Author: Uri Blass
Date: 09:18:14 06/11/02
Go up one level in this thread
On June 11, 2002 at 09:47:31, Tina Long wrote: >On June 11, 2002 at 09:23:20, Jorge Pichard wrote: > >>On June 11, 2002 at 08:55:58, Tina Long wrote: >> >>>On June 11, 2002 at 08:27:01, Jorge Pichard wrote: >>> >>> >>>> >>>>You can run any program against itself 30 times or 1000 times and the difference >>>>would never be more than 30%. It doesn't matter how many times you let a match >>>>run up to, if the difference is over a margin of 30%, then you should stop it. >>>> >>>>Jorge >>>> >>>>Jorge. >>> >>>30 games, result 17-13, difference 30.7% = 17/13 -1. >>> >>>You say this would never happen. >>>I say it would happen around 22% of the time. >>> >>>Tina Long >> >>I miscalculated I meant to say Not over 50% like in Crafty vs Comet B45. The >>chance of this happenning is very limited. > >30 games, result 18-12, difference 50% = 18/12 -1. >You say this would never happen (revised to "very limited") >I say it (or greater than 50% difference) would happen around 16.6% of the time. > >Tina Long I think that it is dependent on the programs that you test. If the program that you test draw 90% of the games then I expect it almost never to happen. If the programs that you test win with white in 90% of their games and lose with black in 90% of the games then I also expect it almost never to happen(I assume that you are fair and gives every program 15 games with white. In the practical case I expect it to happen but I did not calculate how often it is going to happen. In the match between Fritz and Junior the result was 5-0 after 5 games and Fritz won the next 19 games 13-6. I believe that 18-12 has bigger probability than both 5-0 and 13-6 between equal programs and if you assume that the programs were not equal then it is clear that in at least one of these results the inferior program won. 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.