Reporting

Every result, report and comparison exports to text, JSON, pandas, Markdown, LaTeX, HTML and CSV. save() picks the format from the file extension.

import evalsuite as es

result = es.evaluate(y_true, y_pred, y_prob=y_prob)

result.to_latex(caption="Model evaluation results", label="tab:results")
result.save("report.html")     # standalone page: metrics, per-class tables, confusion matrix
result.save("results.csv")     # columns: metric, name, label, value
result.save("results.json")

report = es.classification_report(y_true, y_pred)
report.save("report.md")

Plots

With the plot extra installed, es.plot draws ROC, precision-recall, calibration, confusion-matrix, residual and model-comparison figures. Each returns a Matplotlib Axes, and every number in a legend comes from EvalSuite's own metrics.

ax = es.plot.roc(y_true, {"baseline": prob_a, "candidate": prob_b})
ax.figure.savefig("roc.png", dpi=300)

es.plot.calibration(y_true, y_prob)
es.plot.confusion_matrix(y_true, y_pred)
es.plot.residuals(y_true, y_pred, kind="predicted")

Later releases add es.plot.decision_curve (v0.2.0) and the vision figures es.plot.segmentation, es.plot.per_class and es.plot.detection_pr (v0.3.0). Segmentation and detection reports export to the same formats with save().

HTML reports contain no scripts and escape all text. Raw predictions are never included.