mengli11235 / Bootstrapped-DQN-with-NP

Bootstrapped DQN with NP

Bootstrap DQN with Noisy Priors

This repo is the implementation of the paper Improving the Diversity of Bootstrapped DQN via Noisy Priors on ALE games.

Requirements

atari-py installed from https://github.com/kastnerkyle/atari-py
torch='1.0.1.post2'
cv2='4.0.0'
matplotlib='3.3.2'

Usage

run_bootstrap.py is the main file. visual_plot is the file to plot results.

To run experiment, set info in run_bootstrap.py. To run different games, set game names in info['GAME']. Set info['IMPROVEMENT'] to use noisy priors or not.

python run_bootstrap.py

Credit

Credit to which the project is built on:

https://github.com/johannah/bootstrap_dqn