muzero-general

Why we included this project

MuZero General is one of the most approachable reference implementations of DeepMind's MuZero algorithm, the model-based reinforcement learning method that mastered Chess, Go, Shogi and Atari without ever being told the rules. If you are trying to understand how modern RL agents combine Monte Carlo tree search with a learned model of the environment, the heavily commented code and the wiki documentation are a much gentler entry point than the original pseudocode. Adding a new game only takes a single game file with the right hyperparameters, which makes it genuinely easy to experiment with board games or classic gym environments. It also ships with Ray-based multi-threaded and cluster training plus multi-GPU support, so toy experiments can be scaled well beyond a single laptop.