gauss-clb/DQN

Tensorflow implementation of Human-Level Control through Deep Reinforcement Learning.

DQN

Prerequisites

• Python3.5
• Numpy, Scipy, Matplotlib, Shutil
• Tensorflow
• Arcade-Learning-Environment

Training

python main.py

The 200 epoches will take more than 3 days, you can early stop it. The records of training process are in the breakout.csv.

Testing

python main.py --train=False

Saving the best result

python main.py --train=False --save=True

The best result is saved in the directory best_result, we can use ffmpeg to get a video.

Best result

References

• simple_dqn

License

MIT License