This is an attemp to run A3C using Predictive Net agent.
- Python 3.5.1
- chainer 1.8.1
- cached-property 1.3.0
- h5py 2.5.0
- Arcade-Learning-Environment
- main
You can use FF/LSTM agent (made by @muupan).
- PredNet
You can use Predictive Net agent (learn only actor and critic).
- PredS
You can use Predictive Net agent (learn actor and cricit, and predic next state).
python a3c_ale.py <number-of-processes> <path-to-atari-rom>
This code is mostly based on muupan's project.