Some exploration scripts and notebooks into RL world with OpenAI/gym and Keras or Pytorch. Keras-RL is also explored along with my proper DQN implementation.
The toy example environment chosen is the Taxi-v3 for its simplicity and the possibility to work directly with a finite length Q-table.
This project requires Python 3.7 and pipenv. One simple way to set this up, is to use the provided devcontainers spec that can be opened with VSCode or another tool that is supporting it.
Then run
pipenv installto install the required python dependencies and
pipenv shellto activate the virtual environment.
Then start a training with
python debug.py- https://towardsdatascience.com/reinforcement-learning-lets-teach-a-taxi-cab-how-to-drive-4fd1a0d00529
- https://www.learndatasci.com/tutorials/reinforcement-q-learning-scratch-python-openai-gym/
- https://levelup.gitconnected.com/build-a-taxi-driving-agent-in-a-post-apocalyptic-world-using-reinforcement-learning-machine-175b1edd8f69
- https://medium.com/@anirbans17/reinforcement-learning-for-taxi-v2-edd7c5b76869
- https://towardsdatascience.com/reinforcement-learning-explained-visually-part-4-q-learning-step-by-step-b65efb731d3e
- https://towardsdatascience.com/reinforcement-learning-explained-visually-part-5-deep-q-networks-step-by-step-5a5317197f4b