incerto.conformal.jackknife_plus

Contents

incerto.conformal.jackknife_plus#

incerto.conformal.jackknife_plus(model_fn, train_dataset, alpha)[source]#

Jackknife+ Intervals — Barber, Candès, and Ramdas, Ann. Stat. 2021.

Produces prediction intervals using leave-one-out cross-validation with the Jackknife+ aggregation rule. All n LOO models are retained and used at prediction time to construct valid intervals.

Parameters:

  • model_fn (Callable[[Dataset], Module]) – Function that trains a model on a provided dataset subset.

  • train_dataset (Dataset) – Training dataset used for LOO calibration.

  • alpha (float) – Miscoverage rate (e.g., 0.1 for 90% coverage).

Return type:

Callable[[Tensor], Tuple[Tensor, Tensor]]

Contents