After trying to contribute to Stockfish, it’s time for me to try to build my own chess engine. This one won’t be UCI-compliant per se; it’s more like a library where you can add interfaces, etc later on. I plan to train the neural network and so on in Julia itself.
This engine is highly experimental. It’s neither A/B nor Monte Carlo, and instead is based on my own search algorithm. This engine would use a GPU. Also, instead of using SPRT to test and commit patches one at a time, I plan to use simulated annealing and/or something along the line to optimize for the combination of patches that play the best (which might not be the best approach), at least with the limited resources I have. It’s probably not very strong, but gonna be a fun ride.
You can submit your wacky ideas here.
With the way this engine is developed, even ideas that lose elos will probably get accepted lots of the time. More on that later, when I managed to build a framework that allows you to try wacky stuff. I intend this engine to be a breeding ground for wacky and potentially dubious ideas. Reject immediate elo gains or “proper” testing for creativity!
However, on the very off chance that all the mainstream chess engines are stuck in some local optima and I find something better and it ends up beating all the mainstream engines, I will conclude that we live in a funny timeline.
Stay tuned.
Tagging @Tord @Oscar_Smith the chess engine enthusiasts, to say the least, who also happened to be in the Julia community.