Round-robin with swiss pairings.

[benlessard 0]

I want to start that i know nothing about sumo but I do know how shogi and go league tend to be in Japan. I have watched "Freakonomics"

and understood that tournament or league are mostly round-robin. In my experience all discipline where there is round-robin or league where some are sure to be qualified well before others there is always the possibility of match-fixing or its often advantageous to lose a game (dumping) or at least not give your full 100%.

I think I have a solution for minimize dumping and match fixing.

Assume you have a 16 players round-robin, there is 15 rounds so each player play 15 match vs all the other players and there is 225 matchs in total.

You start with a predetermine ranking or random for a coule of early rounds. After a couple of early round you do swiss pairings (anti-swiss to be more precise) http://en.wikipedia.org/wiki/Swiss-system_tournament

where the players in the top rank or in the bottom rank tend to play vs the middle rank players. In short you tried to avoid game of middle-rank vs middle-rank) yopu do that every round except the 3 last round where you must make sure that you have a suitable schedule for the end of the tournament. If done properly at the end of the tournament players with extreme score 10-4 or 4-10 will play vs themselves and players with close score 7-7 or 8-6 will also play themselves.

[Asashosakari 19,669]

The divisions below juryo do use the Swiss system, and have for decades. The top two divisions aren't nearly large enough to apply it in a useful way.

[Koorifuu 1,020]

  On 02/10/2013 at 18:09, Asashosakari said:

The divisions below juryo do use the Swiss system, and have for decades. The top two divisions aren't nearly large enough to apply it in a useful way.

Nor it would make sense. By day 5 / 6 an undefeated yokozuna / ozeki would be facing an undefeated low ranker. It'd be completely different from what we're used to watch, the top guns brawling each other unless something extraordinary comes from below.

[Asashosakari 19,669]

  On 02/10/2013 at 19:35, Koorifuu said:

Nor it would make sense. By day 5 / 6 an undefeated yokozuna / ozeki would be facing an undefeated low ranker. It'd be completely different from what we're used to watch, the top guns brawling each other unless something extraordinary comes from below.

Or, depending on what "a couple of early rounds" means, that 5-0 yokozuna would meet a 3-2 or 2-3 rikishi - but so would (by rule) a random guy from the middle of the division who just happened to start 5-0, too. That may make sense in tournament fields where the relative strength of the competitors is not well known. But in top division sumo we actually do know ahead of time that some rikishi are strong championship favourites while others are just along for the ride. And so it doesn't make much sense to treat rikishi as equivalent for the scheduling just because they happen to have the same number of wins.

And of course, unlike most events that do Swiss scheduling Ozumo is a spectator sport, so the matches need to be audience-friendly. There's little point in continually matching up each well-performing maegashira with opponents who are 3-3, 3-4, 4-4, 5-4 etc. just because that's what the system says, if the audience would rather watch those maegashira go head-to-head to see who's the real threat to the top-rankers this basho.

But the biggest issue is that it doesn't actually address the match-fixing threat in a meaningful way, because at the core of this suggestion lies the following supposition:

Given suitable win-loss records,

- Rikishi A meeting Rikishi C early, and Rikishi B late, is less prone to match-fixing than

- Rikishi A meeting Rikishi B early, and Rikishi C late

That idea breaks down as soon as Rikishi A can reasonably assume that he will be meeting B and C at some point during the tournament. And that's exactly what was revealed during the big yaocho scandal - rikishi were simply trading wins round-robin style (so e.g. three guys would all finish their internal "rounds" at 1-1), regardless of when they were actually scheduled.

It appears the OP is mainly influenced by the Freakonomics stuff, but as that focused on 7-7 charity and the associated (suspected) repayment of those wins in later tournaments, it's pretty much old hat. Sadly, the reality of match-fixing turned out much worse and much more complex, and systematic scheduling per Swiss variants really wouldn't offer enough of an improvement to make the associated disadvantages tolerable, IMHO.

[Asashosakari 19,669]

In the end, probably the single biggest improvement to the makuuchi match-making would be to simply scrap the one-day-ahead scheduling. But that's also one of those things that is subject to real-life economic considerations, because there's value in being able to tell your Day X audience what they'll be able to expect for Day X+1.

[Akinomaki 40,811]

And the kensho sponsors want to know in advance as well.

[sashimaru 10]

  On 03/10/2013 at 07:24, Akinomaki said:

And the kensho sponsors want to know in advance as well.

and sumogamers too

[benlessard 0]

"a random guy from the middle of the division who just happened to start 5-0" is it likely ? This middle guy would have at least 2 matchs vs the top ranking Rikishi.

"three guys would all finish their internal "rounds" at 1-1". I must admit I didnt think about this one... If they agree that A>B>C and C>A and you have some gambling money on top of it there not much you can do about it. But still if these matchs are at the end of the tourney its more annoying for the cheaters. EX im C and im supposed to lose against B but im 77 and play B who is 86 will I accept to lose ?

can you give me a link of where I can see a cross-tabe of sumo results ?

In general out of 16 players how many get money prize ? How many get promoted ? how many got demotion ?

Is getting at least 8 wins always important ?

[Asashosakari 19,669]

  On 04/10/2013 at 00:06, benlessard said:

"a random guy from the middle of the division who just happened to start 5-0" is it likely ? This middle guy would have at least 2 matchs vs the top ranking Rikishi.

Well, you didn't really clarify what

  On 02/10/2013 at 17:33, benlessard said:

You start with a predetermine ranking or random for a coule of early rounds.

would entail, so I was just guessing what you would have in store for Day 6. That 5-0 is a lot more likely if "a couple of early rounds" means 3, than if it means 5.

  On 04/10/2013 at 00:06, benlessard said:

But still if these matchs are at the end of the tourney its more annoying for the cheaters. EX im C and im supposed to lose against B but im 77 and play B who is 86 will I accept to lose ?

These guys have lengthy careers. There's always a later opportunity to return the favour.

  On 04/10/2013 at 00:06, benlessard said:

can you give me a link of where I can see a cross-tabe of sumo results ?

Top division from two tournaments ago. (The one from the just-concluded tournament is less useful as a top-ranker withdrew halfway through.)

  On 04/10/2013 at 00:06, benlessard said:

In general out of 16 players how many get money prize ?

Everybody in the "professional" ranks gets a monthly base salary. Other monetary rewards are rather limited in scope: The championship winner gets a large prize, between (usually) 1 and 3 others get smaller ones in recognition of their overall performance, and then there are awards on specific bouts, but those are concentrated on bouts involving at least one very high-ranking rikishi.

  On 04/10/2013 at 00:06, benlessard said:

How many get promoted ? how many got demotion ?

I'm not sure I understand the question. Except those at the very top who don't have much room to move, everybody gets either promoted or demoted after every tournament.

  On 04/10/2013 at 00:06, benlessard said:

Is getting at least 8 wins always important ?

That depends on how you define "important". It's always useful, but sometimes it's more useful than at other times.

Edited October 4, 2013 by Asashosakari

[Doitsuyama 1,192]

Not much to add here, but I would like to clearly state that the original assumption of a round-robin tournament in Ozumo is totally wrong. Compare the matrices in above links. Basically the current system already is matching up same level ranking rikishi in the first week and matching up rikishi with similar records in the second week. That's not perfectly possible in small divisions like Juryo and bottom half of makuuchi. The top half of makuuchi (top 16 or so) actually is more or less a round robin, but cheating is not much of an issue there (or at least looked away from, not even Freakonomics is dealing with the OBSC iirc).

The economic reason to cheat is biggest in low juryo because dropping to makushita means dropping from a fat paycheck to almost nothing. The pay differences between low juryo and sanyaku are even smaller then that drop. So it is easily explainable that most of the expelled rikishi in the scandal were juryo rikishi.

The Freakonomics discovery regarding 7-7 rikishi was economically seen pretty much meaningless, this one is more about stroking the egos of the rikishi as a 8-7 positive final record is seen as a success, while a 7-8 is regarded as a failure - but actually there is very small monetary compensation to that view. In fact, it can be very favourable to drop some ranks with a 5-10 or so, and then get a 10-5 in the next basho, this is much better than a 7-8, 8-7 in the long run because of a quirk in the complicated salary system as positive records increase your salary (so +5 is better than +1), but negative records don't decrease the salary. There actually were a few rikishi suspected of riding this quirk, called elevator rikishi because of the big swings up and down - probably no match fixing involved here; could in theory be possible (trading a loss on the down way with a win on the up way), in praxis difficult because these rikishi usually lose to different opponents than they beat.

[krindel 675]

Another fact that I have not seen mentioned is that in sumo, getting to fight the Yokozuna and / or the Ozeki is largely seen as a badge of honor. You need to get to at least M5 / M6 to be able to fight a Yokozuna, or to have a very exceptional performance in one tournament from lower down and pass the test of defeating some sanyaku too before getting to that.

Also of note is that the Swiss system more or less implies that or participants in a tournament enter the competition with the same goal in mind, i.e. to win the tournament. That is DEFINITELY not the case in sumo, a very big number of rikishi just want an 8-7 score and they are more than fine with it. With that in view, the Swiss system can be exploited by losing a few initial bouts to get some easier opponents down the road (just ask all the Komusubi in history if that would help compared to the current system).

As Asashosakari and Doitsuyama clearly stated earlier, I also think that a straight up Swiss system would not help sumo at all, especially when thinking that the potentially championship deciding matches between the top rikishi would end up getting buried on random weekdays, and that is probably the worst aspect of the whole thing.

  On 02/10/2013 at 17:33, benlessard said:

If done properly at the end of the tournament players with extreme score 10-4 or 4-10 will play vs themselves and players with close score 7-7 or 8-6 will also play themselves.

Also, while I understand why that may sound to you like its a good thing, by sumo standards, it really isn't. For the top rikishi who want to advance, a 10-5 score is very different than an 11-4 and a world away from a 12-3. If a Sekiwake-ranked wrestler at 10-4 gets to fight a low ranked rikishi on last day, it would completely screw up the entire promotion system of the high rankers.

Edited October 4, 2013 by krindel

[Asashosakari 19,669]

  On 04/10/2013 at 10:22, krindel said:

Also of note is that the Swiss system more or less implies that or participants in a tournament enter the competition with the same goal in mind, i.e. to win the tournament.

Or that they're at least all acting under the premise that every single win is very valuable to everybody. That's why it works in makushita and below, because there you have the dual goals of winning the tournament and of ultimately getting promoted to juryo, so even if winning the current tournament is no longer possible there's still a long-term goal. (And every win pays some money directly, too, which might matter to those who have maxed out their careers already.)

  On 04/10/2013 at 10:22, krindel said:

For the top rikishi who want to advance, a 10-5 score is very different than an 11-4 and a world away from a 12-3. If a Sekiwake-ranked wrestler at 10-4 gets to fight a low ranked rikishi on last day, it would completely screw up the entire promotion system of the high rankers.

That's in effect the main issue - in the top two divisions there are too many "pockets" that really require context-dependent scheduling, not a rote system. And once you exclude S-M3, M12-J3 and J10-J14, you don't have many rikishi left to work with for any strict system. Not to mention that the boundaries between "pockets" and "system areas" would be difficult to handle in a hybrid approach. That difficulty is already seen when we compare low juryo to high makushita, actually, though at least they have the "they're completely different divisions" excuse there when things go out of whack.

  On 04/10/2013 at 09:51, Doitsuyama said:

The Freakonomics discovery regarding 7-7 rikishi was economically seen pretty much meaningless, this one is more about stroking the egos of the rikishi as a 8-7 positive final record is seen as a success, while a 7-8 is regarded as a failure - but actually there is very small monetary compensation to that view.

And it was based on a sample of results from a very special period of time in which the positive (ranking) value of an 8-7 was in fact much greater than the negative value of a 7-8. That's no longer true now, the banzuke movements are generally much more symmetrical, so there's no way to arbitrage that single win in a meaningful way anymore. (Which shifts the problem to 5-10 / 10-5 style elevatoring as you mentioned, but that's indeed harder to exploit, at least in a cooperative manner as opposed to a single rikishi simply deciding to dump a few bouts without the agreement of his opponents.) Edited October 4, 2013 by Asashosakari

[shumitto 418]

  On 04/10/2013 at 09:51, Doitsuyama said:

The economic reason to cheat is biggest in low juryo because dropping to makushita means dropping from a fat paycheck to almost nothing. The pay differences between low juryo and sanyaku are even smaller then that drop. So it is easily explainable that most of the expelled rikishi in the scandal were juryo rikishi.

But I think they stopped digging after reaching a certain point in the investigation, otherwise some other names might have been dragged into the mess.

  On 04/10/2013 at 09:51, Doitsuyama said:

The Freakonomics discovery regarding 7-7 rikishi was economically seen pretty much meaningless,

But doesn't a kachi-koshi add some money to the hoshokin ?

[Asashosakari 19,669]

  On 04/10/2013 at 14:18, shumitto said:

But doesn't a kachi-koshi add some money to the hoshokin ?

8-7 = +2000 yen per future basho, but I do think that qualifies as pretty meaningless, especially given the "street value" of wins that was revealed in the yaocho scandal. (What was it, 400,000 yen?) Let's not forget that the basis of the 7-7 charity isn't just 7-7 vs. 6-8 type bouts (where only the 7-7 rikishi can improve his hosho-kin), but just as much the 7-7 vs. 8-6 bouts (where the 8-6 guy is giving up +4000 yen himself by rolling over).

[shumitto 418]

+2000 yen is indeed meaningless, even if it is early in one's career and thus multiplied many times. I just mentioned it for I thought it had more of an impact in their earnings.

That is what lies behind yaocho, I guess. Some bouts don't add much career-wise, but if dropped they make your pocket heavier.

I've just pulled the following situation out of thin air, though it might work as an example.

An 8-4 mid-maegashira rikishi who is certain to get a beating in the meatgrinder might "trade" a couple of his remaining bouts with a 4-8 M12 or a 6-6 M14.

A possible solution would be to pay according to each rank or cluster of ranks, making every bout worthwhile, or more likely to be.

Based on the given numbers for the "street value" of bouts, I have no idea how or if such a system could be efficiently implemented though.

[svijetsuma 1]

Funny, I was just thinking about all of this for a last few days... Being a Go player I tried to come up with a system that would describe how Makuuchi pairings were decided... and I have failed. There is no clear way or system that could do that. Also, I live in Europe, so my views are defined by this experience.

In Chess (I don't know much about Shogi), Swiss system is mostly used. Basic problem with Swiss and larger number of participants (players) is the same like in a Makuuchi - what if Endo actually takes first 13 bouts? Or 7? Or 9? Should sanyaku waist a bout in the middle of the championship with a lower maegashira rikishi? So, in chess tournament are usually divided into smaller tournaments, for instance Makuuchi would be separated into 3 tournaments with sanyaku playing their own.

In Go, a derivation of Swiss system is used called McMahon. It is a Swiss system with added starting points, in which points are given according to players rating. For instance, both Yokozuna could start with 22 points, Ozeki 1 with 21, and m13 Endo with 4 points. So, Endo would never meet any sanyaku, which is obviously good and bed at the same time. Basically McMachon does the same thing as division of Makuuchi into separate tournaments, and does so by creating "layers" of rikishi separated by huge point differences (Endo's maximum might be only 19 points). In this manner players with the similar strength play each other most of the time. But, in Sumo number of wins decides the winner, not number of points, so...

McMahon could be used differently. If we give first 15 players the same starting points and say that pairs are made in a way that top rated player is paired with lowest rated, then we could get something very similar to Makuuchi pairings. For instance Y1e would have 22 starting, as well as m3e (m3w would have 21), and on day1 y1e would play m3e, and y1e would play ozeki players day 10 or later. Also, if we set smaller point differences (m3-m5 21 points, m6-m8 20, m9-m11 19, and so on...) theoretically Endo could play sanyaku by the end of the basho.

This setting has 2 major problems: a) strange results, when stronger rikishi loose too many bouts, for instance Harumafuji might have to fight against Yoshikaze on day 6. Also, Hakuho might have to fight against Kotooshu as early as day 7. b) In Sumo number of wins decides yushio, not number of points. Endo might never get to play with Hakuho or any ozeki and have more wins and take yushio in the end...

So, the closest thing that I have come up with is using different rules in different stages of the basho. Day 1-8 starting points decide pairings. Day 9-11 maegashira, konusubi and sekiwake pair on Swiss system (win-loss ratio). Day 12-15 everybody plays Swiss system. This one looks good, but we might end up loosing yokozuna vs ozeki bouts because yokozuna might have to play mid-maegashira rikishi with lots of wins, and yokozuna vs yokozuna bout would almost never happen on the last day. Biggest problem with "changeable" rules is that they should be equally applied to all, and if they are "changeable" this is not possible...

In conclusion, with the present way that pairings are decided, there is no system capable of simulating Makuuchi pairings.

Also, and I presume that this was the initial idea of this post, this magical pairing system should help us avoid "unfair" or "less-corrupt" results and pairings. But any system that is being used has its own structure and flows that come from it. So any system can be abused, as Neo and The White Rabbit should know.

Edited October 4, 2013 by svijetsuma

[Asashosakari 19,669]

  On 04/10/2013 at 15:45, shumitto said:

That is what lies behind yaocho, I guess. Some bouts don't add much career-wise, but if dropped they make your pocket heavier.

...

A possible solution would be to pay according to each rank or cluster of ranks, making every bout worthwhile, or more likely to be.

I'd be in favour of that on the principle of better rank = better pay alone, though I don't harbour much hopes that it would work as a major incentive to not throw bouts. Maybe taking a page out of the lower divisions' system and switching to a more merit-based payment system would work (lower base salaries, lots of money for each win), but I suspect that introducing something like that would be a PR disaster. (Edit: Would probably allow them to reintroduce a kosho system though, as it would mean absent rikishi would lose significant parts of their earnings, but without completely sending them to the poorhouse like a demotion to makushita does.)

One idea I've seen half-seriously suggested on Japanese places would be to make the sekitori rankings much more dynamic, say dropping Tokushoryu (M10e 6-9) and promoting Azumaryu (J4w 8-7) for the next banzuke. I don't think that's workable though, as you really can't do it at the very top without "confusing the marketplace" (3 new ozeki every year?), and it would probably result in rikishi constantly fighting out of proper rank, leading to lots of foregone conclusions on the dohyo.

Edited October 4, 2013 by Asashosakari

[benlessard 0]

Its clear for me now that these sumo tournament are not round robin at all since there are several classes of sumo in the same tournament. To an untrained eyes your tournaments pairings just look like a big mess begging for unfair pairings and dumping waiting to happen.

Anyway here is an example of what a conventionnal RR with swiss pairings would look like. The chart didnt post well you can find it here ..

http://www2.teu.ac.jp/gamelab/SHOGI/oshomain.html

Osho League (shogi) No Name 1 2 3 4 5 6 7 R T.Kubo, 9-dan x 2 M.Toyoshima, 7-dan x C A.Watanabe, Ryu-O x 4 Y.Habu, 2K x R M.Goda, Kio x 6 K.Fukaura, 9-dan x R E.Iijima, 7-dan x

You start with a predetermined ranking or random. I prefer predetermined based on rank/titles or previous tournament

4-- (2K) not sure what the 2K mean but Habu had at least 2-3 titles so im putting him first.

3-- (Ryu-o) title is more prestigious than kio I think)

5-- (Kio)

1-- (9dan)

6-- (9 dan)

2-- (7 dan)

7-- (7dan)

8-- (ghost player ----bye) because of the bye its a 7rounds tournament not 6.

Now you have to define a goal

do you want that the stronger player play together at the end (breaking point is between 1&2 since you want a decisive match in round 6-7)

do you want that players close to a certain point play together at the end. (example only 4 players qualify so the breaking point is between 4-5)

lets choose the first goal so you do an anti-swiss (top score play vs lowest score, middle play vs middle)

ROUND1

4-bye ---1-0( Ive considered a bye an automatic win for pairings purposes but maybe 0.5 is better)

3-7-------1-0

5-2-------0-1

6-1-------1-0 (so the new ranking is 4,3,6,2,5,1,7,bye

Round 2

4vs7-----2-0

3vsbye--2-0

6vs5-----2-0

2vs1-----1-1 (so the new ranking is 4,3,6,1,2,5,7,bye)

Round 3

4vs5-------2-1

3vs2-------3-1

6vsbye----3-0

1vs7-------2-0 (new ranking is 3,6,4,1,5,2,7,bye)

Round 4

6vs4------4-2

3vs5----- 4-1

1vsbye---3-0

2vs7------2-0 (new ranking is 3,6,1,4,2,5,7,bye)

With 3 round left you have to be a bit careful because there is some triangle position that may block (A need to play B+C and BC need to play together needing 3 rounds even if there is 2 matchs/rounds to play)

so a recap...

1=3,4,5 ---------(1 need to play 3+4+5)

2=4,6,bye

3=1,4,6

4=1,2,3

5=1,7,bye

6=2,3,7

7=5,6,bye

and the ranking is 3,6,1,4,2,5,7,bye.

the 3 last rounds are going to be "forced" but interchangeable & obviously you want the 2 leaders (4-4) on last round.

You put 3vs6 on the last round, and 3vs 4 before 3vs 1. With a little work (or a computer) all the rest will fall into place.

5vs1

3vs4

6vs2

7 vs bye

3vs1

6vs7

2vs4

5 vs bye

3vs6

2 vs bye

5vs7

4vs1

So youve managed a swiss where at least you have an interesting match in the end.

3(Watanabe) with a score of 5 playing vs 6(Fukaura) with a score of 4. Not doing the swiss pairings its very likely the tournament winner would be knowed at round 6 and round 7 would just be a boring formality.

Edited October 5, 2013 by benlessard

[Asashosakari 19,669]

I don't want to discourage your enthusiasm, but that's rapidly losing any traces of applicability to top-division sumo...

I'm fairly sure everybody interested in scheduling here does know the basic method of the Swiss system from sumo anyway, even if it's not commonly referred to as such: Lower divisions Round 1, Round 2, Round 3, Round 4, Round 5, Round 6, Round 7, "Round 8" (divisional playoffs where needed).

Those divisions don't hold any commercial appeal and offer little to no reason for intentional losing, so a straight-up systematic approach works there.

[benlessard 0]

Like ive said my assumption was for pure round robin and those do not exist in Sumo. I have assumed that since there is some RR in Go and Shogi there would be some in sumo (especially when I saw 16 Sumo around the circle and heard the tournament last 15 days).

But its like somebody comparing pairings in Olympics vs pairings for professionnal boxing (where its mostly $$ and contracts that decide who is the next challenger to the titles holders)

[yorikiried by fate 2,029]