Nested conformal prediction and quantile out-of-bag ensemble methods

Conformal prediction is a popular tool for providing valid sets classification and regression problems, without relying on any distributional assumptions the data. While traditional description of conformal starts with nonconformity score, we provide an alternate (but equivalent) view that sequence nested calibrates them to find set. The framework subsumes all scores, including recent proposals based quantile density estimation. these ideas were originally derived sample splitting, our seamlessly extends other aggregation schemes like cross-conformal, jackknife+ out-of-bag methods. We use derive new algorithm (QOOB, pronounced cube) combines four ideas: regression, cross-conformalization, ensemble methods predictions. develop computationally efficient implementation also used by QOOB. In detailed numerical investigation, QOOB performs either best or close simulated real datasets. Code available at https://github.com/aigen/QOOB.