Tensorflow implementation of deep Q reinforcment learning for two games from OpenAI Gym: Breakout and Ms Pacman.
To run the games with the default hyperparameters, use the following commmands and specify a run_num to create a new log directory:
python breakout.py --run_num 1
python mspacman.py --run_num 1
The default hyperparameters were selected through experimentation, but can be adjusted by adding arguments to a game launch.
To visualize the average total episode reward at each epoch, launch tensorboard with the following command:
tensorboard --logdir=breakout/logs/[run_num]
To evaluate a model by playing 1000 games and calculating the total reward, run the following command:
python breakout_eval.py --model_path path/to/model.ckpt
python mspacman_eval.py --model_path path/to/model.ckpt