init

Author: GaetandeCast

examples/model_evaluation/plot_estimator_report.py

@@ -397,14 +397,12 @@ def operational_decision_cost(y_true, y_pred, amount):
 # %%
 # We can normalize the confusion matrix to get percentages instead of raw counts.
 # Here we normalize by true labels (rows):
-cm_display = report.metrics.confusion_matrix()
 cm_display.plot(normalize="true")
 plt.show()
 
 # %%
 # More plotting options are available via ``heatmap_kwargs``, which are passed to
 # seaborn's heatmap. For example, we can customize the colormap and number format:
-cm_display = report.metrics.confusion_matrix()
 cm_display.plot(heatmap_kwargs={"cmap": "Greens", "fmt": ".2e"})
 plt.show()
 

skore/src/skore/_externals/_sklearn_compat.py

@@ -869,3 +869,20 @@ def parametrize_with_checks(
         check_X_y,  # noqa: F401
         validate_data,  # noqa: F401
     )
+
+
+########################################################################################
+# Upgrading for scikit-learn 1.8
+########################################################################################
+
+if sklearn_version < parse_version("1.8"):
+
+    def confusion_matrix_at_thresholds(y_true, y_pred, pos_label):
+        from sklearn.metrics._ranking import _binary_clf_curve
+
+        fps, tps, thresholds = _binary_clf_curve(y_true, y_pred, pos_label=pos_label)
+        fns = tps[-1] - tps
+        tns = fps[-1] - fps
+        return tns, fps, fns, tps, thresholds
+else:
+    from sklearn.metrics._ranking import confusion_matrix_at_thresholds  # noqa: F401

skore/src/skore/_sklearn/_estimator/metrics_accessor.py

@@ -2057,6 +2057,7 @@ def confusion_matrix(
         data_source: DataSource = "test",
         X: ArrayLike | None = None,
         y: ArrayLike | None = None,
+        threshold: bool = False,
     ) -> ConfusionMatrixDisplay:
         """Plot the confusion matrix.
 
@@ -2080,6 +2081,14 @@ def confusion_matrix(
             New target on which to compute the metric. By default, we use the target
             provided when creating the report.
 
+        threshold : bool, default=False
+            Whether to enable decision threshold support for binary classification.
+            When True, the display will precompute confusion matrices at all possible
+            decision thresholds, allowing you to specify a threshold in `.plot()` or
+            `.frame()` methods. This is only applicable for binary classification and
+            requires the estimator to have `predict_proba` or `decision_function`
+            methods.
+
         Returns
         -------
         display : :class:`~skore._sklearn._plot.ConfusionMatrixDisplay`
@@ -2095,16 +2104,32 @@ def confusion_matrix(
         >>> split_data = train_test_split(X=X, y=y, random_state=0, as_dict=True)
         >>> classifier = LogisticRegression(max_iter=10_000)
         >>> report = EstimatorReport(classifier, **split_data)
-        >>> report.metrics.confusion_matrix()
+        >>> display = report.metrics.confusion_matrix()
+        >>> display.plot()
+
+        With decision threshold support for binary classification:
+
+        >>> display = report.metrics.confusion_matrix(threshold=True)
+        >>> display.plot(threshold=0.7)
         """
-        display_kwargs = {"display_labels": self._parent.estimator_.classes_.tolist()}
+        display_kwargs = {
+            "display_labels": self._parent.estimator_.classes_.tolist(),
+            "pos_label": self._parent.pos_label,
+            "threshold": threshold,
+        }
+
+        response_method: str | list[str] | tuple[str, ...]
+        if threshold and self._parent._ml_task == "binary-classification":
+            response_method = ("predict_proba", "decision_function")
+        else:
+            response_method = "predict"
         display = cast(
             ConfusionMatrixDisplay,
             self._get_display(
                 X=X,
                 y=y,
                 data_source=data_source,
-                response_method="predict",
+                response_method=response_method,
                 display_class=ConfusionMatrixDisplay,
                 display_kwargs=display_kwargs,
             ),