Skip to content

A Rainbow implementation with PARL and Pytorch backend

License

Notifications You must be signed in to change notification settings

TonyLianLong/Rainbow

Repository files navigation

An Rainbow implementaion with PARL

This is an rainbow implementation which uses PARL and its backend for PyTorch. Currently you could run the following command to train and test:

python main.py

After 20000 steps, you will likely have:

[07-01 18:19:54 MainThread @DQN_agent.py:190] Frame: 20000, Score: 169.2, loss: 4.88

[07-01 18:20:02 MainThread @DQN_agent.py:179] score: 200.0

Demo

Start

Note that the angle becomes larger than 15 deg quickly. start

End

Note that even after 200 units of time the cart is still on the screen. end

Note

This project is heavily inspired by Rainbow is what you need. The PARL document and source code are also referenced in the development of this project.

In addition, PARL selects backend based on whether paddlepaddle is installed, and paddlepaddle backend has a higher priority compared to pytorch backend.

To use pytorch backend, you need to have pytorch installed and paddlepaddle not installed.

About

A Rainbow implementation with PARL and Pytorch backend

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages