All Tools
P
AgentsFreeOpen Source
PGX
Vectorized JAX environments for classic board games
Apache-2.0
ABOUT
Board-game RL usually steps one match at a time on CPU, which makes AlphaZero-scale self-play expensive. PGX implements Chess, Go, Shogi, and more as JAX-native environments so thousands of games can run in parallel on GPU or TPU.
INSTALL
pip install pgxINTEGRATION GUIDE
1. Train an AlphaZero-style agent on Chess or Go with batched JAX self-play
2. Benchmark game RL algorithms across Shogi, Backgammon, and poker environments
3. JIT-compile game step functions for hardware-accelerated research
4. Prototype a new board-game agent without writing a custom simulator
TAGS
pythonjaxboard-gameschessgoreinforcement-learning