Author: Robert Hyatt
Date: 15:00:24 02/05/99
Go up one level in this thread
On February 05, 1999 at 00:28:11, James B. Shearer wrote: >On February 03, 1999 at 10:15:37, Robert Hyatt wrote: > >>On February 02, 1999 at 00:24:59, James B. Shearer wrote: >> >>>On January 31, 1999 at 10:43:10, Robert Hyatt wrote: >>> >>> <snip> >>>> >>>>No idea why it takes 16 plies however... but I'll look... >>>> >>> >>> A big part of the reason is crafty 16.3 is far too optimistic about g5. >>>This can lead to the positions like that shown below which crafty scores -2.35. >>>This is quite optimistic considering that material is even and the position is a >>>dead draw. Crafty 15.15 scores this position -1.19 (see below) which is still >>>optimistic but better. >>> >> >> >> >>I agree with the evaluation, in that if anybody wins, black wins. The >>protected passed pawn completely limits white's king to a few squares. >>But the fact that black's king is 'locked out' is an issue, which i try to >>resolve via search rather than by evaluation. On a normal machine, it can >>get to 16 plies very quickly and discover that it can force entry by pushing >>the other pawn... but it needs the depth to 'see' this rather than trying >>to 'evaluate' this... > > How can you resolve the locked out king by search? The position will >continue to score 2.35 or so until the 50 move or 3 rep rules kick in. This >means that to select h4 in preference to g5 (which leads to positions like this) >crafty must see more than that h4 is good, it must see that it is better than >the 2.35 (actually 1.71) it is incorrectly scoring g5 at. This takes much >longer than if the g5 positions were more accurately evaluated. Too long for >fitter so I drew a lost game. I think 15.15 would have won. > James B. Shearer Not at all. At 16 plies, it can see that h4 allows it to eat the kingside pawns which is all it needs. It doesn't have to find h4 just because g5 draws eventually, it can find h4 because it _wins_...
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.