fix(skore): Follow up on FeatureImportanceCoefficientsDisplay

skore/src/skore/_sklearn/_comparison/feature_importance_accessor.py

@@ -10,8 +10,8 @@
 from skore._sklearn._base import _BaseAccessor
 from skore._sklearn._cross_validation import CrossValidationReport
 from skore._sklearn._estimator import EstimatorReport
-from skore._sklearn._plot.metrics.feature_importance_display import (
-    FeatureImportanceDisplay,
+from skore._sklearn._plot.metrics.feature_importance_coefficients_display import (
+    FeatureImportanceCoefficientsDisplay,
 )
 from skore._utils._accessor import _check_comparison_report_sub_estimators_have_coef
 
@@ -29,21 +29,27 @@ def __init__(self, parent: ComparisonReport) -> None:
         super().__init__(parent)
 
     @available_if(_check_comparison_report_sub_estimators_have_coef())
-    def coefficients(self) -> FeatureImportanceDisplay:
+    def coefficients(self) -> FeatureImportanceCoefficientsDisplay:
         """Retrieve the coefficients for each report, including the intercepts.
 
-        If the compared reports are `EstimatorReport`s, the coefficients from each
+        If the compared reports are ``EstimatorReport``\s, the coefficients from each
         report's estimator are returned as a single-column DataFrame.
 
-        If the compared reports are `CrossValidationReport`s, the coefficients
+        If the compared reports are ``CrossValidationReport``\s, the coefficients
         across all cross-validation splits are retained and the columns are prefixed
         with the corresponding estimator name to distinguish them.
 
         Comparison reports with the same features are put under one key and are plotted
         together.
         When some reports share the same features and others do not, those with the same
         features are plotted together.
-        """
+
+        Returns
+        -------
+        FeatureImportanceCoefficientsDisplay
+            The feature importance display containing model coefficients and
+            intercept.
+        """  # noqa: D301
         similar_reports = defaultdict(list)
 
         for name, report in self._parent.reports_.items():
@@ -89,7 +95,12 @@ def coefficients(self) -> FeatureImportanceDisplay:
         else:
             raise TypeError(f"Unexpected report type: {self._parent._reports_type}")
 
-        return FeatureImportanceDisplay(self._parent, coef_frames)
+        return FeatureImportanceCoefficientsDisplay(
+            "comparison-estimator"
+            if self._parent._reports_type == "EstimatorReport"
+            else "comparison-cross-validation",
+            coef_frames,
+        )
 
     ####################################################################################
     # Methods related to the help tree

skore/src/skore/_sklearn/_cross_validation/feature_importance_accessor.py

@@ -6,8 +6,8 @@
 from skore._externals._pandas_accessors import DirNamesMixin
 from skore._sklearn._base import _BaseAccessor
 from skore._sklearn._cross_validation.report import CrossValidationReport
-from skore._sklearn._plot.metrics.feature_importance_display import (
-    FeatureImportanceDisplay,
+from skore._sklearn._plot.metrics.feature_importance_coefficients_display import (
+    FeatureImportanceCoefficientsDisplay,
 )
 from skore._utils._accessor import _check_cross_validation_sub_estimator_has_coef
 
@@ -22,9 +22,15 @@ def __init__(self, parent: CrossValidationReport) -> None:
         super().__init__(parent)
 
     @available_if(_check_cross_validation_sub_estimator_has_coef())
-    def coefficients(self) -> FeatureImportanceDisplay:
+    def coefficients(self) -> FeatureImportanceCoefficientsDisplay:
         """Retrieve the coefficients across splits, including the intercept.
 
+        Returns
+        -------
+        FeatureImportanceCoefficientsDisplay
+            The feature importance display containing model coefficients and
+            intercept.
+
         Examples
         --------
         >>> from sklearn.datasets import make_regression
@@ -34,15 +40,16 @@ def coefficients(self) -> FeatureImportanceDisplay:
         >>> report = CrossValidationReport(
         >>>     estimator=Ridge(), X=X, y=y, splitter=5, n_jobs=4
         >>> )
-        >>> report.feature_importance.coefficients().frame()
+        >>> display = report.feature_importance.coefficients()
+        >>> display.frame()
                     Intercept	Feature #0	Feature #1	Feature #2
         Split index
         0       	0.064837	74.100966	27.309656	17.367865
         1       	0.030257	74.276481	27.571421	17.392395
         2       	0.000084	74.107126	27.614821	17.277730
         3       	0.145613	74.207645	27.523667	17.391055
         4       	0.033695	74.259575	27.599610	17.390481
-        >>> report.feature_importance.coefficients().plot() # shows plot
+        >>> display.plot() # shows plot
         """
         combined = pd.concat(
             {
@@ -58,7 +65,7 @@ def coefficients(self) -> FeatureImportanceDisplay:
         ).T
         combined.index.name = "Split index"
 
-        return FeatureImportanceDisplay(self._parent, combined)
+        return FeatureImportanceCoefficientsDisplay("cross-validation", combined)
 
     ####################################################################################
     # Methods related to the help tree

skore/src/skore/_sklearn/_estimator/feature_importance_accessor.py

@@ -17,8 +17,8 @@
 from skore._externals._pandas_accessors import DirNamesMixin
 from skore._sklearn._base import _BaseAccessor
 from skore._sklearn._estimator.report import EstimatorReport
-from skore._sklearn._plot.metrics.feature_importance_display import (
-    FeatureImportanceDisplay,
+from skore._sklearn._plot.metrics.feature_importance_coefficients_display import (
+    FeatureImportanceCoefficientsDisplay,
 )
 from skore._sklearn.types import Aggregate
 from skore._utils._accessor import (
@@ -158,9 +158,15 @@ def __init__(self, parent: EstimatorReport) -> None:
         super().__init__(parent)
 
     @available_if(_check_estimator_has_coef())
-    def coefficients(self) -> FeatureImportanceDisplay:
+    def coefficients(self) -> FeatureImportanceCoefficientsDisplay:
         """Retrieve the coefficients of a linear model, including the intercept.
 
+        Returns
+        -------
+        FeatureImportanceCoefficientsDisplay
+            The feature importance display containing model coefficients and
+            intercept.
+
         Examples
         --------
         >>> from sklearn.datasets import load_diabetes
@@ -171,7 +177,8 @@ def coefficients(self) -> FeatureImportanceDisplay:
         >>> split_data = train_test_split(X=X, y=y, random_state=0, as_dict=True)
         >>> regressor = Ridge()
         >>> report = EstimatorReport(regressor, **split_data)
-        >>> report.feature_importance.coefficients().frame()
+        >>> display = report.feature_importance.coefficients()
+        >>> display.frame()
                    Coefficient
         Intercept     152.4...
         Feature #0     21.2...
@@ -184,7 +191,7 @@ def coefficients(self) -> FeatureImportanceDisplay:
         Feature #7    112.6...
         Feature #8    250.5...
         Feature #9     99.5...
-        >>> report.feature_importance.coefficients().plot() # shows plot
+        >>> display.plot() # shows plot
         """
         parent_estimator = self._parent.estimator_
 
@@ -239,7 +246,7 @@ def coefficients(self) -> FeatureImportanceDisplay:
             columns=columns,
         )
 
-        return FeatureImportanceDisplay(self._parent, df)
+        return FeatureImportanceCoefficientsDisplay("estimator", df)
 
     @available_if(_check_has_feature_importances())
     def mean_decrease_impurity(self):

skore/src/skore/_sklearn/_plot/metrics/feature_importance_display.py → skore/src/skore/_sklearn/_plot/metrics/feature_importance_coefficients_display.py

@@ -3,14 +3,15 @@
 from skore._sklearn._plot.base import DisplayMixin
 
 
-class FeatureImportanceDisplay(DisplayMixin):
+class FeatureImportanceCoefficientsDisplay(DisplayMixin):
     """Feature importance display.
 
     Each report type produces its own output frame and plot.
 
     Parameters
     ----------
-    parent : EstimatorReport | CrossValidationReport | ComparisonReport
+    parent : {"estimator", "cross-validation", "comparison-estimator",
+            "comparison-cross-validation"}
         Report type from which the display is created.
 
     coefficient_data : DataFrame | list[DataFrame]
@@ -76,16 +77,12 @@ def frame(self):
             - If a ``ComparisonReport``, the columns are the
             models passed in the report, with the index being the feature names.
         """
-        from skore import ComparisonReport, CrossValidationReport, EstimatorReport
-
-        if isinstance(self._parent, EstimatorReport):
+        if self._parent == "estimator":
             return self._frame_estimator_report()
-        elif isinstance(self._parent, CrossValidationReport):
+        elif self._parent == "cross-validation":
             return self._frame_cross_validation_report()
-        elif isinstance(self._parent, ComparisonReport):
-            return self._frame_comparison_report()
         else:
-            raise TypeError(f"Unrecognised report type: {self._parent}")
+            return self._frame_comparison_report()
 
     def _frame_estimator_report(self):
         return self._coefficient_data
@@ -110,53 +107,71 @@ def plot(self, **kwargs) -> None:
         return self._plot(**kwargs)
 
     def _plot_matplotlib(self, **kwargs):
-        from skore._sklearn._comparison import ComparisonReport
-        from skore._sklearn._cross_validation import CrossValidationReport
-        from skore._sklearn._estimator import EstimatorReport
-
-        if isinstance(self._parent, EstimatorReport):
+        if self._parent == "estimator":
             return self._plot_estimator_report()
-        elif isinstance(self._parent, CrossValidationReport):
+        elif self._parent == "cross-validation":
             return self._plot_cross_validation_report()
-        elif isinstance(self._parent, ComparisonReport):
-            return self._plot_comparison_report()
         else:
-            raise TypeError(f"Unrecognised report type: {self._parent}")
+            return self._plot_comparison_report()
 
     def _plot_estimator_report(self):
         self.figure_, self.ax_ = plt.subplots()
-        self._coefficient_data.plot.bar(ax=self.ax_)
-        self.ax_.set_title(f"{self._parent.estimator_name_} Coefficients")
+        self._coefficient_data.plot.barh(ax=self.ax_)
+        self.ax_.set_title("Coefficients")
         self.ax_.legend(loc="best", bbox_to_anchor=(1, 1), borderaxespad=1)
+        self.ax_.grid(False)
+        self.ax_.spines["top"].set_visible(False)
+        self.ax_.spines["right"].set_visible(False)
+        self.ax_.spines["left"].set_visible(False)
+        self.ax_.tick_params(axis="y", length=0)
         self.figure_.tight_layout()
         plt.show()
 
     def _plot_cross_validation_report(self):
         self.figure_, self.ax_ = plt.subplots()
-        self._coefficient_data.boxplot(ax=self.ax_)
+        self._coefficient_data.boxplot(ax=self.ax_, vert=False)
         self.ax_.set_title("Coefficient variance across CV splits")
+        self.ax_.grid(False)
+        self.ax_.spines["top"].set_visible(False)
+        self.ax_.spines["right"].set_visible(False)
+        self.ax_.spines["left"].set_visible(False)
+        self.ax_.tick_params(axis="y", length=0)
         self.figure_.tight_layout()
         plt.show()
 
     def _plot_comparison_report(self):
-        if self._parent._reports_type == "EstimatorReport":
-            for coef_frame in self._coefficient_data:
-                self.figure_, self.ax_ = plt.subplots()
-                coef_frame.plot.bar(ax=self.ax_)
-                self.ax_.legend(loc="best", bbox_to_anchor=(1, 1), borderaxespad=1)
-
-                self.ax_.set_title("Coefficients")
-                self.figure_.tight_layout()
-                plt.show()
-        elif self._parent._reports_type == "CrossValidationReport":
-            for coef_frame in self._coefficient_data:
-                self.figure_, self.ax_ = plt.subplots()
-                coef_frame.boxplot(ax=self.ax_)
-                self.ax_.set_title(
+        self.figure_, self.ax_ = plt.subplots(
+            nrows=1,
+            ncols=len(self._coefficient_data),
+            figsize=(5 * len(self._coefficient_data), 6),
+            squeeze=False,
+        )
+        self.ax_ = self.ax_.flatten()
+
+        if self._parent == "comparison-estimator":
+            self.figure_.suptitle("Coefficients")
+            for ax, coef_frame in zip(self.ax_, self._coefficient_data, strict=False):
+                coef_frame.plot.barh(ax=ax)
+                ax.legend(loc="best", bbox_to_anchor=(1, 1), borderaxespad=1)
+                ax.grid(False)
+                ax.spines["top"].set_visible(False)
+                ax.spines["right"].set_visible(False)
+                ax.spines["left"].set_visible(False)
+                ax.tick_params(axis="y", length=0)
+
+        elif self._parent == "comparison-cross-validation":
+            for ax, coef_frame in zip(self.ax_, self._coefficient_data, strict=False):
+                coef_frame.boxplot(ax=ax, vert=False)
+                ax.set_title(
                     f"{coef_frame.columns[0].split('__')[0]} Coefficients across splits"
                 )
-                plt.xticks(rotation=90)
-                plt.tight_layout()
-                plt.show()
+                ax.grid(False)
+                ax.spines["top"].set_visible(False)
+                ax.spines["right"].set_visible(False)
+                ax.spines["left"].set_visible(False)
+                ax.tick_params(axis="y", length=0)
         else:
-            raise TypeError(f"Unexpected report type: {type(self._parent.reports_[0])}")
+            raise TypeError(f"Unexpected report type: {self._parent}")
+
+        self.figure_.tight_layout()
+        plt.show()

sphinx/reference/report/comparison_report.rst

@@ -56,3 +56,16 @@ get the common performance metric representations.
     ComparisonReport.metrics.rmse
     ComparisonReport.metrics.roc_auc
     ComparisonReport.metrics.timings
+
+Feature importance
+------------------
+
+The `feature_importance` accessor helps you evaluate the importance
+used to train your estimator.
+
+.. autosummary::
+    :toctree: ../api/
+    :template: autosummary/accessor_method.rst
+
+    ComparisonReport.feature_importance.help
+    ComparisonReport.feature_importance.coefficients

sphinx/reference/report/cross_validation_report.rst

@@ -77,3 +77,16 @@ estimator across cross-validation splits.
     CrossValidationReport.metrics.roc
     CrossValidationReport.metrics.roc_auc
     CrossValidationReport.metrics.timings
+
+Feature importance
+------------------
+
+The `feature_importance` accessor helps you evaluate the importance
+used to train your estimator.
+
+.. autosummary::
+    :toctree: ../api/
+    :template: autosummary/accessor_method.rst
+
+    CrossValidationReport.feature_importance.help
+    CrossValidationReport.feature_importance.coefficients

sphinx/user_guide/reporters.rst

@@ -84,6 +84,16 @@ Refer to the :ref:`displays` section for more details regarding the `skore` disp
 API. Refer to the :ref:`estimator_metrics` section for more details on all the
 available metrics in `skore`.
 
+Model interpretability
+^^^^^^^^^^^^^^^^^^^^^^
+
+:obj:`EstimatorReport.feature_importance` is the entry point to interpret and explain a
+predictive model. This accessor provides methods that return a `skore` :class:`Display`
+object. As with other display objects, they expose three methods:
+(i) `plot` that visualizes the information.
+(ii) `set_style` that sets graphical options.
+(iii) `frame` that returns a `pandas.DataFrame` with the underlying data.
+
 Caching mechanism
 ^^^^^^^^^^^^^^^^^