get_model_evaluation

Scope: read · Signature: get_model_evaluation(experiment_id: str, split: str | None = None, class_key: str | None = None) -> Any

Use for numeric model quality: metrics, curves, confusion — not prose.

Returns per-split metric tiles + ROC/PR/calibration/lift/gains/KS curves, score distributions, confusion matrix, regression scatter, and confusion_curve: tp/fp/fn/tn at the thresholds that flag the top 1, 2, 5, 10, 15, 20, 30 and 50% of rows and at the decision threshold — precision and recall at a review budget. Curves reach you as 20 evenly spaced points (the panel draws every point and threshold). Defaults to macro splits only: pass split to narrow to one split, class_key for one multiclass one-vs-rest slice. 404 → no holdout evaluation attached. For prose use get_autodoc.

Typed output: schema model_evaluation — generated from the backend pydantic model by scripts/export_mcp_schemas.py (the same SSOT as the React client types).

Powered by: mli evaluation & trust (mli/docs/guides/evaluation-trust.md).