Skip to content

Pytorch implementation of the PAAC algorithm presented in Efficient Parallel Methods for Deep Reinforcement Learning https://arxiv.org/abs/1705.04862

Notifications You must be signed in to change notification settings

qbx2/PAAC.pytorch

Repository files navigation

PAAC.pytorch

Pytorch implementation of the PAAC algorithm presented in "Efficient Parallel Methods for Deep Reinforcement Learning". PAAC is the abbreviation of Parallel Advantage Actor-Critic.

Currently, because the PAAC network is not using LSTM, the evaluation result is not very good. I'm working on the LSTM version of PAAC (waiting for a new graphic card due to lack of current gpu's memory.)

The original paper is here: https://arxiv.org/abs/1705.04862

Requirements

PAAC.pytorch requires torch, torchvision, PIL, gym.

Libraries used in this project:

  • torch==0.1.12+32e6665
  • torchvision==0.1.8
  • Pillow==4.1.1
  • gym@797a25d1b1a8823b305fdb575c4378a5c288b432

Result (BreakoutDeterministic-v4 training log)

log

https://www.youtube.com/watch?v=6FMzNaL88wQ

About

Pytorch implementation of the PAAC algorithm presented in Efficient Parallel Methods for Deep Reinforcement Learning https://arxiv.org/abs/1705.04862

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages