calibration

SuperLSS.calibration(

X: object | EagerFrame,

y: NDArray,

*,

sample_weight: NDArray | None = None,

offsets: Mapping[str, NDArray] | None = None,

**kwargs: Any,

) → CalibrationPayload

Return interval coverage, tail totals, quantile calibration and reliability.

Four questions in one payload, in the sense of Gneiting, Balabdaoui and Raftery (2007): does the central interval hold at each level, does the expected count of exceedances match the realised one at each threshold, does each predicted quantile leave the right fraction of rows above it, and is the exceedance forecast reliable (the CORP diagram of Dimitriadis, Gneiting and Jordan 2021). sample_weight is the aggregation weight; levels, thresholds and quantile_grid reach the builder.