offlinerllib 0.1.1

A python module desgined for Offline RL algorithms developing and benchmarking.

OfflineRL-Lib

OfflineRL-Lib provides unofficial and benchmarked PyTorch implementations for selected OfflineRL algorithms, including:

• In-Sample Actor Critic (InAC)
• Extreme Q-Learning (XQL)
• Implicit Q-Learning (IQL)
• Decision Transformer (DT)
• Advantage-Weighted Actor Critic (AWAC)
• TD3-BC
• TD7

For Model-Based algorithms, please check OfflineRL-Kit!

Benchmark Results

• We benchmark and visualize the result via WandB. Click the following WandB links, and group the runs via the entry task (for offline experiments) or env (for online experiments).
• Available Runs
  • Offline:
    • TD7 :chart_with_upwards_trend:
    • XQL :chart_with_upwards_trend:
    • InAC :chart_with_upwards_trend:
    • AWAC :chart_with_upwards_trend:
    • IQL :chart_with_upwards_trend:
    • TD3BC :chart_with_upwards_trend:
    • Decision Transformer :chart_with_upwards_trend:
  • Online Runs
    • SAC :chart_with_upwards_trend:
    • TD3 :chart_with_upwards_trend:
    • TD7 :chart_with_upwards_trend:
    • XSAC :chart_with_upwards_trend:

Citing OfflineRL-Lib

If you use OfflineRL-Lib in your work, please use the following bibtex

@misc{offinerllib,
  author = {Chenxiao Gao},
  title = {OfflineRL-Lib: Benchmarked Implementations of Offline RL Algorithms},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/typoverflow/OfflineRL-Lib}},
}

Acknowledgements

We thank CORL for providing finetuned hyper-parameters.