🥟 Gyozas 🥟 released: a pure-Python RL library for combinatorial optimization

We are releasing Gyozas, a pure-Python reinforcement learning library for combinatorial optimization, built on top of PySCIPOpt and Gymnasium.

Gyozas provides a clean, Gymnasium-compatible interface to train RL agents that control SCIP’s branch-and-bound solver — either for variable selection (branching) or node selection. It ships with bipartite graph observations, pluggable reward functions, and built-in instance generators for standard combinatorial benchmarks (Set Cover, Independent Set, Combinatorial Auction, Facility Location).

pip install gyozas

Links: GitHub · Documentation · PyPI · Project