A reinforcement-learning agent to compete in C1Games' Terminal.
Feature goals:
- Initialized from high-level replays, optimized on self-play.
- Lightweight and fast on CPU.
- Competitive on the Terminal leaderboard.
-
git clone https://github.com/wz-ml/termite.git
-
Create a new conda environment:
conda create -n termite python=3.11
conda activate termite
- Navigate to the
termite
directory and install thetermite
package.
pip install -e .
Note: Avoid using
python setup.py
as it doesn't install the package in editable mode, which is required if you don't want to reinstall the package every time you update its files.
Termite is currently under development. If you spot a bug or have an idea, please feel free to open an issue on Github.
Todo:
- Port the Terminal environment from java to Python.
- Verify that the ported Terminal matches the behavior of the real game.
- Write test cases with Pytest.
- Build an OpenAI Gymnasium wrapper around the environment.
- Train a baseline model.
Thank you for contributing! Please make a fork of the directory and submit a pull request. Type hinting and reST docstrings are recommended.