Reference

API reference generated from docstrings via mkdocstrings. The three public estimator entry points are:

• kpe() — K-Fold Personalization Estimator.
• train_eval() — TrainEval baseline (single 50/50 split AIPW).
• papd() — PAPD baseline (3-fold centered IPW; Eq. S18).

All three return a KPEResult dataclass.

The registry and user-supplied-object protocols for reward_model, contextual_policy, and best_arm are documented in Learners.