Training Atari2600 by Reinforcement Learning

Train Atari2600 and check how it works!


How to Setup

You can setup packages on your local env.

$ make setup

or you can run the docker image.

$ make docker-run

How to Run

$ python --help

usage: [-h] [--env ENV] [--checkpoint CHECKPOINT] [--n-iters N_ITERS] [--n-workers N_WORKERS] [--gpu]

optional arguments:
  -h, --help            show this help message and exit
  --env ENV             Atari-2600 env name (Default: Breakout-v0)
  --n-iters N_ITERS     Training iteration number (Default: 10)
  --n-workers N_WORKERS
                        Number of workers for sampling (Default: 4)
  --checkpoint CHECKPOINT
                        Checkpoint path for inference
  --gpu                 Use GPU (Default: False)
  --render              Render env during eval

More Atari2600 environments can be found at:

For Training

$ python --gpu  # GPU
$ python  # CPU

For Evaluation

$ python --render --checkpoint path-to-checkpoint

For Developers

For clean code, you can run formatting or linting.

$ make format  # formatting
$ make lint  # linting


View Github