davide97l / rl-traingenerator

Automatic code generator for training Reinforcement Learning policies

rl-traingenerator

Automatic code generator for training Reinforcement Learning policies:joystick:

Try it out:
https://rl-traingenerator.herokuapp.com/

Generate custom template code to train you reinforcement learning policy using a simple web UI built with streamlit. It includes different environments and can be expanded to support multiple policies and frameworks with an high level of flexible hyperparameters customization. The generated code can be easily downloaded as .py file or Jupyter Notebook so to immediately start training your model or use it as a baseline for your project. The backbone of this project has been taken from this repository where you can contribute with your own template.

Features

Supported frameworks:

• Tianshou

Supported policies:

• DQN
• More to come...

Compatible environments:

• Atari: Pong, Breakout, MsPacman, ...
• Classic Control: CartPole, Acrobot
• Box2D: LunarLander

---

Usage

Clone this repository and set up the environment:

git clone https://github.com/davide97l/rl-traingenerator/
cd rl-traingenerator
pip install -r requirements.txt

Run it locally:

streamlit run app/main.py

Deploying to Heroku:

heroku create
git push heroku main
heroku open

Update deployed app:

git push heroku main

Support

If you found this project interesting please support me by giving it a ⭐, I would really appreciate it 😀

rl-traingenerator's Issues

Next steps

• A2C
• PPO
• Generate file atari_wrapper.py