Author: James Robertson
Date: 17:46:32 05/17/99
Go up one level in this thread
On May 17, 1999 at 03:08:59, Christophe Theron wrote: >On May 17, 1999 at 02:45:51, Jouni Uski wrote: > >>According to Chessbits magazine MCP9 beta is getting 70% against Hiarcs7 and >>also won Fritz5.32 by 4-1! > >4-1??? > >Is that a scientific experiment? > >Has the 70% score against Hiarcs7 been done in a 5 games match too? > >Remember that in order to compare chess programs that are close in strength you >need A LOT of games. I would give no credit to a match with less than 30 games. >If they are very close, you will not be able to decide even after 100 games. > > >>Is this book tuning or what? > >I would call this lack of games. > > >>BTW I remember funny >>article in CSS magazine after Paris WMCCC - almost EVERY programmer told, that >>current beta has positive score against ALL opponents, heh, heh, heh... > >You get the point. :) > > > Christophe I did some calculations to see how many games I needed to play to accurately estimate different version strengths with my program. The result was rather discouraging: 360 games must be played to be 95% confident that the sample percentage of points is within [I cannot remeber, but this sounds right] 5% of the true percentage. James
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.