Author: Robert Hyatt
Date: 06:45:43 06/21/99
Go up one level in this thread
On June 21, 1999 at 02:59:33, Ingo Althofer wrote: >On June 20, 1999 at 22:44:34, Robert Hyatt wrote: > > >>note that their tiebreak was 1.0 points different. Note that shredder was >>seeded very low (incorrectly) which paired it against a good opponent in >>round 1, > >That is wrong. Shredder was seeded to low, but in round 1 it got Neurologic >which made altogether 1.0 points in the whole tournament. So without round 1 >Shredders lead in Buchholz would have been even higher. > Sorry.... I omitted (or deleted) one line: "consider the following possibility when seedings are wrong": and then the above still stands. IE a 1 point difference in tie-break is not significant after 7 rounds. And the initial seedings _can_ affect the final standing when the difference is only 1 point... I didn't look at the final standings, not wanting to try to decipher who benefited and who didn't, from initial seeding errors. The main place where sum-of-opponent's scores work nicely is with reasonably accurate seedings, in a small number of rounds, where one player does the 'swiss gambit' and loses in round one and then wins the remaining games and ties with someone that lost in a late round. The 'gambiteer' should have a lower tiebreak. In fact, round-by-round scores are often used to break ties also, as this also accomplishes the same thing... Someone ought to analyze statistically what the probability is that ifyou have a sum-of-opponent's score that is 1 higher than someone else that is tied for first after 8 rounds (counting the playoff draw) what is the statistical probability that that program is actually the best. IE maybe the FIDE approach makes sense if more long games can't be added... maybe another faster game would be better? sum-of-opponent's scores in a field of 30 after 7 rounds is not a real good indicator because you basically eliminate common opponents. And after 7 rounds, the two top programs will have several strong opponents in common, meaning the outcome depends on how two weaker programs do to give the tiebreak advantage to someone... > >>while ferret was paired high (correctly) and was paired against a >>weaker opponent in round 1. That could account for the 1.0 difference in >>tie-break instantly... >> >>IE nothing says that shredder played better opponents... try the tiebreak >>without round 1 scores and see what happens... and then decide whether the >>sum-of-opponents is meaningful when they are that close. Both played well. >>And a coin-toss would have been just as accurate to pick the better program >>since they drew twice...
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.