skiftet_statistical

https://github.com/Skiftet/skiftet_statistical

A small, dependency-free toolkit for online decision-making under uncertainty: register arms, ask which to play, record rewards, and let a pluggable policy balance exploration and exploitation. Ships Thompson Sampling, epsilon-greedy, UCB1 and Softmax; state serialises to a plain Hash for persistence and every policy is deterministic under an injected RNG for testing.