chore(skore/estimator-report): Clean some parts of the metrics accessor (probabl-ai#2331)

In preparation for a larger refactor of `_MetricsAccessor`

Author: auguste-probabl

reproduce_output.txt

@@ -0,0 +1,4 @@
+
+
+Plotting coefficients...
+Done.

reproduce_warning.py

@@ -0,0 +1,27 @@
+from sklearn.datasets import load_iris
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import PolynomialFeatures
+from sklearn.linear_model import LogisticRegression
+from skore import ComparisonReport, CrossValidationReport
+import matplotlib.pyplot as plt
+import warnings
+
+# Ensure warnings are shown
+warnings.simplefilter("always")
+
+X, y = load_iris(return_X_y=True)
+estimator_1 = LogisticRegression(max_iter=10000, random_state=42)
+estimator_2 = Pipeline(
+    [
+        ("poly", PolynomialFeatures()),
+        ("predictor", LogisticRegression(max_iter=10000, random_state=0)),
+    ]
+)
+report = ComparisonReport(
+    [CrossValidationReport(estimator_1, X, y), CrossValidationReport(estimator_2, X, y)]
+)
+display = report.inspection.coefficients()
+print("Plotting coefficients...")
+display.plot()
+plt.close('all')
+print("Done.")

skore/src/skore/_sklearn/_plot/inspection/coefficients.py

@@ -288,7 +288,7 @@ def plot(
         subplot_by: Literal["auto", "estimator", "label", "output"] | None = "auto",
         select_k: int | None = None,
         sorting_order: Literal["descending", "ascending", None] = None,
-    ) -> None:
+    ) -> Any:
         """Plot the coefficients for the different features.
 
         Parameters
@@ -354,7 +354,7 @@ def _plot_matplotlib(
         subplot_by: Literal["estimator", "label", "output"] | None = None,
         select_k: int | None = None,
         sorting_order: Literal["descending", "ascending", None] = None,
-    ) -> None:
+    ) -> Any:
         """Dispatch the plotting function for matplotlib backend."""
         frame = self.frame(
             include_intercept=include_intercept,
@@ -409,9 +409,9 @@ def _categorical_plot(
         barplot_kwargs: dict[str, Any] | None = None,
         boxplot_kwargs: dict[str, Any] | None = None,
         stripplot_kwargs: dict[str, Any] | None = None,
-    ):
+    ) -> Any:
         if "estimator" in report_type:
-            self.facet_ = sns.catplot(
+            facet_ = sns.catplot(
                 data=frame,
                 x="coefficient",
                 y="feature",
@@ -421,7 +421,7 @@ def _categorical_plot(
                 **barplot_kwargs,
             )
         else:  # "cross-validation" in report_type
-            self.facet_ = sns.catplot(
+            facet_ = sns.catplot(
                 data=frame,
                 x="coefficient",
                 y="feature",
@@ -451,14 +451,15 @@ def _categorical_plot(
         )
         for ax, n_feature in zip(self.ax_.flatten(), n_features, strict=True):
             _decorate_matplotlib_axis(
-                ax=ax,
+                ax=axis,
                 add_background_features=add_background_features,
                 n_features=n_feature,
                 xlabel="Magnitude of coefficient",
                 ylabel="",
             )
-        if len(self.ax_.flatten()) == 1:
-            self.ax_ = self.ax_.flatten()[0]
+        if len(ax_.flatten()) == 1:
+            ax_ = ax_.flatten()[0]
+        return facet_
 
     def _plot_single_estimator(
         self,
@@ -470,7 +471,7 @@ def _plot_single_estimator(
         barplot_kwargs: dict[str, Any],
         boxplot_kwargs: dict[str, Any],
         stripplot_kwargs: dict[str, Any],
-    ) -> None:
+    ) -> Any:
         """Plot the coefficients for an `EstimatorReport` or a `CrossValidationReport`.
 
         An `EstimatorReport` will use a bar plot while a `CrossValidationReport` will
@@ -528,7 +529,7 @@ def _plot_single_estimator(
             barplot_kwargs.pop("palette", None)
             stripplot_kwargs.pop("palette", None)
 
-        self._categorical_plot(
+        facet_ = self._categorical_plot(
             frame=frame,
             report_type=report_type,
             hue=hue,
@@ -541,7 +542,8 @@ def _plot_single_estimator(
         title = f"Coefficients of {estimator_name}"
         if subplot_by is not None:
             title += f" by {subplot_by}"
-        self.figure_.suptitle(title)
+        facet_.figure.suptitle(title)
+        return facet_
 
     @staticmethod
     def _has_same_features(*, frame: pd.DataFrame) -> bool:
@@ -565,7 +567,7 @@ def _plot_comparison(
         barplot_kwargs: dict[str, Any],
         boxplot_kwargs: dict[str, Any],
         stripplot_kwargs: dict[str, Any],
-    ) -> None:
+    ) -> Any:
         """Plot the coefficients for a `ComparisonReport`.
 
         Parameters
@@ -657,7 +659,7 @@ def _plot_comparison(
                     "different axis using `subplot_by='estimator'`."
                 )
 
-        self._categorical_plot(
+        facet_ = self._categorical_plot(
             frame=frame,
             report_type=report_type,
             hue=hue,
@@ -670,7 +672,8 @@ def _plot_comparison(
         title = "Coefficients"
         if subplot_by is not None:
             title += f" by {subplot_by}"
-        self.figure_.suptitle(title)
+        facet_.figure.suptitle(title)
+        return facet_
 
     @classmethod
     def _compute_data_for_display(

skore/src/skore/_sklearn/_plot/inspection/impurity_decrease.py

@@ -30,17 +30,6 @@ class ImpurityDecreaseDisplay(DisplayMixin):
             "comparison-cross-validation"}
         Report type from which the display is created.
 
-    Attributes
-    ----------
-    ax_ : matplotlib Axes
-        Matplotlib Axes with the plot.
-
-    facet_ : seaborn FacetGrid
-        FacetGrid containing the plot.
-
-    figure_ : matplotlib Figure
-        Figure containing the plot.
-
     Examples
     --------
     >>> from sklearn.datasets import load_iris
@@ -209,7 +198,7 @@ def plot(self) -> None:
         """
         return self._plot()
 
-    def _plot_matplotlib(self) -> None:
+    def _plot_matplotlib(self) -> Any:
         """Dispatch the plotting function for matplotlib backend.
 
         This method creates a bar plot showing the mean decrease in impurity for each
@@ -221,7 +210,7 @@ def _plot_matplotlib(self) -> None:
         boxplot_kwargs = self._default_boxplot_kwargs.copy()
         frame = self.frame()
 
-        self._plot_single_estimator(
+        return self._plot_single_estimator(
             frame=frame,
             estimator_name=self.importances["estimator"].unique()[0],
             report_type=self.report_type,
@@ -300,7 +289,7 @@ def _plot_single_estimator(
         self.figure_, self.ax_ = self.facet_.figure, self.facet_.axes.squeeze()
         self.ax_ = self.ax_[()]  # 0-d array
         _decorate_matplotlib_axis(
-            ax=self.ax_,
+            ax=ax_,
             add_background_features=False,
             n_features=frame["feature"].nunique(),
             xlabel="Mean Decrease in Impurity (MDI)",

skore/src/skore/_sklearn/_plot/inspection/permutation_importance.py

@@ -34,16 +34,6 @@ class PermutationImportanceDisplay(DisplayMixin):
     report_type : {"estimator"}
         Report type from which the display is created.
 
-    Attributes
-    ----------
-    facet_ : seaborn FacetGrid
-        FacetGrid containing the permutation importance.
-
-    figure_ : matplotlib Figure
-        Figure containing the permutation importance.
-
-    ax_ : matplotlib Axes
-        Axes with permutation importance.
     """
 
     _default_boxplot_kwargs: dict[str, Any] = {
@@ -166,7 +156,7 @@ def plot(
         *,
         subplot_by: str | tuple[str, str] | None = "auto",
         metric: str | list[str] | None = None,
-    ) -> None:
+    ) -> Any:
         """Plot the permutation importance.
 
         Parameters
@@ -197,7 +187,7 @@ def _plot_matplotlib(
         *,
         subplot_by: str | tuple[str, str] | None = "auto",
         metric: str | list[str] | None = None,
-    ) -> None:
+    ) -> Any:
         """Dispatch the plotting function for matplotlib backend."""
         boxplot_kwargs = self._default_boxplot_kwargs.copy()
         stripplot_kwargs = self._default_stripplot_kwargs.copy()
@@ -213,7 +203,7 @@ def _plot_matplotlib(
         elif "output" in frame.columns and frame["output"].isna().any():
             raise ValueError(err_msg.format("outputs"))
 
-        self._plot_single_estimator(
+        return self._plot_single_estimator(
             subplot_by=subplot_by,
             frame=frame,
             estimator_name=self.importances["estimator"].unique()[0],
@@ -229,7 +219,7 @@ def _plot_single_estimator(
         estimator_name: str,
         boxplot_kwargs: dict[str, Any],
         stripplot_kwargs: dict[str, Any],
-    ) -> None:
+    ) -> Any:
         """Plot the permutation importance for an `EstimatorReport`."""
         if subplot_by == "auto":
             is_multi_metric = frame["metric"].nunique() > 1
@@ -295,7 +285,7 @@ def _plot_single_estimator(
             # deprecation warning if passing palette without a hue
             stripplot_kwargs.pop("palette", None)
 
-        self.facet_ = sns.catplot(
+        facet_ = sns.catplot(
             data=frame,
             x="value",
             y="feature",
@@ -317,9 +307,9 @@ def _plot_single_estimator(
         add_background_features = hue is not None
 
         metrics = frame["metric"].unique()
-        self.figure_, self.ax_ = self.facet_.figure, self.facet_.axes.squeeze()
-        for row_index, row_axes in enumerate(self.facet_.axes):
-            for col_index, ax in enumerate(row_axes):
+        figure_, ax_ = facet_.figure, facet_.axes.squeeze()
+        for row_index, row_axes in enumerate(facet_.axes):
+            for col_index, axis in enumerate(row_axes):
                 if len(metrics) > 1:
                     if row == "metric":
                         xlabel = f"Decrease in {metrics[row_index]}"
@@ -331,18 +321,19 @@ def _plot_single_estimator(
                     xlabel = f"Decrease in {metrics[0]}"
 
                 _decorate_matplotlib_axis(
-                    ax=ax,
+                    ax=axis,
                     add_background_features=add_background_features,
                     n_features=frame["feature"].nunique(),
                     xlabel=xlabel,
                     ylabel="",
                 )
-        if len(self.ax_.flatten()) == 1:
-            self.ax_ = self.ax_.flatten()[0]
+        if len(ax_.flatten()) == 1:
+            ax_ = ax_.flatten()[0]
         data_source = frame["data_source"].unique()[0]
-        self.figure_.suptitle(
+        figure_.suptitle(
             f"Permutation importance of {estimator_name} on {data_source} set"
         )
+        return facet_
 
     def frame(
         self,