Skip to content

Implementation of the Dueling-Double-Deep-Q-Learning algorithm with Prioritized Experience Replay from scratch.

License

Notifications You must be signed in to change notification settings

Skeletonboi/dueling-ddqn-atari-rl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

49 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

dueling-ddqn-atari-rl

Implementation of the Dueling-Double-Deep-Q-Learning algorithm with Prioritized Experience Replay from scratch. Solution was iteratively extended from my DQN implementation, to include:

  • Double Q Learning - a second value network with delayed weight updates to perform the bootstrap estimate of the expected cumulative reward (i.e. error term) to reduce value overestimation
  • Dueling Networks - replacing the direct projection of state-action Q-values with separate projections for state-value and action advantage respectively

The DDQN training and testing was performed in both OpenAI gym and Atari environments, however Dueling DDQN was only trained for OpenAI gym environments due to limited compute.

Original goal is to expand upon this algorithm until it reaches the Rainbow DQN architecture from DeepMind.

About

Implementation of the Dueling-Double-Deep-Q-Learning algorithm with Prioritized Experience Replay from scratch.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages