Skip to content

Original Deep Q-Network code integrated with Facebook MazeBase environment

Notifications You must be signed in to change notification settings

SimsGautam/DQN-MazeBase

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DQN-MazeBase

Original DeepMind Deep Q-Network code integrated with Facebook MazeBase environment.

What?

Takes DQN agent from Deepmind-Atari-Deep-Q-Learner and MazeBase environment from Facebook MazeBase. Instead of using alewrap to interface with the Atari environment, we create a wrapper (mbwrap) to interface with MazeBase.

Why?

Facebook's MazeBase allows for building simple, custom 2D games inside which you can train neural network models to perform tasks. In addition to various toy games, MazeBase also includes a variety of neural models, including convolutional network and memory network, but these are trained via policy gradient. By integrating DQN code with MazeBase, we can additionally train Q-network models with deep Q-learning on MazeBase environments.

About

Original Deep Q-Network code integrated with Facebook MazeBase environment

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published