NOTE: This repository was initially created as part of an exam project,
but has recently undergone a major refactoring.
Currently, we're in the process of documenting the project and creating
some templates/examples of how you can extend it and implement
your own agents and training strategies. If you're interested, please come
back in a a month or so!
The script "training_script.py" serves as a basic template for learning an
agent to play the game.
The script defines the function train(), which can train any two implemented
player classes against each other. Ie., you can also use the function both for
training one agent against an opponent using the Minimax-strategy.
The script should give an idea of how to write training scripts using our
environment.
We use Neptune to track the training progress of agents. To use our logging
functions, you therefore need a Neptune user and write access to a project.
You also need to have your personal API token set as a system variable on your
PATH.
See the Neptune docs for
instructions.
This project was built using Python version 3.9.13.
The used packages are listed in requirements.txt and req_no_cuda.txt,
which can be used to create a virtual environment as described
here.
Use requirements.txt if you have a CUDA-enabled GPU available for
computations and use req_no_cuda.txt if not.