API

SelfCalibratingConformalPredictor

Use this class when you already have a point predictor for y and want:

• a calibrated point prediction
• a Venn-Abers style set of calibrated predictions
• a prediction interval induced by those calibrated predictions

Core methods:

• fit(...) / calibrate(...)
• predict(...) / predict_point(...)
• predict_venn_abers(...)
• predict_interval(...)
• check_coverage(...)
• plot(...)

Customization points:

• predictor can be a callable or any object with .predict(...)
• calibrator can be swapped for a user-supplied calibrator
• conformity_score can replace the built-in regression residual score logic

VennAbersQuantileConformalPredictor

Use this class when you have:

• a predictor for the (1 - alpha) quantile of a conformity score
• optionally a center predictor mu(x) for the default score |y - mu(x)|

Core methods:

• fit(...) / calibrate(...)
• predict_score_quantile(...)
• predict_score_venn_abers(...)
• predict_interval(...)
• check_coverage(...)
• check_threshold_calibration(...)

Customization points:

• score_quantile_predictor can be a callable or any object with .predict(...)
• score_fn(x, y) can replace the default absolute-residual conformity score
• calibrator can be replaced as long as it returns a transform over predicted score quantiles
• interval_solver_params controls the numeric y grid used for custom score level sets

Configuration types

• RegressionAlgoConfig
• QuantileAlgoConfig
• IntervalSolverConfig
• HistogramConfig

Dict-based configs still work for backward compatibility, but they emit deprecation warnings so users can migrate gradually to typed configs.