Author: Ron Langeveld
Date: 15:46:15 07/31/02
Go up one level in this thread
On July 31, 2002 at 17:34:43, Maurizio De Leo wrote: >On July 31, 2002 at 16:34:47, Gian-Carlo Pascutto wrote: > >>>Actually also #7 Junior is in the "confidence range". >> >>It isn't. (You can't simply add the error margins) >> >>sqrt (30^2 + 30^2) = 42 >> >>Fritz 7 is with >95% confidence better than Junior 7. > >You are right. > >With the ipotesis that ssdf ranges are based on a standard normal distribuition > >Fritz >average M1 = 2741 >medium square error s1 = 15,306 > >Junior >average M2 = 2689 >medium square error s2 = 14,796 > >so Z = (M1-M2) / sqrt (s1^2 + s2^2) = 52 / 21.29 = 2.44 > >and being this also a standard normal it leaves indeed way less than 5% >probability that Junior has the same strenght of Fritz. > >Maurizio > >P.S. Thank you for letting me take off a little rust from my math. >P.S.2 So after all the SSDF list isn't so unuseful : it rules out two big >pretender (Junior and Hiarcs) for the trone of best computer program. An interesting thought! Without a doubt Fritz7 is a very strong engine. I use it a lot. Compared to others it has a reliable evaluation except for a few endgame positions with opposite bishops. From a pure statistical point of view Fritz7 is very interesting. The results speak for themselves, however somehow i feel pleased by the fact that some engines are not able to show these results as well. Next to statistics there is this aspect of understanding crucial positions. Where evaluations of the same postion seem to drift apart things become really interesting. From my observations I can only conclude that these positions often proof that engines like Diep, Shredder, Hiarcs are more capable of smelling strategical errors by the opponent. Not that i want to diminish the importance of tactics, every top program deserves full credentials in this respect, and some people claim tactics determine 90+ % of the outcome, but even if Fritz is the uncrowned king in this respect, these other engines surprise me on more occasions with a dead-on evaluation in difficult positions than Fritz does. You may call this unpredictability if you like, but there are other aspects to consider besides statistics when you want to determine the best computer program. Aspects that may seem less objective, but with regard to objective measurements I don't hear a lot of people complain that engines use different (quality) books. Sometimes I wish there was a seperate SSDF list that is not compiled based on games but on a big collection of testpositions. If these positions reflect an appropriate amount of positional bonusses than Fritz will no longer top the ranks, that's my conviction.
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.