I was interested in understanding the workings of AlphaZero. So back in 2017/2018, I read up on the papers and implemented it for the game of Othello (aka Reversi) in Javascript.
I didn't have time to wrap it a web app at the time, so I wanted to do so now. The idea is to create an AI Gaming platform where game providers can submit their AI games, starting with simple classic board games. Mine being the first.
I was able to get this working fairly quicky, but it order for it to work on a more set of board games (and one day beyond board games), much more needs to be done. Thus, it is a good time to pause on it at this state.
So, about the game. Luckily, I was able to find model weights so I didn't have to spend time and money to train to produce models. You can find them here. The system is however implemented in python, so I won't be able to make use of the code. Which isn't all lost, as I wanted to truly understand the AlphaZero algorithm anyway. Having a working code in python did help trumendously in understanding the algorithm, but it was still lots of roadblocks since it is the first time I implemented something from a scientific paper. Special thanks also for the board and piece styling from Lea Verou.
Ready to get defeated? Let's go!