Tournament Structures

Discuss anything you like about chess related matters in this forum.
User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Tue Apr 10, 2018 1:25 am

Roger de Coverly wrote:
Mon Apr 09, 2018 11:37 pm
That view was obsolete fifty years ago when Stewart Reuben and others devised any number of deterministic systems for making pairings based on the rankings of players.
One of the references cited in that paper is this: http://emis.ams.org/journals/DM/v71/art3.pdf, which treats pairing a round of a Swiss as a stable roommates problem where the aim is to get a stable matching (no two players would rather play each other than the opponents they've got) while minimising both the maximum score difference and the maximum absolute colour preference. They don't do a very good job of it, but there are some numbers in there that could be fine-tuned to make it better.

I wonder what would happen if you also tried to minimise the deviations of the difference in ranking from the ideal value, which would be half the size of the score group. I might try that at some time that isn't 1:25am.

EDIT - I've knocked something up in Excel, and discovered a possible problem with treating pairing as a stable roommates problem: floats. If you had an odd number in the (1) score group of a Swiss, your downfloat would be forced to play someone from the (0.5) group, even if more stable pairings existed within the (0.5) group. A stable roommates solution would potentially lock in all the pairings of the (0.5) group and force the downfloat to play someone from the (0) group. More fiddling needed...
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

User avatar
IM Jack Rudd
Posts: 3862
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Tournament Structures

Post by IM Jack Rudd » Tue Apr 10, 2018 1:31 pm

Try minimizing the square of the maximum score difference. That should solve that problem.

Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Tue Apr 10, 2018 2:12 pm

Enormous tournaments can stress a pairing algorithm particularly amongst the players with a half point in round 2.

Take round 2 of the European Championships in Aix-les-Bains in 2011

http://chess-results.com/tnr45941.aspx? ... CO&flag=30 and look at the pairings of the players who drew in round 1.

I think it was suggested at the time that a minor rewrite of the FIDE rules was needed. Either that or the implementation of the rules had failed under the circumstances given.

For practical purposes and to avoid accusations of bias, a pairing method has to produce a unique solution. The "stable roommates problem", although parallel, doesn't do this according to the quoted material.

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Tue Apr 10, 2018 2:21 pm

IM Jack Rudd wrote:
Tue Apr 10, 2018 1:31 pm
Try minimizing the square of the maximum score difference. That should solve that problem.
Having re-read the Finnish paper, I think the issue is that the roommates algorithm doesn't know how to minimise that function, because it only compares two edges of the graph against each other without considering the knock-on effect.

Maybe it's possible to improve on the roommates algorithm. Then again, I've now tracked down the Olafsson article that both papers cite, and it seems pretty definitive. The Finns were trying to find an improvement to Olafsson where it didn't matter in what order you considered the score groups

"Contrary to Olafsson’s proposal, we reject the assumption that a pairing system should imitate the manual system described in the FIDE rules"

but if such an improvement exists, they didn't find it.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Tue Apr 10, 2018 3:01 pm

Roger de Coverly wrote:
Tue Apr 10, 2018 2:12 pm
Enormous tournaments can stress a pairing algorithm particularly amongst the players with a half point in round 2.

Take round 2 of the European Championships in Aix-les-Bains in 2011

http://chess-results.com/tnr45941.aspx? ... CO&flag=30 and look at the pairings of the players who drew in round 1.

I think it was suggested at the time that a minor rewrite of the FIDE rules was needed. Either that or the implementation of the rules had failed under the circumstances given.
Board 108, why has player 324 been paired with player 249? Do you remember what it was about the FIDE rules that caused that, or that they wanted to rewrite?
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

User avatar
IM Jack Rudd
Posts: 3862
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Tournament Structures

Post by IM Jack Rudd » Tue Apr 10, 2018 3:07 pm

My guess is that it's the usual computer problem of failing to look above the problem pairing to find ways to sort it out.

Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Tue Apr 10, 2018 3:18 pm

Chris Goodall wrote:
Tue Apr 10, 2018 3:01 pm
Board 108, why has player 324 been paired with player 249? Do you remember what it was about the FIDE rules that caused that, or that they wanted to rewrite?
The pairing method used would have built on the ideas described in the 1990 paper linked earlier. It's designed so that at least in principle it can be done without the use of a computer, which is likely to rule out goal seeking methods. Arbiters, or at least those involved with the specification and implementation of pairing methods should know the later status of the Aix feature. With players on half a point in short supply given the rating range of the tournament, the speculation was that the algorithm ran out of potential sensible opponents. The "best" pairing was likely the illegal one of players who had met and drawn in the first round.

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Tue Apr 10, 2018 4:14 pm

I hadn't looked at the July 2017 FIDE Swiss pairing rules. Wow, that's a lot of rules. I guess that makes all previous research into chess pairings irrelevant, then. Given the stipulation that there is now only one valid pairing and any human or computer applying the rules must arrive at that one.

EDIT - okay, so we can't compare random pairings with "the Swiss system as it's actually implemented in chess tournaments", but I still think it can be implemented better than the Wharton article managed.
Paul Dargan wrote:
Mon Apr 09, 2018 12:35 am
I have a feeling that GM Smerdon (a serious aceademic economist, when not playing/writing about dodgy Bg4 Scandi lines) has written about the efficiency (or otherwise) of swiss systems... perhaps on chess.com?

Paul
I can find various writings by GM Smerdon on tie-break systems (Fixing Flaws and Stopping Draws). Is that what you were thinking of?
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Tue Apr 10, 2018 6:12 pm

Chris Goodall wrote:
Tue Apr 10, 2018 4:14 pm
Given the stipulation that there is now only one valid pairing and any human or computer applying the rules must arrive at that one.
It's slightly looser than that as you can have your own set of pairing rules as long as they are declared. Tournaments with alternative rules would likely be valid for rating, but possibly not norms. One practical and valid improvement is to give the bye to the player in the middle of a field, rather than the lowest ranked. The subtle reason for doing this is that if you don't, a player taking part in a series of tournaments can have more than their fair share of byes. It's not just that they are the lowest rated, all they actually need is to be not rated and at the end of the alphabet because of FIDE's alphabetic order rule for players without ratings.

User avatar
IM Jack Rudd
Posts: 3862
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Tournament Structures

Post by IM Jack Rudd » Tue Apr 10, 2018 6:35 pm

Almost any pairing system is valid for FIDE rating, because, as the relevant officials always say, "FIDE rates games, not tournaments".

Paul Dargan
Posts: 512
Joined: Sun May 13, 2007 11:23 pm

Re: Tournament Structures

Post by Paul Dargan » Tue Apr 10, 2018 6:43 pm

and tournaments have a lot of latitude too - remember IoM round1 pairings last year.

That's declared - but impossible to reproduce

User avatar
IM Jack Rudd
Posts: 3862
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Tournament Structures

Post by IM Jack Rudd » Tue Apr 10, 2018 6:50 pm

It has since been banned for norm events for that very reason.

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Tue Apr 10, 2018 8:47 pm

IM Jack Rudd wrote:
Tue Apr 10, 2018 6:35 pm
Almost any pairing system is valid for FIDE rating, because, as the relevant officials always say, "FIDE rates games, not tournaments".
They'll never be able to rate games in a vacuum, because there's a feedback loop between the level of volatility introduced by the k factor and the prevalence of draws. If you come up with a tournament format that produces lots more decisive games than the standard formats, FIDE are going to complain that you're making everyone's rating more volatile than they want it to be. This could be to the benefit of a player who needed to make 2300 for 1 day to get a title.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

User avatar
IM Jack Rudd
Posts: 3862
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Tournament Structures

Post by IM Jack Rudd » Tue Apr 10, 2018 9:10 pm

Chris Goodall wrote:
Tue Apr 10, 2018 8:47 pm
IM Jack Rudd wrote:
Tue Apr 10, 2018 6:35 pm
Almost any pairing system is valid for FIDE rating, because, as the relevant officials always say, "FIDE rates games, not tournaments".
They'll never be able to rate games in a vacuum, because there's a feedback loop between the level of volatility introduced by the k factor and the prevalence of draws. If you come up with a tournament format that produces lots more decisive games than the standard formats, FIDE are going to complain that you're making everyone's rating more volatile than they want it to be. This could be to the benefit of a player who needed to make 2300 for 1 day to get a title.
They rate matches and Swisses on the same system. I think it's fair to say it's probably not a major concern on their radar.

(More generally, FIDE's reaction time to things that disrupt the assumptions of their rating system is long. Look at how long it took for them to introduce higher K factors for juniors.)

User avatar
Michael Farthing
Posts: 1780
Joined: Fri Apr 04, 2014 1:28 pm
Location: Morecambe, Europe

Re: Tournament Structures

Post by Michael Farthing » Tue Apr 10, 2018 9:36 pm

Gentle men,

Though with considerable time and effort I would probably be able to understand this thread, it is currently totally opaque to me.

However, the manner of its conduct is a model of scholarship, decorum and good manners and I heartily congratulate you on this grand achievement and pronounce it Thread of the Year.

Thus, the at first sight erroneous presence of a space in the first line of my post is not a fault but of deliberate intent.

Post Reply