60-second decision tree

Use this page to pick the correct first quickstart.

1. Model compatibility

Need a scikit-learn compatible estimator with fit and prediction methods.

2. Calibration split

Need a held-out calibration split: x_cal, y_cal.

3. Choose mode

• Classification: Classification quickstart

• Percentile or interval regression: Regression quickstart

• Probabilistic or thresholded regression: Regression quickstart

• Guarded explanations: Guarded explanations quickstart

Semantics are mode-specific. Use Calibrated interval semantics.

4. Minimal flow

from calibrated_explanations import WrapCalibratedExplainer

explainer = WrapCalibratedExplainer(model)
explainer.fit(x_proper, y_proper)
explainer.calibrate(x_cal, y_cal, feature_names=feature_names)
explanations = explainer.explain_factual(X_query)

Entry-point tier: Tier 1.