Reference Issues/PRs

See also #16880 (users wanting heatmap colors to reflect classification performance rather than sample counts).

What does this implement/fix? Explain your changes.

This adds metrics.ClassificationReportDisplay, a visual (heatmap) rendering of a classification report, analogous to the existing ConfusionMatrixDisplay. It covers the gap that third-party libraries such as Yellowbrick currently fill.

The display shows the per-class precision, recall and f1-score on a shared, fixed 0–1 color scale, so a cell's color reflects classification quality directly (unlike a raw confusion matrix, where color tracks the number
of samples). support is shown as a text-only column since it is a count, not a score. Rows are the individual classes followed by the macro avg and weighted avg summary rows; the scalar accuracy entry is intentionally left out of the grid.

Following the established Display conventions, three entry points are provided:

• ClassificationReportDisplay(report) — visualize an existing report dict
  (the output of classification_report(..., output_dict=True));
• ClassificationReportDisplay.from_predictions(y_true, y_pred, ...);
• ClassificationReportDisplay.from_estimator(estimator, X, y, ...).

Metric computation is delegated entirely to classification_report; the display only handles layout. No new dependencies.

Included:

• sklearn/metrics/_plot/classification_report.py — the display class;
• tests in sklearn/metrics/_plot/tests/test_classification_report_display.py, including a check that the color scale is fixed to [0, 1] independent of support magnitude;
• exports in sklearn/metrics/__init__.py, and entries in doc/api_reference.py and doc/visualizations.rst;
• a changelog fragment;
• a gallery example (examples/model_selection/plot_classification_report.py) and a paragraph in the Classification report section of the user guide.

First time contributor introduction

I'm a data scientist working in cyber security (and PMC chair for Apache Drill), and CEO at GTK Cyber. I use scikit-learn regularly and kept wanting a built-in visual companion to classification_report the same way ConfusionMatrixDisplay complements confusion_matrix, so I don't have to pull in an extra dependency for it. This is my attempt to contribute that upstream.

AI usage disclosure

I used AI assistance for:

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

Any other comments?

Thank you for your consideration.