ahavenoname/ERL

Exploration based Reinforcement Learning. (Montezuma Revenge)

ERL

Exploration based Reinforcement Learning

PixelCNN DQN

A tensorflow implementation of 'Count-Based Exploration with Neural Density Model'. It uses PixelCNN to measure the sparsity of one state and give corresponding bonus.

DQN

DQN is modified from 'DQN-tensorflow'

PixelCNN

PixelCNN is modified from 'gated-pixel-cnn'

How to use ?

python main.py --mode top-pixelcnn or python main.py --mode pixelcnn

To reproduce the result of args/6744.json, run python main.py --mode top-pixelcnn --test The autoencoder mode is not verified yet.

References

• PixelCNN++ paper with Code
• DQN Code