Author: Robert Hyatt
Date: 18:56:14 08/25/00
Go up one level in this thread
On August 25, 2000 at 01:29:22, Uri Blass wrote: >On August 25, 2000 at 01:17:06, Alvaro Polo wrote: > >>On August 24, 2000 at 23:17:53, Bruce Moreland wrote: >> >>>On August 24, 2000 at 22:48:47, Michael Fuhrmann wrote: >>> >>>>On August 24, 2000 at 22:01:39, Bruce Moreland wrote: >>>> >>>>>On August 24, 2000 at 18:53:53, Eran wrote: >>>>> >>>>>>Different chess programs use different processor speed, and the results may not >>>>>>be fair. If Junior uses Pentium III 1000 Mhz instead of 700 Mhz, Junior might be >>>>>>a bit stronger and belong to the group of top chess programs, Shredder, Fritz, >>>>>>and Nimzo. I assume that if all chess programs use the same processor speed >>>>>>exactly, for instance Pentium III 1000 Mhz, the wmcc results will be enough >>>>>>reliable and fair. Do you agree with it? >>>>>> >>>>>>Eran >>>>> >>>>>As for "fair", this is not a uniform platform tournament. If you do a uniform >>>>>platform tournament, you get "fair". If you go to one that isn't, and you >>>>>expect "fair", it's better to stay home, because it won't be. It's not uniform >>>>>platform. >>>>> >>>>>If you expect "reliable", even in a uniform platform tournament, you aren't >>>>>going to get that, either. If you deduce a perfectly accurate rating for each >>>>>participant, and simulate the tournament a few dozen times, you'll get wildly >>>>>different results. The "best" program won't win every time. The "best" program >>>>>might not even finish in the top half. >>>>> >>>>>bruce >>>> >>>> >>>>Sounds like there's no good reason for holding the event at all, since the >>>>result doesn't tell us anything meaningful. >>> >>>Right. There isn't a reason to hold the World Series, the Olympics, the World >>>Cup, or any other sporting event, either. >>> >> >>It is curious what you say. In plain chess (not computer chess), if there is a >>tournement where Kasparov participates in, you can expect him to win. Sometimes >>he wont, of course, but I don't think it is probable that he won't finish on the >>top half. It looks like chess is much more reliable than computer chess. I >>wonder why. > >If the difference in the rating between the first program and the other programs >is big enough then you can be practically sure that the top program is going to >finish in the top half and the same is for humans. > >Uri I don't see how. Try flipping a coin 5 times, and repeat it 100 times. In more than one of those 100 trials, you will get 5 heads or 5 tails... which is statistically correct on one hand, and statistically impossible on the other. If you play 10 tournaments in a row, the best program will tend to do better overall. But just one tournament? Or just one game? Too much statistical variance.
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.