Accel is an open source library for reinforcement learning.
Accel is based on PyTorch. You must install PyTorch by yourself in advance.
Then, you can install Accel via pip (>=19.0) by
$ pip install git+https://github.com/waffoo/accel
You can also install it from the source code:
$ pip install .
Some examples are constructed on top of D4RL. Please install it if necessary.
Following algorithms have been implemented:
- DQN (including Double DQN, Dueling DQN, Prioritized Experience Replay, and Multi-step DQN)
- Soft Actor-Critic (SAC)
- PPO
- Behavioral Chroning (BC)
- Only examples are provided
- Conservative Q-Learning (CQL) (+ Double DQN, SAC)