Alchemy is an open-source benchmark environment for meta-reinforcement learning. Because it involves a compositional latent structure that is both rich and well-defined, Alchemy presents a substantive meta-learning challenge, while at the same time affording principled analyses of agent behavior. The release includes multiple versions of the game (including a simplified, symbolic version, and a human-playable version), along with Bayes-optimal and other benchmarks, and numerous other resources and analysis tools. Check out the GitHub repository here.


09 Feb 2021