risk_curves

SuperLSS.risk_curves(

reference: Mapping[str, Any] | Series,

covariate: str,

*,

X_train: object | EagerFrame | None = None,

**kwargs: Any,

) → RiskCurves

Return predicted response quantiles along one covariate, with bands.

One reference row is swept over the covariate’s training range while every other column is held at reference (or at its training centre), and each requested quantile of the predictive law is reported with a posterior band drawn from one shared draw set, so the curves are coherent with one another. weights here are per swept point, not per training row: they state the exposure the curve is priced at.