All Tools
M
AgentsFreeOpen Source
MINARI
Standard dataset API for offline reinforcement learning
MIT
ABOUT
Offline RL experiments stall when every paper ships a different dataset format, download path, and episode schema. Minari standardizes Gymnasium trajectories with a CLI, Hugging Face hosting, and sampling utilities so researchers can share D4RL-style datasets and train algorithms against the same episodes instead of rewriting loaders.
INSTALL
pip install minariINTEGRATION GUIDE
1. Download a D4RL door or hopper dataset with the Minari CLI and train an offline RL algorithm
2. Sample and filter episodes from a hosted dataset before fitting CQL or IQL
3. Generate a custom Gymnasium dataset from your own policy rollouts in the Minari format
4. Share an offline RL benchmark on Hugging Face Hub that other Minari users can download in one command
TAGS
pythonoffline-rldatasetsgymnasiumfaramahuggingfacereinforcement-learning