actual_expected

SuperLSS.actual_expected(

X: object | EagerFrame,

y: NDArray,

covariate: str | NDArray,

*,

name: str | None = None,

sample_weight: NDArray | None = None,

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

**kwargs: Any,

) → ActualExpected

Return the realised against the predicted total per bin of a covariate.

Every number in the table is a ratio of weighted sums – sum w y over sum w mu_hat – so on a rate target with exposure weights it reads as total cost over total expected cost, never as the mean of per-row ratios. sample_weight is that aggregation weight, and it also fixes the standard error’s law: prior weights put the weight inside each row’s distribution, frequency weights replicate the row.