All Tools
T
AgentsFreeOpen Source
TIANSHOU
PyTorch reinforcement learning platform with a Pythonic API
MIT
ABOUT
Many RL libraries bury algorithms behind nested classes or slow TensorFlow stacks, so researchers rewrite trainers instead of comparing policies. Tianshou exposes a Pythonic PyTorch API with production-quality algorithm implementations, prioritized and HER replay, and imitation-learning methods so you can assemble agents from documented building blocks.
INSTALL
pip install tianshouINTEGRATION GUIDE
1. Train PPO, SAC, or TD3 on a Gymnasium control environment with Tianshou's trainer loop
2. Compare Rainbow DQN, C51, and IQN on Atari without rewriting the replay and collection stack
3. Run offline RL with CQL, BCQ, or TD3+BC on a logged dataset when online interaction is unavailable
4. Add hindsight experience replay or an intrinsic curiosity module to an existing Tianshou agent
TAGS
pythonpytorchreinforcement-learningagentsdqnpposac