Slowpoke is a checkboard playing program for my 3rd year dissertation. It's inspired by Blondie24 but also includes a series of modifications that allow it to be a system that can be taken seriously in 2017. It revolves around GANNs (Genetic Algorithms / Neural Networks) and moves are evaluated using a modified Monte-Carlo Tree Search.

Training is called by running the python file:

python3 simulate.py heavy

The program will attempt to utilise as many cores that the computer running the program has. I'm currently running this on a 128-core machine, which takes around 20 hours to finish (200 generations, 15 players per generation, 6ply).

python3 play.py

The program above also allows arguments; so you can quickly test the system:

python3 play.py b=slowpoke w=slowpoke ply=8

python3 evaluate.py

There is currently a variety of configured games that slowpoke will try to win. They're currently set to play for 256 games (128 on both sides - black and white). Statistics are scored in root/results/evaluations in the form of a .json and a .csv.