Author: Dave Gomboc
Date: 09:19:45 11/30/99
Go up one level in this thread
On November 30, 1999 at 02:01:16, Dann Corbit wrote: >On November 30, 1999 at 01:28:49, Dave Gomboc wrote: >[snip] >>>To use the Swiss in this way (without proper seeding) is incompetent. >> >>That's a pretty strong statement. >I meant mathematically, not from a business standpoint. They may or may not >amount to the same thing. > >>If Sampras and Agassi are playing against a bunch of weakoes, and they happen to >>be paired in the first round because the organizers didn't have a clue who they >>were and assumed everybody was equally bad, then after the first game, both >>players would likely go on to win the rest of their games anyway, and finish >>1-2. > >But suppose that all the other entries are random also. Maybe one chain of >events would be rafter -> agassi -> sampras and another wimp -> doofus -> dolt. >Even that single event renders the contest unfair. In Swiss format, every error >in seeding is magnified greatly since there are so few matches. The swiss system is about finding a winner, not ranking everyone in the tournament exactly correctly. It will still do this in log_base_2 (number_of_players) rounds. If you want accurate second-place, third-place, et cetera, you need to play a few more rounds than n, but that's true even if you did have reliable pre-event rankings. >The point is that the Swiss algorithm has internal assumptions about the >participants: >Either they are all exact peers >Or: >The strengths are known and the weakest is pitted against the strongest, the >second weakest against the second strongest and so on (or at least a near >approximation). This isn't how swisses are typically paired in chess. The population in a score group is divided into two, and the top person in the group plays the fellow just below the median of the group. The fellow just above the median of the group plays the bottom person of the group. >If this pairing system is used and these assumptions are violated, then the >tournament is unfair. The only real benefit of Swiss over Round-Robin format is >that you can greatly reduce the number of comparisons (matches) to complete the >data structure. However, since we only did a small fraction of the matches >needed for a Round-Robin where everyone plays everyone there is information >missing from this format. That missing information is implicit in the two >assumptions above. > >That's why I said that Swiss -> Round-Robin is backwards. >If anything, it should be Round-Robin -> Swiss. At least that is the case for >chess programs because contests invariably have new or modified entries and the >strength is an unknown. > >If Swiss is doable, then Round-Robin seems redundant. Round-Robin -> Swiss just doesn't make sense. If you have already had everyone play each other once (or twice, if it's a double round-robin), why would you then have a swiss for the top spots? Your tournament is already over. Dave
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.