This repo contains the source code of the CleanRL experiments presented in EnvPool: A Highly Parallel Reinforcement Learning Environment Execution Engine
- Paper url: https://arxiv.org/abs/2206.10558
- Tracked Weights and Biases experiments: https://wandb.ai/openrlbenchmark/envpool-cleanrl
If you like this repo, consider checking out CleanRL (https://github.com/vwxyzjn/cleanrl), the RL library that we used to build this repo.
Prerequisites:
- Python 3.9+
- Poetry
Install dependencies:
poetry install
Train agents:
poetry run python ppo_atari.py
Train agents with experiment tracking:
poetry run python ppo_atari.py --track
Train agents:
poetry run python ppo_continuous_action.py
Train agents with experiment tracking:
poetry run python ppo_continuous_action.py --track
See benchmark.sh
@article{envpool,
title={EnvPool: A Highly Parallel Reinforcement Learning Environment Execution Engine},
author={Weng, Jiayi and Lin, Min and Huang, Shengyi and Liu, Bo and Makoviichuk, Denys and Makoviychuk, Viktor and Liu, Zichen and Song, Yufan and Luo, Ting and Jiang, Yukun and Xu, Zhongwen and Yan, Shuicheng},
journal={arXiv preprint arXiv:2206.10558},
year={2022}
}