Scalable Offline Reinforcement Learning for Mean Field Games

Official code for Offline Munchausen Mirror Descent.

Dependency installation: pip install -r requirements.txt.

The main file for training is src/train.py.

Note that in order to train Off-MMD, you need to generate the datasets first. We provide the policy checkpoints to generate the datasets via src/collect_dataset.py.

Configuration files are located in config.