drib861204 / dqn-atari-pytorch

DQN-atari-pytorch

Pytorch implementation of DQN/DDQN for atari games.

Requirement

• python 3.6
• pytorch >= 1.0.0
• OpenAI Gym (Atari)
• PIL
• NumPy
• OpenCV

Run Python Code

# Training
python main_pong -t train

# Evaluation
python main_pong -t test

Run on Google Colaboratary

• Login your google drive account and create a “DQN_tutorial/” folder at the outermost directory.
• Put the “dqn_tutorial.ipynb” into the "DQN_tutorial/" folder.
• The model and generated GIF will be saved at "DQN_tutorial/save/" and "DQN_tutorial/gif/" respectively.

Demo

After 400k iterations (~3hr on RTX2070).