTransformer):
         Period to be used as basis of the trigonometric function.
     """
 
-    def __init__(self, period=24):
-        self.period = period
-
     def fit_transform(self, X, y=None):
         """Fit the encoder and transform a column.
 
@@ -679,9 +703,3 @@ def transform(self, X):
         ]
 
         return self._post_process(X, new_features)
-
-    def _post_process(self, X, result):
-        result = sbd.make_dataframe_like(X, dict(zip(self.all_outputs_, result)))
-        result = sbd.copy_index(X, result)
-
-        return result

skrub/_minhash_encoder.py

@@ -297,3 +297,27 @@ def transform(self, X):
         result = sbd.make_dataframe_like(X, dict(zip(names, X_out.T)))
         result = sbd.copy_index(X, result)
         return result
+
+    def get_feature_names_out(self, input_features=None):
+        """Return a list of features generated by the transformer.
+
+        Each feature has format ``{input_name}_{n_component}`` where ``input_name``
+        is the name of the input column, or a default name for the encoder, and
+        ``n_component`` is the idx of the specific feature.
+
+        Parameters
+        ----------
+        input_features : None
+            The input features. Ignored, only here for compatibility.
+
+        Returns
+        -------
+        list of str
+            The list of feature names.
+        """
+        check_is_fitted(self, "n_components_")
+        num_digits = len(str(self.n_components_ - 1))
+        return [
+            f"{self.input_name_}_{str(i).zfill(num_digits)}"
+            for i in range(self.n_components_)
+        ]

skrub/_string_encoder.py

@@ -7,6 +7,7 @@
     TfidfVectorizer,
 )
 from sklearn.pipeline import Pipeline
+from sklearn.utils.validation import check_is_fitted
 
 from . import _dataframe as sbd
 from ._apply_to_cols import SingleColumnTransformer
@@ -263,3 +264,27 @@ def _post_process(self, X, result):
         result = sbd.copy_index(X, result)
 
         return result
+
+    def get_feature_names_out(self, input_features=None):
+        """Return a list of features generated by the transformer.
+
+        Each feature has format ``{input_name}_{n_component}`` where ``input_name``
+        is the name of the input column, or a default name for the encoder, and
+        ``n_component`` is the idx of the specific feature.
+
+        Parameters
+        ----------
+        input_features : None
+            The input features. Ignored, only here for compatibility.
+
+        Returns
+        -------
+        list of str
+            The list of feature names.
+        """
+        check_is_fitted(self, "n_components_")
+        num_digits = len(str(self.n_components_ - 1))
+        return [
+            f"{self.input_name_}_{str(i).zfill(num_digits)}"
+            for i in range(self.n_components_)
+        ]

skrub/_text_encoder.py

@@ -413,3 +413,27 @@ def __getstate__(self):
                 del state[prop]
 
         return state
+
+    def get_feature_names_out(self, input_features=None):
+        """Return a list of features generated by the transformer.
+
+        Each feature has format ``{input_name}_{n_component}`` where ``input_name``
+        is the name of the input column, or a default name for the encoder, and
+        ``n_component`` is the idx of the specific feature.
+
+        Parameters
+        ----------
+        input_features : None
+            The input features. Ignored, only here for compatibility.
+
+        Returns
+        -------
+        list of str
+            The list of feature names.
+        """
+        check_is_fitted(self, "n_components_")
+        num_digits = len(str(self.n_components_ - 1))
+        return [
+            f"{self.input_name_}_{str(i).zfill(num_digits)}"
+            for i in range(self.n_components_)
+        ]

skrub/tests/test_datetime_encoder.py

@@ -369,3 +369,11 @@ def test_error_checking_periodic_encoder(a_datetime_col):
 def test_error_dispatch(func):
     with pytest.raises(TypeError, match="Expecting a Pandas or Polars Series"):
         func(np.array([1]))
+
+
+def test_n_splines_default_value(df_module):
+    """Check that when `n_splines is None`, it defaults to the `period` value."""
+    period = 15
+    enc = _SplineEncoder(period=period)
+    result = enc.fit_transform(df_module.make_column("when", [20, 20, 20]))
+    assert sbd.column_names(result) == [f"when_spline_{i:02d}" for i in range(period)]