Skip to content
Minimal TensorFlow implementation of the Advantage Actor-Critic model for Atari games
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
imgs
models
src
.gitignore
LICENSE
README.md
main.py
play.py

README.md

Ms Pacman Breakout Space Invaders

Advantage Actor-Critic

Minimal TensorFlow implementation of the Advantage Actor-Critic model for Atari games.

As an alternative to the asynchronous implementation, researchers found you can write asynchronous, deterministic implementation that waits for each actor to finish its segment of experience before performing an update, averaging over all of the actors. One advantage of this method is that it can more effectively use of GPUs, which perform best with large batch sizes. This algorithm is naturally called A2C, short for advantage actor-critic.

The gym environment wrappers used are from Open AI baseline

You can’t perform that action at this time.