Skip to contents

Changelog

Source: NEWS.md

kpe 0.1.0

  • Initial release. Point estimate, confidence interval, and one-sided p-value for the personalization effect via the two-fold K-Fold Personalization Estimator of Li and Brunskill (2026).
  • Three estimators — kpe(), train_eval(), and papd() — sharing one argument signature and returning a "kpe" S3 object with print(), summary(), confint(), and coef() methods.
  • Pluggable reward, contextual-policy, and best-arm learners via built-in strings (backed by ranger, policytree, grf, glmnet in Suggests) or user-supplied fit/predict objects.
  • The estimate is reproducible regardless of n_cores: the per-shuffle fold split and the forest seeds are pinned to be independent of the parallel backend.