Friday, February 2, 2007

Guild Wars: What the Heck is a Swiss Tournament?

With the upcoming tournament and the ATS beyond that, Guild Wars players are going to be dealing with a Swiss Tournament format. So, I thought it best to do some looking into that and exactly what it entails.

Simply, to have a Swiss tournament you divide the pool of available teams into various groups (This can be done randomly or through some criteria – ladder rank, for example. . Then you pair them against each other. The results of those matches you use to create new groups in the next round. And so on until you play enough games to decide a winner. Every team can play in every round. And, indeed, that's one of the big reasons to switch to this format.

One of the advantages of a Swiss Tournament is that it encourages more participants – unlike a single elimination tournament, teams are guaranteed a number of matches equal to the base 2 log of the number of participants. Or, in other words for 2^n participants you have n rounds of play (I knew that Math class was going to pay off!). That's the same as the number of rounds in a single elimination tourney but there a team that has a bad day or runs into something that hates them out can be gone after a single round. That they get to play more means more experience and, hopefully, a better environment for all competitors.

One of the disadvantages of a Swiss Tournament is that it encourages more participants – especially the bad ones. Meat, in so many words, for the better teams to crush in between the matches against good opposition. The more teams in the pool, the more diluted it becomes. As Trevor Reznik points out in the aforelinked thread at TGH – that leads to a situation not unlike the ladder where good teams wade through hordes of bad opponents, occasionally get bombed out by something out of left field, and rarely get a match that really tests them. Noobsmashing isn't fun for anyone, after all (I disagree with his assessment - it really shouldn't be as bad as that even with a lot of teams playing. But the central concern behind his typical TGH "sky is falling" attitude is, I think, a valid one.).


However, a Swiss tournament isn't the completely random pairing of the old ELO ladder – while the first round or so can be completely randomized (They can also be based on some kind of ranking – the only viable measure at the moment is the ladder rank and, you know, I'd avoid that like the plague) each subsequent round is based on the results of the previous one. Winners face winners and losers face losers and you play until you've had enough rounds for one winner to be determined.


For example, consider this simple Swiss Tournament of four teams called A, B, C, D. That's four teams and so you only need to play two rounds. The first round will be completely randomly paired (In this case, alphabetically, and when the games are played the higher letter always wins.) but the second round will pit the winners of the first round against each other. So, you get the following matchups:


Seed

Team

Round 1 Opponent

Round 2 Opponent

Final Score

1

A

C

B

2

2

B

D

A

1

3

C

A

D

1

4

D

B

C

0


Notice that A doesn't play D – the worst team – at all. The rule in Swiss Tournaments is that the top half gets matched against the bottom half. So, in the first round the first seed plays the seed that's 1/2 the total participants +1. In this case, the third. But in an 8 team tournament it would be 1vs5. 16 it would be 1vs9. And so on down the line. This is a difference from the typical single elimination tournament where the highest seed faces the lowest – to give that higher seed a better chance at getting deeper into the tournament. That's not a concern in Swiss Tournaments, though.


So, in the first round, Team A dominates Team C. While Tean B has a solid if unspectacular match against Team D. With matching 1-0 records Team A and Team B has a showdown in the second round while Team C and D have a consolation match in the loser's bracket.


More teams means more rounds. If there were 8 teams (From A to H) taking part there'd be a third round and Team A would face Teams E, C, and then B to take the title with three points and a 3-0 record. And Team H would have gone up against D, F, and G in amassing a 0-3 record. While at the same time there'd be three teams with a 2-1 record and three teams with a 1-2 record. The top team gets to see only decent competition, the worse teams get a peak at the middle of the pack, and the middle gets a view of each. The lesser qualified teams are included and get needed experience against better teams but, at the same time, the better teams are insulated from facing the dregs of the matching – no matter how big or small the tournament is.


In Guild Wars, we're being promised the “largest tournament ever” which means looking at past results at least 128 teams being involved. Indeed, the word is that it's going to be essentially open so we might have 50 or 5000 teams taking part. The Swiss Tournament expands as needed, adding another round if necessary (When, for example, there's an uneven number of participants.) As I figure it to keep the math simple 50 teams means 7 rounds. And however the seeding's done the top team's going to face the 26th overall team and so on. Assuming the higher seed wins out, they'll face the 13th, 7th, 4th, and 2nd rated teams, too. And if they can beat those, I don't think you could argue they're not the best team around.


Things get a bit trickier when you have 500 teams because that means your #1 team is going to face your 251st best team to start. And it's going to take them a while before they're facing the better teams. It's here that the concern of noobstomping becomes relevant because, after all, the higher seed winning isn't a sure thing (When that happens you rely on the points. In this case I'm awarding a single point for a win. But it could easily be different. 3 points for a win, 1 point for taking a match to VoD and losing, maybe, and 0 for a loss otherwise. There are a lot of possibilities here and, unlike in a single elimination tournament, you can win the tournament before the final round is played.). While you can recover from flaming out in a match due to someone running something janky that's your specific counter (In M:TG this would be a rogue deck – something that only works because it hates out the current metagame. Against people not running the bleeding edge stuff it gets destroyed. In Guild Wars, however, skillful play can make up such a build advantage but you can still get hated out if you happen to get too predictable.) unlike in a single elimination tournament it puts you in a hole and means that, most likely, you're going to be facing lesser opponents who you should be crushing.


One way Swiss Tournaments have of dealing with this quandary is called “accelerated pairings”. Basically, for the first few rounds of the tournament, the records of the top seeds are artificially inflated (Typically by the simple expedient of adding a point to their score.). The result of that is to have the first quarter of the bracket play the second and the third to play the forth instead of the top half being matched against the bottom. When there are a lot of teams or there are enough teams to make the rounds lopsided, this is a good way of providing the better teams with better opponents earlier while still giving the lower half of the bracket a chance to win their way into the higher realms because after a few rounds, that artificial point is done away with and the pairings are again based on who's actually winning.


So, our original tournament would look like this:


Seed

Team

Round 1 Opponent

Round 2 Opponent

Total Score

1

A

B

C

2

2

B

A

D

1

3

C

D

A

1

4

D

C

B

0


Essentially, here you flip those first few rounds around. Which doesn't do anything to the results. But we're talking about an extreme here. In a tournament with more teams involved the effect is to consistently pair the better teams against the better teams.


In order to illustrate the point, let's take a look at a larger tournament. We'll bump up to 2^3 participants instead of the 2^2 we're been dealing with. Larger tournaments are certainly possible and maybe more illustrative but I'm trying to simplify things and all. Here's how the 8 person tournament would look in a standard Swiss Tournament:


Seed

Team

Round 1 Opponent

Round 2 Opponent

Round 3 Opponent

Total Score

1

A

E

C

B

3

2

B

F

D

A

2

3

C

G

A

E

2

4

D

H

B

F

2

5

E

A

G

C

1

6

F

B

H

D

1

7

G

C

E

H

1

8

H

D

F

G

0


Here's the round by round records:


Seed

Team

Round 1 Record

Round 2 Record

Total Record

1

A

1-0

2-0

3-0

2

B

1-0

2-0

2-1

3

C

1-0

1-1

2-1

4

D

1-0

1-1

2-1

5

E

0-1

1-1

1-2

6

F

0-1

1-1

1-2

7

G

0-1

0-2

1-2

8

H

0-1

0-2

0-3


Now, here's what it might look like (More on this later) under accelerated pairing. The * in the score column represents the fictional point that creates the unequal matchings it's added in the first two rounds but not the third.



Seed

Team

Round 1 Opponent

Round 2 Opponent

Round 3 Opponent

Total Score

1

A

C

B

E

3*

2

B

D

A

F

2*

3

C

A

E

G

2*

4

D

B

F

H

2*

5

E

G

C

A

1

6

F

H

D

B

1

7

G

E

H

C

1

8

H

F

G

D

0


You'll note that Team A, the #1 seed has the same exact opponents (Again, we're assuming that everything works out with the higher seed always winning. This won't always be the case, of course, but I'm trying to illustrate a point here.) - the 2nd, 3rd, and 5th seeds. They just play them in a different order. Interestingly, the arguably “best” matchup of the tournament #1 vs #2 happens in the second round. That's because this is an 8 team tournament which is unlikely to require accelerated pairing in reality – it's really for very large tournaments. If it were a 16 team tournament, that would be a lower ranked team (The 3rd, I believe. Still a good match, though!) and that #1 vs #2 would happen later on.


The same thing happens for every other team – same opponents, different order. And the end results are the same. However, take a look at the round by round records and their point totals (again, with the teams getting an artificial points marked with a *):



Seed

Team

Round 1 Record

Round 1 Score

Round 2 Record

Round 2 Score

Total Record

1

A

1-0

2*

2-0

2

3-0

2

B

1-0

2*

1-1

1

2-1

3

C

0-1

1*

1-1

1

2-1

4

D

0-1

1*

1-1

1

2-1

5

E

1-0

1

1-1

1

1-2

6

F

1-0

1

1-1

1

1-2

7

G

0-1

0

1-1

1

1-2

8

H

0-1

0

0-2

0

0-3


I think this is the way it goes, anyway. There are, apparently, many different ways of getting this done and many different rules used in creating pairings – there's no one way, in other words, that everyone agrees on and the way the Guild Wars developers come up with could be just about anything for all I know at this point. Basically, I'm following the rule of thumb that the bottom half of a score group faces the top half. Meaning at the start of Round 3 when there are six teams at 1-1, Team A picks from the middle of that pack (Or, to put it in a more specific way – to start with there are four quarters. The 1st having Teams A and B, the 4th having G and H. In the first round, normally Quarter 1 plays Quarter 3. In accelerated pairing Quarter 1 plays Quarter 2 to start with. Then while Quarters 2 and 3 sort themselves out Quarters 1 and 4 play theirselves. What I'm not sure about is the next round – I'm guessing here that the best team from Quarter 3 then gets to play the best team from Quarter 1 and so on down the line. But, it might be that just the best team overall – that a team hasn't faced – is taken. In which case the top of Quarter 1 would be playing the bottom of Quarter 2. Um, okay, I can see why this thing has been delayed now...)


There's another rule of thumb that winners face winners – and you can't play the same team twice - in which case they'd actually face Team D. But that kinda punishes Team D for being good. They'd wind up with a 1-2 record and only a point – and those points are generally used for ranking them in the next tournament. I'm not sure I like that (It means that accelerated pairing's good for the upper echelon but it's going to warp the picture in the middle tier, essentially.) but if that's how it goes, here's how it looks:



Seed

Team

Round 1 Opponent

Round 2 Opponent

Round 3 Opponent

Total Score

1

A

C

B

D

3*

2

B

D

A

F

2*

3

C

A

E

G

2*

4

D

B

F

A

1*

5

E

G

C

H

2

6

F

H

D

B

1

7

G

E

H

C

1

8

H

F

G

E

0


Which, again, even though it means #1 has a tougher road to getting the crown. But it doesn't much affect the picture at the top or bottom of the brackets but changes around the picture in the middle. I think with more teams involved and more rounds after the arbitrary points go away that this would be more self-correcting – Team D gets hurt here but if this tournament went five rounds and only the first two featured accelerated pairing then the teams in that range wouldn't be as much.


Either way, unlike under the normal Swiss Tournament rules, there's a lot more teams with equal records after two rounds of play. The best team and the worst team are already separating themselves but the top and bottom of that middle ground is yet to be determined. With more teams in the pool those teams in the middle are going to be facing each other more often in subsequent rounds – theoretically providing them with better competition and limiting the influence of any “upsets” along the way.


If the developers are concerned about the elite teams fearing to participate because they'll have to wade through lesser competition then accelerated pairing is a good way to prevent that. It means the good teams will be fighting better teams earlier and the better teams from what's left in the middle later on. I, for one, think it would be a good way of keeping things competitive while still allowing for more teams to participate.


The only issue I can see in such a system is how you determine the seedings – you need accurate seedings in order to accelerate pairings. The only measure now is the ladder rank which is, frankly, not a very good measure at all. So, the initial tournament might have to be completely random draw in the first round with standard Swiss pairing in subsequent rounds. The scores from that tournament can then be used to create seedings for the next tournament and the next (This also, by the way, creates a situation where if teams win enough they might not need to play every round – they can quit once they reach the point total needed to place high enough in the next tournament. Having a variable number of teams from round to round is a variant of the Swiss system known as Konrad. Some Go tournaments also give byes in the form of bonus points to higher ranked players. And different amounts of points can be added over different amounts of rounds for varying degrees of acceleration to the pairings. Then there's tie breakers and awarding points based on the record of your opponent and all other sorts of fun stuff. There are, again, a lot of ways of doing this.) - and these accumulated points can be used to determine the entrants into a smaller championship tournament.

2 comments:

Clamatius said...

You may want to dock your sub-editor's pay. Rouge? Eqaul?

I know, I know. It's so hard to find the staff these days.

Sausaletus Rex said...

Would you believe all typos I make are carefully constructed and deliberate acts designed to draw more comments?