Author: Severi Salminen
Date: 10:27:24 11/24/00
Go up one level in this thread
>>It really makes me happy to see someone understanding that 10 or 20 games are >>not enough to find who's better. And when programs are very close in strenght >>(+-50 Elos) we need many games. 300 might be statistically significant, so thank >>you for not makeing this another "Nimzo beat Fritz 7-3" match! >> >>I'm looking forward to see results. 150-150 is my guess. >> >>Severi > > >It would also be good if the numbers were correct. >Games played 42. not 72. Games left 258 not 228. >Also how does 300 say become "statistically significant" >when the games are 30 min, ponder off (assumed). > >Myself I would have more faith in the results of >the 10 games if they were autoplayed at tourney controls. It means that after the match we can say that: in 30/G games Nimzo 8 scores approximately x% against Fritz6 with ponder off. Of course this means nothing in other time controls and also if we use ponder. 10 games means nothing, when programs are close in strength. Do you really think that 10 games match at tournament time controls give better info on strength comparison between 2 programs than a 300 game match at 30/G? Interesting. 10 games in torn. t-control gives no info at all about 30/G strength and very little info about torn. strength. Severi
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.