All Tools
D
AgentsFreeOpen Source
D3RLPY
Offline deep reinforcement learning with a scikit-learn-style API
MIT
ABOUT
Online RL needs live environment interaction, which is unsafe or impossible in robotics, healthcare, and production systems that only have logged data. d3rlpy trains CQL, BCQ, and related offline algorithms from those datasets through a scikit-learn-style fit API, so teams can learn policies without rolling out untrained agents.
INSTALL
pip install d3rlpyINTEGRATION GUIDE
1. Fit Conservative Q-Learning on a D4RL or Minari dataset and evaluate the resulting policy
2. Train Discrete CQL on Atari replay when you cannot collect new game frames online
3. Compare online SAC against offline CQL on the same continuous-control task with one API
4. Prototype an offline RL pipeline on PyBullet hopper data before moving to a production log
TAGS
pythonpytorchoffline-rlreinforcement-learningcqlbcqdatasets