Skip to content

xffxff/endorphin

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

60 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

endorphin

Dopamine is my favorite research framework of reinforcement learning, for its very readable code and easy to try out new ideas. This project is mainly inspired by dopamine, and can also be said I am learning coding from dopamine.

My design principles are:

  • Flexible development: Make it easy to try out research ideas.
  • Concentration: Only focus on policy gradient or actor critic style algorithms.

Also inspired by:

Prerequisites

Ubuntu

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev

Requirements

Training models

python -m endorphin.atari.train --agent_name=<name of the agent> --base_dir=<name of the base directory> --game_name=<name of the game>

Examples

To train a fully-connected network controlling Classic Control using ppo

python -m endorphin.classic.train --agent_name='ppo' --base_dir=/tmp/endorphin/ppo/CartPole --game_name='CartPole-v0'

To train a conv network controlling Atari using a2c

python -m endorphin.atari.train --agent_name='a2c' --base_dir=/tmp/endorphin/a2c/Breakout --game_name='Breakout'

References

Marc G. Bellemare, Pablo Samuel Castro, Carles Gelada, Saurabh Kumar, Subhodeep Moitra. Dopamine, https://github.com/google/dopamine, 2018.

Mnih et al. Asynchronous Methods for Deep Reinforcement Learning. ICML 2016

Schulman et al. Proximal Policy Optimization Algorithms

Releases

No releases published

Packages

No packages published

Languages