jeanharb / option_critic

Implementation of the Option-Critic Architecture on the Atari (ALE) environment
170 stars 50 forks source link

The Option-Critic Architecture

Code for the Option-Critic Architecture https://arxiv.org/pdf/1609.05140v2.pdf.

Installation

Here's a list of all dependencies:

Training

To train, run following command:

python train_q.py --rom pong --num-options 8 --folder-name pong_tempmodel

To view a list of available parameters, run:

print train_q.py --help

To speed up training, we highly suggest using cudnn(CUDA).

Testing

To watch model after training, run:

python run_best_model.py models/pong_tempmodel/last_model.pkl