wileyyugioh / reversi-ai

CSUS 2018 Reversi AI Project

An attempted implementation of n-step q-learning detailed @ https://coach.nervanasys.com/algorithms/value_optimization/n_step/index.html

Requirements for this project:

numpy

tensorflow

keras

To play text-based game:

python wytest.py

To train:

python wytrain.py

The model is saved as reversi3.h5

Misc Notes

Currently, the model has trained for ~150,000 games on my macbook.

I would have liked to get 80M, which is a nice number for our model to converge to a good approximation.