Author: Robert Hyatt
Date: 18:35:53 03/27/03
Go up one level in this thread
On March 27, 2003 at 15:32:09, Matthew White wrote: >On March 26, 2003 at 11:34:59, Robert Hyatt wrote: > >>On March 26, 2003 at 11:11:30, Peter Berger wrote: >> >>>On March 26, 2003 at 11:02:06, Robert Hyatt wrote: >>> >>>>On March 26, 2003 at 06:28:52, Peter Berger wrote: >>>> >>>>>These can be tough or very easy depending on engine - some engines seem to know >>>>>too much here. They are also interesting for humans I think. >>>>> >>>>>6k1/p4p2/6p1/1P2p2p/4q2P/4Q1P1/P4P2/6K1 b - - 0 1 >>>>> >>>>>1..Qxe3! wins for black. >>>> >>>>I don't think these are about "knowledge". They are about "search" since the >>>>first position >>>>has blacking giving white an outside passed pawn with the two to one majority on >>>>the >>>>queenside. >>> >>>Maybe. Crafty is one of the engines that has quite a hard time with this >>>position as is Ruffian (that has even more problems). So I think it is about >>>knowledge, or maybe about tuning?! >> >>I don't think so, for the reason that white has what appears to be a won ending >>if >>queens are removed, because of the 2-1 distant majority. Crafty understands >>this >>perfectly well. And it takes it a while to see that the king is in just the >>right position >>so that black's eventual threats can't be met properly. >> >>Both have the same basic motif, a 2-1 queenside majority. The second is simply >>not a >>problem for Crafty because of its knowledge about the potential distant passed >>pawn. > >>The first is more of a problem in that it requires a deep search to discover >>that the >>distant passed pawn knowledge really doesn't apply. >> > >Couldn't most positional knowledge be found given a deep enough search? If a >program has the knowledge, a deep search isn't necessary. For example, a pawn >move in the opening may cause a permanent weakness that may not be exploitable >for 30 moves, however, without searching to that depth, a program might make >that move and never see the problem coming. > >(SNIP!) > >Regards, >Matt Of course depth is the answer to knowledge, because search is perfect knowledge within the depth you can reach. But since we can't go deep enough, we employ static evaluation tricks to try to make up for what the search can't see directly. And in this position, a normal "distant passer wins" idea fails just barely due to king placement and pawn placement. IE this is a pretty rare exception to a very common ending rule. And it takes the search to discover what the evaluation would have a _very_ hard time seeing.
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.