Author: martin fierz
Date: 00:53:31 11/25/03
Go up one level in this thread
On November 25, 2003 at 02:59:12, Mig Greengard wrote: >Otherwise it seems to me that you >just try to cover up the weaknesses of your program by tweaking the book to >avoid the positions it doesn't handle. And that should be contrary to the goal >of making a good chessplaying machine. agreed - but it's the "quick fix" and there are many others in the same direction: evaluating the position as bad if there are many pawns (or many blocked pawns / pawn chains) on the board, evaluating king safety asymmetrically (being more afraid about your own king in order to recognize long-term king attacks by the opponent), then there are some ideas of hard-coding moves to be bad, e.g. crafty recognizes trojans (white B/N on g5, white pawn h4, black pawn h6, black castled) and will not capture on g5, even if it might be good. fritz had an anti-nemeth code which stopped it from capturing *free* pawns on e4 to avoid the nemeth gambit. all of these things are short-term fixes to computer chess problems. all of them work somewhere between reasonably and extremely well, but i feel uncomfortable with all of them. they are all trying to avoid getting into positions the machines don't understand, instead of making the machines understand these positions (of course that's much much harder...). >Have their been significant projects >without books? Or with computer-only generated and tuned books? in computer checkers, we are generating our own opening books by a process called dropout expansion. it works very well there. the idea is based on a paper by thomas lincke, who tried building such a book for chess using crafty (probably version 15 or 16 something) - which didn't work at all. he used a cluster of computers at the university to compute a couple of millions of positions and the book was terrible. i don't think it had anything to do with crafty - it has much more to do with the game of chess itself: in more or less any normal opening, you can play h3 as white and lose a tempo. chess engines will think that this drops perhaps 5 centipawns in evaluation or so, but nothing dramatic. so this automated process of looking at all "decent" moves within a certain limit will also have to look at such a move - we humans will say: this wastes a tempo and achieves nothing, end of the story. checkers is a much more forcing game, and therefore this method works much better there. i thought it would help to let the chess book search more forcing lines, but when we had it look at a ultra-sharp variation of the dragon, it produced nothing but crap :-( cheers martin
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.