Author: Ernest Bonnem
Date: 10:38:31 04/22/05
Go up one level in this thread
On April 22, 2005 at 13:06:00, Kurt Utzinger wrote: >On April 22, 2005 at 11:34:46, Ernest Bonnem wrote: > >>My experience is that even with Nunn(1) if you repeat the 20-game match you can >>obtain quite different results. For instance 14-6, 11-9, even 9-11 ! >>(S9 against Fruit-Toga, 15mn to 30mn KO) > > Hello Ernest > Your statement somewhat surprised me. I remember a test > I made some years ago with the Nunn2 positions and I got > identical results when repeating the 40 games matches. > It's however very important to delete all learn files > and not to use a blitz time control where the engines > must often move "without thinking" at the end of a game. > Kurt Well, Kurt, I was disappointed too by the different results. I analyzed the case of games with different outcome : of course, sooner or later in the game you get a move which is different, maybe the 23th, or the 28th... And you can see that between these 2 moves, the eval difference is very small. For S9 UCI, there was no book and no position learning, for Fruit-Toga (UCI)there is no book and learning anyway. Maybe 15mn or 30mn KO, without time increment, is too small (AMD 2000+). The first matches were with ChessPartner GUI, the last ones with ChessBase GUI. When you did your Nunn2 matches, it would be interesting to know how many duplicate games you obtained. And I should make an experiment in playing 10 times the same Nunn position...
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.