Good format for non-large team regional championship

[PeterGustafsson]

Hi!


My club will reach a milestone in a few months - we will organize our first competition other than our small intramural competitions. It will be the South Swedish team Championships, which we were asked (at a regional meeting) to organize by the other clubs in the region. Since we are the youngest and smallest club in the region, this is cause for pride.

This competition is for bragging rights only, it does not count for any qualification - there are no formal qualification competitions to get into a Swedish competition.

The other clubs will provide refs, since we only have 2 refs with epee qualifications, and no ROW refs. All clubs must provide sufficient number of refs to a competition that they take part in or pay a hefty fine, this is regulated in detail by national rules. Pistes will be provided by the regional fencing federation and possibly some of the larger clubs, but of course we will have to put them up.

If previous years are anything to go by, we can expect at most 8 teams in ME, about the same for WE, MF, MS put together. A club may field as many teams as they like in a given event. I can not envision a WS team competition being held, since rules enacted by the Swedish Sport Federation require that a regional championship must have at least 3 entrant teams representing at least 2 clubs. Since that rule is enacted by the Swedish Sport Federation (RF), which the Swedish Fencing Federation (SvFF) is a member of, it supersedes whatever SvFF states/enacts/wishes. It is possible, but unlikely, that the number of competing MS teams is increased by nearby Danish teams being invited. Non-federation clubs may take part in regional championships, but they can not get a placement in the championship final standings. It would be to give local MS competitors some good (read: overwhelming) competition in order to increase their quality. I expect 3-5 pistes being used, all events total.

In previous years, the regional championships have sometimes been done in a Saturday-Ind/Sunday-team format. Other years, the individual and team events have been held at different weekends, as will be the case for us. I think there was a year when they held epee in one weekend, and the ROW weapons in another, although MS has on numerous years not been held at all. We have yet to hold a WS regional championship event.

I have been to numerous team competitions this size, and in every single one of them the piste utilization has been bad - after a few rounds, less than half of the pistes are being used, and this stays so until the BC stops a round and restarts the competition, after which 1-2 pistes lay idle. I strongly believe that this is due to a combination of the format and the fact that the various events have set starting times so that they do not work as load balancers.

The format used is first 1 round of poules, 3-4 teams strong, followed by 100% promotion to DE and all teams eliminated in the same DE round are put into new DE brackets in order to fight out each placement. In the team competitions that I have been to, placement of teams into poules has always been according to the subjective ranking by the BC. While this has worked without major glitches so far, it is risky and can produce bad blood even if the poules are not mismatched. The small poules also increases the effect of quirky team matchups. This is not so much a factor in individual competitions due to the larger poules, but in team events that is not feasible due to time constraints. The combination of DE stages and one large/several small fields also creates a lot of empty pistes, especially when one event is set to start after several DE stages have finished of the previous event.

Despite this format similarity, the Swedish Fencing Federation does not (to my knowledge) mandate a given format for regional championships. It is required that the competition invitation describes the format, and that the format does not involve anything else than fencing. Competition organizers do describe the format, but they routinely add a note that they reserve the right to change the format at the start of the competition based on unexpected entrant numbers, or during the competition should the need arise. This boilerplate phrase has never caused any controversy, AFAIK.

I dislike the bad piste utilization. Being the only club member who has voiced an interest in the theory of competition formats, I intend to do something about it by getting my club to mandate a new format for the MET event. The other events will be run as complete poules (as they always have been) but their starting times will be concurrent with MET. Until MET is down to L4, bouts in the other events will only be fenced to fill out the pistes.

This is the format that I am thinking of forwarding for MET event:
1. Teams are given ranking by lottery, and then a complete shell-sort algorithm is run to give a good ranking. This sorting algorithm can be shown to provide a quick sort where all elements/teams are correctly placed, under the assumption that inherently better teams always win over weaker teams. That is not always the case, I will come to that later. (This is how the Shell sort algorithm works, for those who did not already know or were not paying attention during undergraduate computer studies: http://en.wikipedia.org/wiki/Shell_sort)

2. The Shell sort produces a ranked list which is close to the true list of team fencing strength. That is, the list that one would get if all teams would bout all others so many times so that quirks are smothered by the average. However, since it is designed to reach this ranked list in as few bouts as possible, a single strange bout result (incompetent/partial ref, strange matchup, strange glitch in the box that is only discovered after the bout ends, etc.) can cause a single team to get ranked relatively far from its true place. The 2nd stage is designed to alleviate that. In this stage, team captains get to challenge other teams, but only if the challenged team is ranked over the challenger, they have not had a match before, and that the challenged team has not challenged already by another team in that challenge round. If the challenger wins the teams exchange places in the ranking list, otherwise the keep their places. A team can issue challenges until they amass a total number of losses equal to the critical number, CN which is dependent on the number of competition entrants, n. This is calculated according to:

CN = INT[2Log(n-1)]+2

This critical number ensures that all teams get to issue at least one challenge, even if they have lost all bouts in the Shell sort stage. Once a team has no more challenges and all teams below it in the ranking list also have run out of their challenge opportunities, all those teams can not change their placement and they are out of the competition.

The challenge stage continues until at least one criterion is met:
a. The number of remaining teams is not more than 4
b. No more challenge matchups are possible
c. The time does not allow for another challenge round before the designated stop time.
If the challenge stage is stopped for reasons b-c, all teams with placements below #4 retain their placements and are out of the competition.

3. As soon as the challenge stage has ended, the L4 stage starts. This consists of 4 bouts - 2 semifinals between rank#1-#4 and rank#2-#3, and the two medal bouts. The results in the previous bouts between the teams in those pairings do not count for anything, as soon as the L4 stage starts there is a clean slate.

So much for the proposed MET format. As should be instantly obvious, the right combination between number of pistes, number of teams, and the combination of prior bouts can lead to situations where it is not possible to fill all pistes in a round with MET bouts. In that case, I propose that the unused pistes should be filled with bouts from the other events. If possible, one should run a bout involving a MF or MS team which has at least one member who is also a member of a ME team not currently fighting, this in order to alleviate conflicts while still letting people fence multiple weapons. I assume that a relatively small proportion of the MF/MS fencers will not be part of a ME team. If this can not be done, the piste should be filled with a WET bout. Once one reaches the L4 stage of the MET event, all but 2 pistes should be used by the other events. Considering their small size and that they already have started, they will finish quickly thereafter.

Oops - too long post. The rest follows below.

PG

[PeterGustafsson]

Continued:


Pros of this format:
1. This format will ensure high piste utilization.
2. Since the BC has no part in deciding which teams will fence against each other, no one can accuse anyone else of impropriety.
3. Since there are no poules, unbalanced poule strength is not an issue.
4. High proportion of competitive bouts. In the Shell sort algorithm, good teams quickly rise to the top, weak teams sink, and bouts quickly become competitive - after early blowouts, teams go to the part of the ranking list where those close to them in fencing strength are. The proportion of all bouts that are competitive can not be this high in a poule-based format, since if a poule consisted of only fencers/teams close in strength, then the poules would be grossly imbalanced. In the challenge stage, it goes against the best interest of the challenging team to challenge a team far too good for it, so one can expect a lot of competitive bouts there.
5. Resistance to effects of really weak teams. A total newbie team can give its poule opponents indicators which overrank them. If a middling team has the luck to get a newbie team in its poule and win 45-0, it can get better indicators and therefore be ranked better than another middling team which is intrinsically better, but got a tougher poule. Since the proposed format does not take indicators into account, it is not subject to this kind of screw-up.
6. The ending time of an event is known beforehand. If one has not reached either few surviving competitors or complete connectivity beforehand, the time stop criterion kicks in and only the top four teams get to fence after that stop time.

So, what do you think of it? If one assumes that I can get the format adopted for this competition, how do you think that it will go? Can you think of any combination of piste numbers, event sizes, and other parameters which will cause the format to run slowly? Here, assume that the BC, which will be trained by me, will understand the format. Can you think of any other possible glitch, even if all involved understand the format?

Actually, I have on purpose left a big glaring non-optimal thing in the format outlined above. There is a simple solution to it, which I plan to have in the final implementation. The first person to find the glitch and post a fix gets rep from me!


Have a nice time!

Peter Gustafsson

[oiuyt]

You expect at most 8 ME teams. Run it as a complete pool. With 8 teams that's 4 strips. Use the others for the other weapons. 7 rounds, with each round ~1 hour isn't unreasonable. Only issue is the potential for a barrage for first.

-B

[oiuyt]

A team can only be challenged once?

"but only if the challenged team is ranked over the challenger, they have not had a match before, and that the challenged team has not challenged already by another team in that challenge round."

So if a low team challenges a high team and loses it becomes impossible for an under-ranked mid-level team to challenge that higher team? Not to mention the various race conditions (Team A and Team B each want to challenge Team C. Who gets to issue the challenge?).

Second issue:

Team A works their way up the list, challenging and beating every team above them to the top. Team B challenges from the bottom and has a fluke upset. Team A is now locked at the bottom, unable to rework their way through the pack again, as they've already beaten all of the teams (except for B).

If there's no limit on the number of teams that can challenge a team then B could be bumped on back down the list as all of the teams A beat on the way up take B back down. The final list would have A at the bottom, with B one slot above. All of the other teams that A beat would be above that.

-B

[PeterGustafsson]

Hi!

Originally Posted by oiuyt

You expect at most 8 ME teams. Run it as a complete pool. With 8 teams that's 4 strips. Use the others for the other weapons. 7 rounds, with each round ~1 hour isn't unreasonable. Only issue is the potential for a barrage for first.

-B

I am not claiming that my proposed format gives a critical run time advantage over poule/DE format at this small field size. However, as field sizes increase, I am quite confident that my format runs somewhat faster (and that the speed difference increases with field size), gives a higher percentage of competitive bouts, and has a considerably better piste utilization.

I can not influence the format used at competitions. However, I can influence the format here, and I intend to used that opportunity to test - and prove - my format. I wish to break the poule/DE think mold, and how else would I do that better than a practical example?


Have a nice time!

Peter Gustafsson

[PeterGustafsson]

Hi!

Originally Posted by oiuyt

A team can only be challenged once?

Only once in each challenge round, and it is bound to take on the challenging team. However, it can be challenged by other teams in subsequent rounds.

Originally Posted by oiuyt

"but only if the challenged team is ranked over the challenger, they have not had a match before, and that the challenged team has not challenged already by another team in that challenge round."

I guess that I should provide an example. Assume that one has a field size of 8. We call the teams A-H, with intrinsic fencing strengths in alphabetical order. Assume that the Shell sort stage has passed so that we have a roughly ordered list, but a few upsets makes it only roughly ordered. The ranking list is like this:

B
A
C
D
G
E
F
H

Team H has 3 losses. The critical number of this field size is 4, so after one challenge loss team H will be out. They can choose whichever team they want, except those that they previously fenced against. Assuming that they chose E, team F gets to choose next. They get to choose among the teams that are not E, H, or any team that they previously have fenced against. And so on - only one team challenges at a time, that is the team which is presently lowest ranked among those not set up for a bout in the present challenge round.

So, in the 1st challenge round, we might get this team pairings:
H-E
F-G
D-A
B and C have previously met, so only 3 bouts in this round. If one only has 2 pistes the D-A bout will not be fought in this round, since challenges can only be issued as long as there are pistes to be filled. If there are 3 pistes, all challenge bouts will be fought. If one has more than 3 pistes, bouts from other events are used as load balancers.

Assuming that the results are H loses, F wins, and A wins we get the following ranking:

B
A
C
D
F
E
G
(H)

H has gotten its 4th loss, and since no team below them can challenge them for their position, they are out of the competition with a 8th as their final placement.

G gets to start challenging (not F or any other team that they have met so far) in the 2nd challenge round. This keeps on going until there are only 4 surviving teams, or that no more challenges can be issued, or that the time runs out.

Originally Posted by oiuyt

So if a low team challenges a high team and loses it becomes impossible for an under-ranked mid-level team to challenge that higher team?

In that challenge round, yes. In subsequent challenge rounds - no, not unless that mid-level team has previously fenced against the high team.

Originally Posted by oiuyt

Not to mention the various race conditions (Team A and Team B each want to challenge Team C. Who gets to issue the challenge?).

Whichever of teams A and B that is lowest on the ranking list when that specific challenge round starts. If team a chooses team C, then C becomes unavailable for all other teams for that challenge round.

Originally Posted by oiuyt

Second issue:

Team A works their way up the list, challenging and beating every team above them to the top. Team B challenges from the bottom and has a fluke upset. Team A is now locked at the bottom, unable to rework their way through the pack again, as they've already beaten all of the teams (except for B).

If there's no limit on the number of teams that can challenge a team then B could be bumped on back down the list as all of the teams A beat on the way up take B back down. The final list would have A at the bottom, with B one slot above. All of the other teams that A beat would be above that.

-B

Actually, that is a designed feature of the format, not a glitch. It ensures that teams are kept on their toes - even if one has fenced well all day, a single bad result can spell disaster. (That is what makes straight DE so nerve-racking, BTW.) It ensures that all teams that have not been eliminated so far still have a chance of winning the gold. With this format, there will be no bout which is not important to either team, and the results of a bout can not directly affect the ranking of teams not taking part of it. Those are desiderata that I have outlined before, BTW.

If you can think of a format which fulfills those desiderata (and the ones that I listed in posts #1-2) in another way, I am all ears.


Have a nice time!

Peter Gustafsson

[Lemonaide]

New Ideas are often met with sceptical puzzlement. I used to try to say the same basic thing "utilization of the piste is bad". They're underutilized and we frequently saw people loitering around waiting to fence until a sharper fencer or ref would tap his foot on an empty strip and get the fencers started again. But it sort of brings the momentum down and is the chief complaint of fencers who want to compete within a reasonable length of time, while others also note that there is little excitment for people who want to watch fencing. Timing being everything, a simple thing like this may be the answer to fencing glamor, rather than all the bells and whistles we're currently treated to by our vendors [whom we love anyway].

Why not use your idea and see how it goes!

We've seen light changes, mask changes, electric saber where there was none, no crossovers and so forth.

[oiuyt]

Okay, I was missing what you meant by challenge round. In the original description you'd referenced a "challenge stage" but had no explanation of what a challenge round is. I now understand.

Is a team required to place a challenge (if possible) in every challenge round?

Taking your example, as given, let's say H managed to pick up their three losses against teams B, C, and D. For whatever reason they know they have a match-up advantage against A. Can they choose to wait for A to gain the top spot before challenging?

Could you give the step-by-step algorithm for the initial sorting? What gaps are you proposing to use? Seems like it could take an awfully long time.

I'm very unclear how the insertion sorting is done, especially in the 1-sort round, without horrible strip utilization.

-B

[PeterGustafsson]

Hi!

Originally Posted by oiuyt

Okay, I was missing what you meant by challenge round. In the original description you'd referenced a "challenge stage" but had no explanation of what a challenge round is. I now understand.

Is a team required to place a challenge (if possible) in every challenge round?

Yes. With that requirement, teams that have done well in the Shell sort stage get rewarded by getting a few byes in the beginning of the challenge stage, if the number of entrants is larger than 2*#pistes. Otherwise, a weaker team could force a better team to enter early into the challenge stage, without the weaker team risking racking up its loss count.

Originally Posted by oiuyt

Taking your example, as given, let's say H managed to pick up their three losses against teams B, C, and D. For whatever reason they know they have a match-up advantage against A. Can they choose to wait for A to gain the top spot before challenging?

Yes, if team A is not selected by another team by then. If H wins, they will then take the high ranking of team A. In the next challenge round, they will most likely be knocked down considerably by one of teams E, F, or G.

Originally Posted by oiuyt

Could you give the step-by-step algorithm for the initial sorting? What gaps are you proposing to use? Seems like it could take an awfully long time.

I'm very unclear how the insertion sorting is done, especially in the 1-sort round, without horrible strip utilization.

-B

I propose the simplest possible gap sequence that works - 64, 32, 16, 8, 4, 2, 1. Start at the largest possible place in the gap sequence. Shell sort is different from insertion sort, even if both are comparison sorts. I can not envision any existing sorting algorithm that could work for sports competitions that is not some type of comparison sort.


Have a nice time!

Peter Gustafsson

[mrbiggs]

It seems to me that for all the time you gain with strip utilization, you lose more because teams must be in constant communication with the BC to determine who they want to challenge, who they are allowed to challenge, and probably to ask a bunch of questions because I'm willing to bet that there will be some confusion about the details of the format, especially pertaining to strategy.

[oiuyt]

Originally Posted by PeterGustafsson

I propose the simplest possible gap sequence that works - 64, 32, 16, 8, 4, 2, 1. Start at the largest possible place in the gap sequence. Shell sort is different from insertion sort, even if both are comparison sorts. I can not envision any existing sorting algorithm that could work for sports competitions that is not some type of comparison sort.

Okay, so with 8 teams we start with 4-gap.

I'll arbitrarily order the list D, G, A, B, E, F, C, H. We get
D G A B
E F C H
We run 1 round of matches (assuming 4 available strips) to get
D G A B
E F C H
Or a list of DGABEFCH

Next is 2-gap
D G
A B
E F
C H

The first group requires D vs. A (flip), D vs. E (no flip), C vs. E (flip), C vs. D (flip), C vs. A (no flip). That requires 5 sequential matches (and plenty of strips available to the other weapons) to get
A B
C F
D G
E H
Or a list of ABCFDGEH

Next is 1-gap.

This requires A vs. B (no flip), B vs. C (no flip), C vs. F (no flip), F vs. D (flip), C vs. D (omit because already run?), G vs. F (omit as already run?), E vs. G (flip), E vs. F (flip), E vs. D (omit as already run?), and H vs. G.

7-10 sequential matches.

Unless you have a modification to the algorithm that allows for the comparisons to take place non-sequentially. Without such modification this sequence takes 13 rounds of matches just to get to the sorted list before the challenge stage begins. Most of those rounds taking place on a single piste.

The number of losses picked up by a team during the sorting will be variable based on the random ordering at the beginning. A weak team that is early in the initial list and gets unlucky with the ditribution of stronger teams can get hammered and might enter the challenge phase with many more losses than they would if they had been late in the initial seeding list. In the example above F picks up 5 losses during the sort.

I assume that I'm just missing something in the algorithm and have a misinterpretation. If you could run a tournament stage-by-stage, as above, it might help me see where the error lies.

-B