This codebase accompanies paper Conditionally Optimistic Exploration for Cooperative Deep Multi-Agent Reinforcement Learning accepted at UAI 2023.
The codebase is based on the open-sourced framework EPyMARL. Our implementations of EMC and MAVEN are based on the original EMC code and MAVEN code, respectively.
Please refer to the original README in EPyMARL for installation instructions.
To run an experiment with default configurations, run a command such as the following:
python3 src/main.py --config=ucb_mix_episode --env-config=gymma with env_args.time_limit=50 env_args.key="lbforaging:Foraging-8x8-2p-3f-v1"where --config refers to the algorithm config file located in src/config/algs, and --env-config refers to the environment config file in src/config/envs.
To perform a hyperparameter search, either use the search.py script provided by authors of EPyMARL, or use scripts we provide in the scripts directory like:
bash scripts/run_ucb_mix.shWe recommend our scripts as they enjoy more flexibility and clearity if experiments are scheduled using the SLURM manager.
We release a colab notebook for the didactic multi-player game described in our paper. Experiment results across MARL benchmarks are available here.