gym-rdm

A Gymnasium environment for training agents on Random Dot Motion (RDM) tasks.

About

This project uses pygame to animate a set of dots in a more or less random fashion, implementing the Random Dot Kinematogram (RDK) psychophysics task. The animation algorithm is inspired by this repository.

The dot motion code is wrapped into a Gymnasium environment. Autonomous agents can be trained to recognize the main dot motion pattern through Reinforcement Learning.

The above demo was generated by the following command.

# Run the task unit tests as a standalone python program
python tests/test_task.py

Development notes

Toolchain

This project is built and tested with the following software:

• Poetry for dependency management and packaging;
• Black for code formatting;
• Pylint and mypy to detect defaults and mistakes in the code;
• pytest for testing the code.
• A GitHub Action to check and test the code upon each push.

Useful commands

# Reformat all Python files
black .

# Check the code for mistakes
pylint gym_rdm/* tests/*

# Run static type checking.
# The -strict flag is... stricter. It should pass on the codebase
mypy [--strict] .

# Run all code examples as unit tests
# The -s flag prints code output
pytest [-s]

License

MIT.

Copyright © 2024-present Baptiste Pesquet.