alternative eval functions and tuning

[tissatussa]

i just saw your video "The Fascinating Programming of a Chess Engine" at https://www.youtube.com/watch?v=w4FFX_otR-4 and so i found Sophia. I was able to compile the engine and it runs fine in CuteChess (i'm on Xubuntu 22.04).

the video is well done!
it gave me an inspiration regarding that evaluation function.

we can define many different values for a piece, eg. not default P=1 BN=3 R=5 Q=9 but maybe P=1 N=3.1 B=3.5 R=4.8 Q=9 .. same for PST's and bonus points, which are often used methods.

then arrange a serious puzzle tournament among those engine versions : let them solve many puzzles (custom usage of EPD Test Suites) within a certain time or depth, then compare the results statistically (did it find the solution?), then conclude which code versions performed best at which type (!?) of positions and finally : try to create a function which determines the "position type" and thereby which set of values to use for finding a "best move" in that piece constellation.

i know "position types" are difficult to distinguish, i'm a club chess player myself for years .. esp. closed pawn structures are a "game changer" when trying to reach higher depth.

btw. the values (and rules) of the old and famous TuroChamp machine / engine can be considered also, they lead to human-like play! See eg. https://www.chessprogramming.org/Turochamp

what do you think ?
are you ready for an adventure ?

[bartekspitza]

Hello!

Glad to hear you liked the video. As you've figured out, the piece values and PSTs are really rudimental. This repo is not currently active however, I simply lack the time right now. Who knows, maybe in the future I'll come back and work on it, but in the mean time, feel free to fork it :)