A Tak AI program with no human provided knowledge. Using MCTS (but without Monte Carlo playouts) and a deep residual convolutional neural network stack.
This is a fairly faithful interpretation of the system described in the Alpha Go Zero paper "Mastering the Game of Go without Human Knowledge".
All you need a PC with a internet connection. Thats it.
Head to the Github releases page at https://github.com/GeneralZero/TakZero/releases, download the latest release and launch TakZero.exe. It will connect to the server automatically and do its work in the background, uploading results after each game. You can just close the window to stop it.
Install the requirements from requirements.txt and run python MonteCarlo.py.
This project is in its early stages and will update with information later.
The code is released under the GPLv3 or later.