Skip to content

mthrok/simple_dqn

 
 

Repository files navigation

Tensorflow Simple DQN

This is Tensorflow version of Simple DQN by tambetm.

For the detail of simple DQN, please refer the original repositry.

DQN Network Architecture

Installation

You need to install Tensorflow in addition to the original requirements.

How to use

Training scripts and related commands should work without any modification.

For example, to train on GPU you can run: ./train.sh Breakout-v0 --environment gym --backend gpu

If you want to run the original Neon version, change the import statements in src/main.py as follow.

from deepqnetwork import DeepQNetwork  # instead of deepqnetwork_tf

from statistics import Statistics  # instead of statistics_tf

Note

  • Tensorflow version is not as fast as Neon Version. It is about 1.5 times slow.

TODO

  • Add tensorboard data
  • Add trained weight

About

Simple deep Q-learning agent.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • HTML 99.1%
  • Other 0.9%