Author: Andrew Dados
Date: 13:30:46 02/13/00
Go up one level in this thread
On February 13, 2000 at 16:02:27, Vincent Vega wrote: >On February 13, 2000 at 01:30:07, Andrew Dados wrote: > >> 90% of 'amateur' programmers do it to try out some ideas authors have in their >>mind. >>And 90% of programmers undertaking this waste of time realizes that neural nets, >>genetic alghoritms,(quantum computers?!? - how can amateur try that one?) suck >>when comes down to chess so they don't bother. Some pattern recognition can be >>found in most programs; while 'knowledge programs' are undef to me, I noticed >>alpha-beta can be found in 100% of programs both amateur and commercial.... My >>advice - try out your own.. preferably skip that boring alpha-beta crap. It's >>fun when you notice chess programming is one of least rewarding hobbies you can >>have :) >> >>-Andrew- > >Actually genetic algorithms already proved useful when it comes to determining >piece values. Amateurs can try quantum algorithms, there are no real quantum >computers except 1-bit ones anyway. And I already wrote a chess program with >alpha-beta years ago - not much knowledge could be fit in 40-odd KB of memory >though :-) Then it was a one-lane road, today it's a 5-lane highway. What I >would like to see is chess programs using methods that can be applied to other >games and AI fields, unfortunately most prefer to just tweak evaluation so the >knowledge gained can only be applied very narrowly. Take one, simplest example: basic ending KP vs K. Don Beal wrote a routine for that; looking at it I can see some 20-30 exceptions which need to be handled specially. Now applying neural nets or any genetic alghoritm would make my program learn that ending after some time, sure; it still would not get to play it perfect in sensible learning time. TDChess does learning using temporal diference algorithm; takes it a while to get to some sensible 'conclusions'. Also finger 'morph' on icc - it's rating is close to 1000. I wonder what rating would totally random-moving program get (or program limited to ply 0 /qsearch/ only). So is writing some 1-3 thousand lines for cute, state of art, hyphe techniques worth it? For me - maybe in 2 places: to adjust some very limited and dependand set of wages (like piece values you mentioned), or to try it to determine best book line upon played games so far. Everything else I can (and want) to adjust manually...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.