8000 GitHub - pzombie/dqn_atari: A python implementation of the DeepMind Atari agent
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

A python implementation of the DeepMind Atari agent

Notifications You must be signed in to change notification settings

pzombie/dqn_atari

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

30 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

dqn_atari

The repository is a learning project; essentially me trying to figure out how Deep Q-learning works.

It's a re-implementation of the DeepMind DQN algorithm by Mnih et al. [1], with the Double Q-Learning modification by van Hasselt et al. [2]. I've tried to keep the code simple, so it's easy to understand, and also fast to keep debugging cycles short... debugging this thing took forever (!).

Here's the agent playing Breakout after about 10M observations. This took about 20 hours (wall time) on my desktop PC (1080 Ti) @ roughly 150 frames per second.

I've learned a lot from the implementations of others. Notably from:

https://github.com/yilundu/DQN-DDQN-on-Space-Invaders

https://github.com/keras-rl/keras-rl

https://github.com/devsisters/DQN-tensorflow

and from the following blogs:

https://becominghuman.ai/lets-build-an-atari-ai-part-1-dqn-df57e8ff3b26

https://yanpanlau.github.io/2016/07/10/FlappyBird-Keras.html

[1] Mnih et al., Human-level control through deep reinforcement learning, Nature (2015)

[2] van Hasselt et al., Deep Reinforcement Learning with Double Q-learning, arXiv:1509.06461v3

About

A python implementation of the DeepMind Atari agent

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published
0