Author: Uri Blass
Date: 11:34:55 01/15/01
Go up one level in this thread
On January 15, 2001 at 14:26:23, Jon Dart wrote: >On January 15, 2001 at 12:24:10, Uri Blass wrote: > >>It is possible to use some games and a lot of computer time to genrate a test >>suite. >> >>The idea is to give some top programs to analyze all the positions in the games >>and to give them 24 hours per position. >> >>The test suite can include all the positions that there was an agreement between >>all the top programs about the best move after 24 hours of search when part of >>them needed more than 3 minutes to converge to the best move(otherwise the >>position is too easy). > >To have a good test, the selected move needs to be better than other >alternatives by a significant amount. If, say Rb8 is good but Rc8 has very close >to the same score, then you can't really say that a program which selects Rc8 is >wrong. It could do just as well in an actual game as one that selected Rb8. I guess that in most of the cases that all the programs converge to the same move there is a good reason for it(the difference may be small but if all top programs can find the same move by a long search of 24 hours the program that can see it in 3 minutes has probably better positional understanding) 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.