Skip to content
No description or website provided.
Branch: master
Clone or download
Latest commit 5945fab Apr 15, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
source change before train gif Apr 15, 2019
.gitignore add Apr 10, 2019
Environment.py first commit Apr 9, 2019
README.md training Apr 15, 2019
core.py training Apr 15, 2019
main.py training Apr 15, 2019

README.md

Implementation of Relational Deep Reinforcement Learning

This Repository is implementation of Relational Deep Reinforcement Learning to Breakout Environment.

The Reinforcement Learning Algorithm is Proximal Policy Optimization

Configuration

  • This paper requires heavy computation power.
  • Left Figure is the map of attention which is produced by self-attention.
  • Though the paper developed 100 environments for experiment, the implementer of this repository created only 16 environments with the limitation of computer resources. So sometimes it's exactly the performance and sometimes it's not.
  • If you want to see more significant attention map, just control CNN function to have less strides and more filters. In this repository, 84, 84 images are processed to have 19, 19 because of my computation limit.

Initial Training status

During Training

Tensorboard

You can’t perform that action at this time.