RLjs is forked from REINFORCEjs by Andrej Karpathy, but more modualized with React in ES2015/ES6. It tries to serve mainly as a playground for learning reinforcement learning (RL) related algorithms. The two main parts are:
- The agents as
in
src/lib/agents/, which implement key RL algorithms such as dynamic programming(Chapter 4), Sarsa, Q-Learning (Chapter 6, Chapter 7). - The grid world which is taken as the example to illustrate how different agents/RL algorithms work.
Currently, only tabular methods have been covered. Approximation methods are planned to be developed developed in the future.
The draft app is available at https://rljs.herokuapp.com/.
There is also another repo for hosting machine-learning related notebooks I created when learning/implementing algorithms at https://github.com/zyxue/sutton-barto-rl-exercises.
A screenshot
The development setup is adopted from create-react-app.
git clone [email protected]:zyxue/rljs.git
cd rljs
yarn install
yarn start
This will generate the static web site and copy it to the docs folder
that is deployed to the web site when you push to GitHub.
yarn build
yarn test