PySkat implements a subset of the Skat card game: a suit game where ♣ is the trump. It uses C++ code that is wrapped to Python using pybind11.
It also provides a PlayerTrainer class that implements a neural network using Tensorflow which learns to play using the policy gradient method.
To build from source, you need: Pip >=10, boost, googletest, a C++ compiler. Clone the git repository, navigate to the root folder and run
pip install .
The root directory contains two scripts to play against a neural network model and to train a model.
Play against the neural network using the pre-trained model contained in skat_model.h5:
python play_against_model.py --model skat_model.h5
The pre-trained model seems to be able to make legal moves consistently. Its actual skill is another matter.
Train the model, starting from the pre-trained state:
python train_policy_gradient.py --start-model skat_model.h5
Run, in the root folder,
python setup.py test