Author: blass uri
Date: 12:09:38 08/31/99
Go up one level in this thread
On August 31, 1999 at 12:18:41, Gerrit Reubold wrote: >On August 31, 1999 at 10:19:35, Robert Hyatt wrote: > >>On August 31, 1999 at 10:02:19, Gerrit Reubold wrote: >> >>>Hi all, >>> >>>Here are two endgame positions which are easy draws (for humans), however, my >>>program (Bringer) lost both. Does your programs play better moves? how long do >>>they take to see the draw? >>> >>>Position from Bringer 1.4 - Crafty 16.11, after 155. ... Bb5 >>> >>>8/8/2k5/1bp5/4p2p/2K1P2P/4B1P1/8 w - - 0 1 >>> >>>How long does your program take to avoid the losing 156. Bc4?? Instead, >>>sacrificing the g2 pawn with Bg4 and keeping the king on c3 should be an easy >>>draw. (Bringer would need 6 min on a PII 300 to avoid Bc4 :-( ) >>> >>> >> >> >>Crafty seens that Bc4 loses in about 30 seconds on my PII/300 notebook, >>after getting to depth=18. Score goes from -1.58 to -5.21 at that depth. >>The quad xeon sees this in 4 seconds however... > >Sure, searching faster an deeper helps. I hoped for an answer like "Do this ... >and that ..., and your engine will see it immediately". Maybe I have to accept >that there are still positions which humans solve easily and progs need some 18 >plies. > >> >> >> >> >>>Position from Crafty 16.11 - Bringer 1.4 >>> >>> - - 0 1 >>> >>>43. ... g5 would be an easy draw, does any program see this move (BTW, Bringer >>>played Kg6 and h5 and lost) >>> >> >> >>I'll have to try this one on the xeon, but it is busy. It has all the 3 vs 2 >>endings and may be able to search this to the end. >>> >>>The second position leads to the question: how should we evaluate positions like >>>this: >>>8/2k5/2n5/8/8/2BN4/2K5/8 w - - 0 1 >>>Don't tell me "5 pieces EGTBs", because there may be another black pawn, which >>>wouldn't change a thing. >>> >>> >>>Greetings, >>>Gerrit >> >> >>Tablebases _will_ help... because often a 20 ply search is more than enough >>to find a way to eliminate that last pawn and get an exact score... otherwise >>such evaluations are difficult, because _all_ KN vs KBN positions are not >>drawn... > >I was looking for a way to *know* it is a draw, not for searching 20 plies to >hit the tablebases. There are programs with some knowledge about these positions Junior evaluates the last position without tablebases only as a small advantage for white(clearly less than a pawn). I know that Rebel evaluated KN vs KBB in WCCC as only 0.xx for the BB(It lost the game when the opponent had tablebases and went for a winning position of KBB vs KN) Uri
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.