Author: Dave Gomboc
Date: 22:14:10 06/08/99
Go up one level in this thread
On June 08, 1999 at 23:50:20, Robert Hyatt wrote: >On June 08, 1999 at 16:13:28, Dave Gomboc wrote: > >>On June 08, 1999 at 14:45:29, Jeremiah Penery wrote: >> >>>On June 08, 1999 at 13:51:13, Dave Gomboc wrote: >>> >>>>On June 08, 1999 at 12:44:04, José de Jesús García Ruvalcaba wrote: >>> >>>>>What I most dislike is that nobody knows which are the strongest entries, so I >>>>>do not see how the accelerated pairings will help to match them more >>frequently. >>>>>I think one of the premises for accelerated pairings to work is to have a good >>>>>ranking of the players, like an established ratings list. But I remember >>>>>somebody said that in these tournaments the entries are ranked according to >>the TD's guesses. I do not think that is a good ranking. >>>>>José. >>>> >>>>There's plenty of background material to rank the players on, including the >>>>result of previous tournaments and, for some entrants, the SSDF list. It's not >>>>as good as it would be in a human tournament, but it is acceptable. >>> >>><snip> >>> >>>>Upsets happen, but since a reasonable ranking can be made before the event, it >>>>is okay to use accelerated pairings. >>>> >>>>Dave >>> >>>I don't really agree that 'reasonable rankings' can be determined before the >>>event in this case. Any previous performances were from older, presumably >>>weaker, versions of these programs, running on slower hardware(?), against >>>other, older, weaker programs on slower hardware. Not to mention that some of >>>these have NOT performed before. (Please correct if I'm wrong. :) >>>How will the TD choose the 'strongest' programs? Will he guess? Pick the >>>programs he 'likes' best? I see no clear way to choose. >>> >>>Jeremiah >> >>Commercial developers are generally good, amateur programs are generally not as >>good. Amateur programs with success in previous WCCC or WMCCC tournaments are >>better than those that haven't. Parallel versions of software running on >>multiple cpus can be expected to perform better than serial versions of the same >>software and serial software of similar strength. SSDF order might be used to >>sort serial versions of commercial programs. > > >totally wrong. In past wccc events commercial programs have not fared well at >all, with the exception of 1992 (no big iron there) and 1995 (only deep thought >participated from the 'big-iron' programs). My first two sentences did not preclude ranking amateur programs with past success above commercial programs. I'm not sure if my wording or your interpretation was "totally wrong", but in any case when I referred to "amateur programs" I was referring to a program like Comet, not something running on a 1024-node transputer or a cray. >ie of 8 wccc events, 2 were won by commercial programs... and that finish >was not 'easy'. I would seed Ferret above any of the commercial programs, >considering its hardware advantage, and how it plays on ICC. I would seed Ferret pretty high too. >IE the idea of seeding these programs is nothing more than putting numbers into >a hat. You "might" be able to divide the field into two groups and make some >good judgement about which half each program belongs in, but even that will be >'dicey' at best... > >> >>The goal is not to rank them perfectly. If that were possible, we might not >>even hold the tournament. The goal is to come up with a reasonable ranking for >>initial pairing decisions, and this is easily done. If 3 programs out of 30 are >>wildly misplaced, it isn't a big deal: the swiss system will take corrective >>action, as always. At the end of the tournament, no one will be able to claim >>that the winner played a bunch of weakoes. >> >>Dave > > > >that is the big point... the seedings are always so out of line, accelerating >the pairing accomplishes nothings. If the seedings are very good, then doing >this is very bad. If the seedings are random, doing this has no effect at >all. Actual circumstance lies in between, of course. and since the balance >shifts from 'no effect' for worst-case, to awful for best seeding, any sort >of accelerated pairing is basically bad. > >Note that the swiss system was not founded by a math dummy. It's been around >a long time. And the AP option was only created for large N sizes with small >numbers of rounds... This doesn't fit either... If the seedings are very good, using accelerated pairings will be _good_, not awful, because more games between the best playing entities will be played. I agree that if the seedings are downright horrible (e.g. random), then the pairing method used is inconsequential. If the seedings are "not bad" but "not perfect" (where I expect them to be), I do not expect worse results from using accelerated pairings than standard pairings. I have already rebutted your second paragraph in another post, specifically quoting the pairing rules from Canada's overseeing chess body (equivalent to USCF.) My investigation of the issue specifically changed my mind from thinking "boy, what a dumb idea" to "this is actually a pretty good idea". Restating your opinion isn't going to change my opinion back, but evidence showing that the use of accelerated pairings on a well-ordered group of 32 players with disparate strength in a 7-round tournament is actually bad would. 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.