Author: Vincent Diepeveen
Date: 06:43:54 03/01/01
Go up one level in this thread
On March 01, 2001 at 03:04:44, José Carlos wrote: >On February 28, 2001 at 16:41:49, Robert Hyatt wrote: > >>On February 28, 2001 at 16:23:35, Sune Larsson wrote: >> >>> >>> [D]8/5b2/p2k4/1p1p1p1p/1P1K1P1P/2P1PB2/8/8 w - - 0 1 >>> >>> >>> This is one of Averbakh's positions from 1954. >>> It's a typical winning position in a good versus bad bishop ending. >>> The black pawns on h5, f5, d5 and a6 are vulnerable and curtail the >>> movements of the black bishop. To seal black's fate, all white need to do >>> is lose a move. That is - repeat the initial position with black to move. >>> This may be accomplished as follows: >>> >>> 1.Be2 Be8 [1.-Bg6 2.Bd3 Bh7 3.Bf1 leads to instant zugzwang, be it after >>> 3.-Bg6 4.Bg2 Bf7 5.Bf3 or after 3.-Bg8 4.Be2 Bf7 5.Bf3] >>> >>> 2.Bd3 Bg6 3.Bc2 Bh7 4.Bb3! Bg8 5.Bd1 Bf7 6.Bf3! and so on >>> >>> >>> Test: The above position is won for white so your program should be able >>> to win it. The evals should differ distinctly between white and black. >>> >>> Sune >> >> >>This is a bad position to test. It doesn't test knowledge at all. I just >>modified crafty to use a pure material-only scoring, at at depth=12, after a >>second or so, it sees winning a pawn. IE this isn't about knowledge, it is >>about search, and as a result, _any_ program should handle this correctly >>on almost any hardware... > > Maybe the good thing of this position is to check static eval. So that the >test could be: "the program should eval this position as very good for white >because of the bishops and the pawn structure". > > José C. Now add 1 pawn to g6 and g3 and retest this position again. Also move pawn from h5 to h6 and retest. He's asking your program also to understand things about kings and things about closedness and entering the position which no prog will clearly know in the next few years either... A search from under a secondhere however brings some interesting things...
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.