This package give python code to compute the reverse information projection in a case of multivariate Bernoulli distributions with Frank-Wolfe algorithm. It is only applicable when projecting on a set parameterized by union of convex sets.
To install, first create a new virtual environment
python3 -m venv test_ripr
then activate it with
source test_ripr/bin/activate
and finally, install the library with pip install git+https://github.com/TimotheeMathieu/evalue-ripr.git.
See examples in the examples folder to get started on using this package.
This code was developped for the paper "Optimal Posterior E-values with Non-Convex Parameter Sets with Applications to Voting Systems" (Adrienne Tuynman and Timothée Mathieu, 2026, https://arxiv.org/abs/2606.29998). Please cite this paper when using this software.