UtilsRL is a reinforcement learning utility python package, which is designed for fast integration into other RL projects. Despite its lightweightness, it still provides a full set of functions needed for RL algorithms development.
Currently UtilsRL is maintained by researchers from LAMDA-RL group. Any bug report / feature request / improvement is appreciated.
You can install this package directly from pypi:
pip install UtilsRLAfter installation, you may still need to configure some other dependencies based on your platform, such as PyTorch.
We are still working on the docs, and the docs will be published as soon as possible.
Here we list some highlight features of UtilsRL:
- Extremely easy-to-use and research friendly argument parsing.
UtilsRL.exp.argparsesupports several handy features for research:- loading arguments from both
yaml,json,pythonfiles and command line - nested argument parsing
- loading arguments from both
- Well-implemented torch modules for Reinforcement Learning
- common network structures: MLP, CNN, RNN, Attention, Ensemble Blocks and etc
- policy networks with various output distributions
- normalizers implemented in
nn.Module, benefiting saving/loading by taking advantage ofstate_dict
- Powerful experiment loggers.
- Super fast Prioritized Experience Replay (PER) buffer. By binding c++-implemented data structures, we boost the efficiency of PER up to 10 times
We provide two examples, namely training PPO on mujoco tasks and training Rainbow on atari tasks as illustrations for integrating UtilsRL into your workflow (see examples/)
We took inspiration for module design from tianshou and Polixir OfflineRL.
We also thank @YuRuiii and @momanto for their participation in code testing and performance benchmarking.