Deep Q-Network Implementation using chainer

Requirement

• Chainer 2.0.0
• Cupy 1.0.0
• PyAutoGUI
• PyQt4

Supported Game

• Winnie the Pooh's Home Run Derby

Usage

Run the game and:

python src/train.py -g 0 -o model/dqn --random 0.4 --random_reduction 0.00002 --min_random 0.1

Options:

• -g, --gpu: (optional) GPU device index (default: -1).
• -i, --input: (optional) input model file path without extension.
• -o, --output: (required) output model file path without extension.
• -r, --random: (optional) randomness of playing (default: 0.2).
• --pool_size: (optional) number of frames of memory pool (default: 50000).
• --random_reduction: (optional) randomness reduction rate per iteration (default: 0.00002).
• --min_random: (optional) minimum randomness of playing (default: 0.1).
• --double_dqn: (optional) use Double DQN algorithm
• --update_target_interval: (optional) interval to update target Q function of Double DQN (default: 2000)

License

MIT License