neural net + genetic algorithm to play ttt
I wrote this in June 2018 when I was starting to look at neural nets. Looking back, my algorithm was incredibly dumb (the net was very small, and no backpropagation happened: the algorithm was evolutionary with unjustified "sexual reproduction") and the results weren't great. I think it had a score closer to 0.5 than 0.8 (perfect) against random play, but at least it made progress.
My more recent attempt with backpropagation seems dismal, not getting past a score of 0.6 despite a much larger net and lots of training. I still don't understand what is wrong with it.