ConfusionMatrixDisplay#

class skore.ConfusionMatrixDisplay(*, confusion_matrix, display_labels, report_type, ml_task, thresholds, data_source, pos_label, response_method)[source]#

Display for confusion matrix.

Parameters:

confusion_matrixpd.DataFrame: Confusion matrix data in long format with columns: “True label”, “Predicted label”, “count”, “normalized_by_true”, “normalized_by_pred”, “normalized_by_all” and “threshold”. Each row represents one cell of one confusion matrix.
display_labelslist of str: Display labels for plot axes.
report_type{“comparison-cross-validation”, “comparison-estimator”, “cross-validation”, “estimator”}: The type of report.
ml_task{“binary-classification”, “multiclass-classification”}: The machine learning task.
data_source{“test”, “train”}: The data source to use.
pos_labelint, float, bool, str or None: The class considered as the positive class when displaying the confusion matrix.
response_methodstr: The estimator’s method that was used to get the predictions. The possible values are: “predict”, “predict_proba”, and “decision_function”.

Attributes:

thresholdsndarray of shape (n_thresholds,): Thresholds of the decision function. Each threshold is associated with a confusion matrix. Only available for binary classification. Thresholds are sorted in ascending order.
facet_seaborn FacetGrid: FacetGrid containing the confusion matrix.
figure_matplotlib Figure: Figure containing the confusion matrix.
ax_matplotlib Axes: Axes with confusion matrix.

frame(*, normalize=None, threshold_value='default')[source]#

Return the confusion matrix as a long format dataframe.

In binary classification, the confusion matrix can be returned at various decision thresholds. This is useful for understanding how the model’s predictions change as the decision threshold varies. Use threshold_value="default" to return the confusion matrix at the default threshold (0.5 for predict_proba response method, 0 for decision_function response method). Use threshold_value="all" to return all available thresholds without filtering.

The matrix is returned as a long format dataframe where each line represents one cell of the matrix. The columns are “true_label”, “predicted_label”, “value”, “threshold”, “split”, “estimator”, “data_source” ; where “value” is one of {“count”, “normalized_by_true”, “normalized_by_pred”, “normalized_by_all”}, depending on the value of the normalize parameter.

Parameters:

normalize{‘true’, ‘pred’, ‘all’}, default=None: Normalizes confusion matrix over the true (rows), predicted (columns) conditions or all the population. If None, the confusion matrix will not be normalized.
threshold_valuefloat, “default” or “all”, default=”default”: The decision threshold(s) to use when applicable (binary classification only). If "default", returns the confusion matrix at the default threshold (0.5 for predict_proba response method, 0 for decision_function response method). If "all", returns all available thresholds without filtering. If a float, returns the confusion matrix at the closest available threshold to the given value.

Returns:

framepandas.DataFrame: The confusion matrix as a dataframe.

help()[source]#: Display display help using rich or HTML.

plot(*, normalize=None, threshold_value='default', subplot_by='auto')[source]#

Plot the confusion matrix.

In binary classification, the confusion matrix can be displayed at various decision thresholds. This is useful for understanding how the model’s predictions change as the decision threshold varies. Use threshold_value="default" to plot at the default threshold (0.5 for predict_proba response method, 0 for decision_function response method).

Parameters:

normalize{‘true’, ‘pred’, ‘all’}, default=None: Normalizes confusion matrix over the true (rows), predicted (columns) conditions or all the population. If None, the confusion matrix will not be normalized.
threshold_valuefloat or “default”, default=”default”: The decision threshold to use when applicable (binary classification only). If "default", plots at the default threshold (0.5 for predict_proba response method, 0 for decision_function response method). If a float, plots at the closest available threshold.
subplot_by: Literal[“split”, “estimator”, “auto”] | None = “auto”,: The variable to use for subplotting. If None, the confusion matrix will not be subplotted. If “auto”, the variable will be automatically determined based on the report type.

Returns:

selfConfusionMatrixDisplay: Configured with the confusion matrix.

set_style(*, policy='update', heatmap_kwargs=None, facet_grid_kwargs=None)[source]#

Set the style parameters for the display.

Parameters:

policyLiteral[“override”, “update”], default=”update”: Policy to use when setting the style parameters. If “override”, existing settings are set to the provided values. If “update”, existing settings are not changed; only settings that were previously unset are changed.
heatmap_kwargsdict, default=None: Additional keyword arguments to be passed to seaborn.heatmap().
facet_grid_kwargsdict, default=None: Additional keyword arguments to be passed to seaborn.FacetGrid.

Returns:

selfobject: The instance with a modified style.

Raises:

ValueError: If a style parameter is unknown.

static style_plot(plot_func)[source]#

Apply consistent style to skore displays.

This decorator: 1. Applies default style settings 2. Executes plot_func 3. Calls plt.tight_layout() to make sure axis does not overlap 4. Restores the original style settings

Parameters: