Author: Ricardo Gibert
Date: 23:47:34 09/16/99
Go up one level in this thread
On September 16, 1999 at 23:59:25, Robert Hyatt wrote: >First, a little background. I have been doing 'outside passed pawn' scoring >for years now, because I got tired of seeing Crafty lose endings where it was >a pawn up, and it traded down to the point where it was a king and pawns vs >king and pawns ending, where the opponent had an outside passed pawn that made >it an easy win... I was interested in a similar topic (for both middlegame & endgame), which is why I posted the following several days ago: Are their any estimates of how many distinct pawn structures get examined by computer programs over the course of a 5 minute blitz game? I wanted to make some approximate calculations of how much space a hashtable of such pawn structures would take, but I got zero response. Evaluating pawn structures for the endgame is tricky: W: Pa3,b2,c2; B: Pa5,b6. 1) Here White can Q a pawn. 2) Shift the a-pawn from a5 to b5 and now White cannot Q a pawn without sacing a pawn and allowing Black to also Q! This can be a problem with pawns on both sides of the board. Your pawn "majority" will let you down. 3) If Blacks pawns are each advanced one square (B: Pa4, b5), then if white tries to get a passed pawn, Black may Q ahead of White e.g 1 b3? (1 b4 is less bad) 1...b4! and Black will Q an a-pawn well ahead of White. I'm assuming of course, that other material on the board do not interfere. > >OK... that was fairly easy to code using bitmaps... and it has worked well. >But once you get past that hurdle, you begin to see endings where you trade >down to a pawn-up ending, but your opponent has a queen-side majority that >turns into an outside passer outside the search horizon, and the same issue >comes up again. > >I am working on addressing this now, and am looking for a discussion on what >might be the best way to do this. > >I have completed a fairly accurate 'candidate passed pawn' analyzer. It is >in the EvaluatePawns() code so that it is all hashable and won't cost a fortune. > >All it does is simply look at each pawn that has no enemy pawn in front of it, >and decides whether pushing that pawn can make a passer or not. Again, not >hard using bitmaps, and in studying the results, it looks reasonable. My intent >is to use this in the absense of any outside passed pawns for one side, to see >if it has any potential outside passed pawns on that side of the board. And >for the usual 3 vs 2 queen-side majorities, it works cleanly and accurately. > >But what about 4 vs 3? Where the passer ends up on the d-file, which might >not be far enough away to cause problems. Or what about 3 vs 3, where one >side has pawns on a-b-c, the other side has pawns on b-c-d, and both end up >with a passer although the abc passer will be more distant. Careful! It is very often possible to Q a pawn on almost any file and not just the d-file by sacing material. This is sometimes relevant. Analogously, this can also happen with 3 vs 2 and 2 vs 1. > >I guess my question is, has anyone given any thought to this? Or is anybody >even dealing with pawn majorities at present? I tend to not actually call this >majority code any longer, because it is _really_ candidate passed pawn >evaluation instead... > >My intention is to recognize that if the kings are on g1/g8, and white has >the a-b-c pawns and black just has b-c pawns, that this is a nearly winning >position. I am going to do just like I do with outside passers, that is, have >their value go up as material goes down, as they don't mean much with queens and >pieces on the board... > >Any comments, suggestions, ideas, etc? > >Obviously necessary, yet I don't see any evidence that any program does much >with this excepting for deep blue... > >Bob
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.