Computer Chess Programming from the Ground up

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
Rebel
Posts: 7025
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Computer Chess Programming from the Ground up

Post by Rebel »

This is an initiative from Discord and Talkchess programmers to produce a document which is intended to be a guide for starters, advanced and expert chess programmers including the latest technologies in programming and testing.

We will start with the starter section and when finished it will be published here.

In the hope it will be useful.
90% of coding is debugging, the other 10% is writing bugs.
chrisw
Posts: 4346
Joined: Tue Apr 03, 2012 4:28 pm

Re: Challenge to the DISCORD invaders

Post by chrisw »

SPSA is hardly difficult, it’s also explained both on the wiki and in “Stockfish tuning method” in the wiki.
I used it in 2003 without knowing what it was called, without any knowledge of it, working out the idea from basic principles. Backgammon engine using neural net, I didn’t know how to do back propagation at the time, so generated a net from random weights given random perturbations, playing fast game batches and keeping a fraction of the perturbation of the winning weights. Worked well enough.
Some things are kind of obvious and people independently think them up. “Modern chess programmers” hardly have to think of anything because all wheels are already invented and perfected.
My 10 Elo came from SPSA on the five piece type values (used in various places in search), they were already “tuned” from years ago, so the 10 Elo was basically from an SPSA fine tune.