Author: Rafael Andrist
Date: 10:35:44 04/08/02
Go up one level in this thread
On April 08, 2002 at 13:11:56, Robert Hyatt wrote: >On April 07, 2002 at 10:41:27, Michael Fuhrmann wrote: > >>[D] 1b6/8/8/7p/6k1/6P1/8/6K1 w - - 0 1 >> >>Crafty, without tablebases, scores this as about -4.5 for white. Shouldn't the >>score be 0.0? > > >Here is what is wrong, after a lot of analysis. > >1. static eval here is 0.00 as it should be. > >However, the search reaches a position where there are two things that >can happen. It is possible for black to zugzwang white into playing g4, >which lets black capture hxg4 and leaves this in a won endgame for black. >Unfortunately, if black plays hxg4 then white is stalemated, but the evaluation >doesn't understand this. It therefore concludes, by pushing this to the end >of any PV, that black can force white to play g4 and get out of the drawn >rook pawn ending. It doesn't realize that for black to do this, white will >be stalemated after black plays hxg4 and so the draw is still there... > >There isn't an easy way to fix this, and it simply shows the weakness of the >current search methodology... > > > >Here is the problematic position: > > +---+---+---+---+---+---+---+---+ > 8 | | | | | | | | | > +---+---+---+---+---+---+---+---+ > 7 | | | | | | | | | > +---+---+---+---+---+---+---+---+ > 6 | | | | | | | | | > +---+---+---+---+---+---+---+---+ > 5 | | | | | | | | *P| > +---+---+---+---+---+---+---+---+ > 4 | | | | | | | P | | > +---+---+---+---+---+---+---+---+ > 3 | | | | | *B| | | *K| > +---+---+---+---+---+---+---+---+ > 2 | | | | | | | | | > +---+---+---+---+---+---+---+---+ > 1 | | | | | | | | K | > +---+---+---+---+---+---+---+---+ > a b c d e f g h > > > >White was just forced to play g4 (the pawn was on g3) since white can't move >his king. If black plays Kxg4 then Crafty knows this is a draw. But at the >last move in the normal search, then black can play hxg4 and the search won't >realize it is a stalemate since the q-search is called next. As the search >goes deeper, black moves the bishop to let the white king go to g1, then you >check and the king has to go back to h1 which burns one ply. You repeat until >the king goes to h1 and hxg4 is done in the q-search which doesn't (in Crafty, >anyway) detect stalemate... > >cute position... With bitboards as you use, a stalemate detection in eval for a lone king wouldn't be difficult. regards Rafael B. Andrist
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.