Author: Uri Blass
Date: 08:51:46 06/23/04
Go up one level in this thread
On June 23, 2004 at 11:38:24, Joachim Rang wrote: >On June 23, 2004 at 11:28:47, Fabien Letouzey wrote: > >>On June 23, 2004 at 11:09:33, Joachim Rang wrote: >> >>>but you aren't exploiting all possibilities of that if you don't tell Fruit that >>>not all perpetuals lead to 3-fold-repetition: >> >>>[D]2R5/p3k3/1p2qrpp/1P2Q3/8/4PP1P/5K2/8 w - - 0 1 >> >>>here Fruit does not see, that Qb8 leads to a perpetual. >> >>>regards Joachim >> >>How many do? >> >>Fabien. > > >okay bad example ;-) > >Here many find the draw in a resonable time: > >[D]7k/4QP2/7p/6rP/8/1K6/6r1/8 b - - 0 1 > >regards Joachim Movei see score close to draw only because of some history based evaluation. In the previous example it does not see it because one of the conditions to detect draw based on suspect to perpetual check is no conversion for some moves and black can capture a pawn after many plies of checks. The scores goes up to 0 and later goes down to small advntage for white. The history based evaluation should be improved to detect at least almost draw also from the previous position and not to have 0 score in the last position when it does not see the draw. It has the same behaviour also in other positions when it can have some local maximum(not necessarily 0 when it suspects perpetual) only to go slightly down later. Uri 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.