Author: Gian-Carlo Pascutto
Date: 04:10:41 09/08/01
Go up one level in this thread
>On June 22, 1998 at 10:02:42, blass uri wrote: > [D]3k4/2p1p1p1/p1P1P1Pp/P6P/4K3/8/8/8 >>white to move. >> >>every human can understand it is a draw but my programs cannot evaluate >>the position as 0.00 >>Is there a program that understand it is a draw? >>( it is enough to use an hash table of 36*9*2=648 positions) >> >>Uri > >This is a good example to see what your hash table can't do for you. > >The problem is that all the positions here evaluate the same. None of them are >obviously drawn or obviously won, so the only way the search can find a draw >score is to find a repetition or 50 moves in the PV. White cannot be forced to >make a repetition, so the only hope is to search this to 100 ply, but this takes >too much time. Mine completes a 101 ply search here in 1.9 seconds, returning a draw by the 50-move rule. It actually made it through 161 plies in 3.0 seconds, but then something broke. I had to relax my normal 40-ply maximum for this, so perhaps I need to increase another compile-time constant. Here is the PV at ply 101. PV 00:00:01.875 101 0 [] Kd5 Ke8 Ke5 Kd8 Kd4 Ke8 Kd5 Kd8 Ke5 Ke8 Kd4 Kd8 Kd3 Kc8 Kc2 Kd8 Kc3 Ke8 Kb2 Kd8 Kc2 Ke8 Kc3 Kd8 Kc4 Kc8 Kb3 Kd8 Kb2 Ke8 Ka3 Kf8 Kb4 Ke8 Kc4 Kd8 Kb3 Ke8 Ka2 Kd8 Kb1 Ke8 Kc1 Kd8 Kd2 Ke8 Ke3 Kd8 Kf3 Ke8 Kf4 Kd8 Ke3 Kc8 Kf2 Kd8 Kg2 Ke8 Kg3 Kd8 Kf2 Ke8 Kg2 Kd8 Kh2 Ke8 Kg1 Kd8 Kf1 Ke8 Ke1 Kd8 Kd1 Ke8 Kd2 Kd8 Kc1 Ke8 Kd1 Kd8 Ke2 Ke8 Kd3 Kd8 Ke4 Ke8 Kf3 Kd8 Kg3 Ke8 Kh2 Kd8 Kg1 Ke8 Kf1 Kd8 Ke1 Ke8 Ke2 Kd8 bruce >If someone has CCC indexed, they might be able to find it if they look for posts >by me that contain "50" or "fifty". > >bruce Perl Power :) -- GCP
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.