Author: Bob Durrett
Date: 20:15:58 12/27/03
Go up one level in this thread
On December 27, 2003 at 22:56:56, Johanes Suhardjo wrote: >To answer that question, I take gm2600.pgn, separate it into whiteall.pgn >where white wins and blackall.pgn where black wins. > >For whiteall.pgn, I run through the games. For every game, when white has >won material for 3 moves, I print out the fen of the position 5 moves earlier >(thus 2 moves before white wins material). The output is whitewins.fen. >For blackall.pgn, I do the same thing and come up with blackwins.fen. > >I don't know what I can figure out of whitewins.fen and blackwins.fen, but >some of you might find them useful. You might be able to extract some >information for piece square table, pawn structure, space control, etc. >Or you might be able to tune your evaluation parameters. > >Of course, this information is not accurate. Among the flaws I can think >of are: >1. Some games are won when the winning side has less material. But, > that's probably more search than evaluation issue. >2. The positions I print out may not be the critical positions. >3. The losing side might have just blunders the games (although not > highly likely with grandmasters rated higher than 2600 ELO points). > >The games are in http://www.nd.edu/~johanes/whiteall.pgn.bz2 and >http://www.nd.edu/~johanes/blackall.pgn.bz2. The "winning" positions >are in http://www.nd.edu/~johanes/whitewins.fen and >http://www.nd.edu/~johanes/blackwins.fen. > >Hope that's useful. > >PS. The extra two numbers at the end of the fen lines are the game and move > numbers where the position occurs. > > Johanes Suhardjo You may wish to examine "Understanding the Sacrifice: sacrifice your way to success" by Angus Dunnington, Copyright 2002, ISBN 1857443128. The underlying idea expressed in that book, which is about positional sacrifices, is that the top GMs are able to recognize positional factors which have value equal to or exceeding a certain amount of material. For chess engine programmers, this translates into the following: The ideal position evaluation code in a chess engine would be able to evaluate positions in such a way as to recognize and quantify positional factors. As an example, the code should be able to determine whether or not material should be given up for some other [positional] advantage or visa versa. The searching should be such that moves [from an internal node] offering material for other advantages [or visa versa] would be identified and pursued. What this means is that evaluation of games/lines based on material alone may not give you the information you want, at least not directly. "Realization of a material advantage" is a recurring theme in the chess literature. Sometimes it's "just matter of technique" and sometimes it takes brilliance. The same could be said for "Realization of XXX" where XXX is some sort of positional advantage. Incidentally, I see no reason why chess engines should not be able to play positional chess, including playing positional sacrifices. A positional sacrifice is MERELY a conversion of one type of advantage to another. Bob D.
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.