My implementation of machine learning for minesweeper solver

Modified Q-Learning implementation from this article (http://cs229.stanford.edu/proj2015/372_report.pdf)
The future reward part of the equation is ignored as I am not interested in end game, but rather in immediate reward for current state.

Made in Unity 2018.2
I save states in scriptable object, whenever I tried to save more than 3000000 unity would get up to 7 GB of ram usage and crash, so I needed to divided into many scriptables with half million records each.The games where a new state was discovered were ignore in the win rate calculation.

the data occupies over 1.1GB on the disc and it takes a minute or so to launch the project :D.

The solver doesnt set the flags, the game is considered a win when all blocks except the bombs are uncovered.

Also this is first time I ever recorded voice so please be understanding.