Appropriate Pairing

#2 - April 17, 2014, 5:52 p.m.

Matchmaking has not changed. We stopped showing ranks because they were being used incorrectly (as if they implied skill rather than time invested), causing a toxic environment where players felt justified (wrongly) while harassing each other.

We are working on changes to matchmaking, but those changes are unrelated to the decision to not show ranks.

That said, there has been a huge influx of new players, and this is likely what you are noticing. They are likely going to need some time to learn the ropes and figure out how to play. May I suggest taking a patient, mentoring approach?

#8 - April 17, 2014, 7 p.m.

EverythingEnds.4261:

… so more experienced players don’t get paired up with them.

Already working on it.

#16 - April 18, 2014, 1:24 a.m.

Steve Whitley.8359:

if matchmaking is/was based on rank, then how is that even relevant? from what i’ve seen before the patch / post patch matchups are not based on rank, but rather just whomever is queued…

Matchmaking has never used rank. We use the Glicko2 MMR algorithm. While we will always be limited to whoever is in the queue at any given time, we do not just throw players together at random.

#17 - April 18, 2014, 1:26 a.m.

Rezz.8019:

So Justin: how exactly will you show skill? Will you use ladder rank or something?

That depends on what the community wants, and ladders do seem to fit.

#18 - April 18, 2014, 1:33 a.m.

bhagwad.4281:

If you matchmake based on ranks, it will be more accurate and effective that what you have now.

I disagree, using ranks instead of MMR may help reduce matchup against new players but it would make matches for everyone else just as bad, if not worse.

We have plans to better ease new players into solo and team arena, which should help with this problem without ruining the game for everyone else.

bhagwad.4281:

Make it happen!

I am, I am!

#33 - April 18, 2014, 1:33 p.m.

Locuz.2651:

You disagree because you have no clue. Do you play this game even? A rank 10 will be worse 99.99% of the cases.

You are entitled to your opinion. We know new players are likely to have lower skill because they are ignorant and don’t know how to play the game, but you’re looking at one part of the problem in isolation. Once you get past the initial ranks, using rank for matchmaking falls apart, i.e. you can have two rank 30s with vastly different skill levels.

We will improve the handling of new players, but using ranks for matchmaking everyone else is just a bad idea.

Locuz.2651:

Yet your system pushes totally new players up to the top after 3-4 lucky wins during their first 10 games.

This happens for a few reasons. 1) Player MMR will fluctuate rapidly as the system tries to figure out where they should be. 2) The number of games required to show on the leaderboards is likely too low.

The first is by design, and should not be changed. The second is something we can investigate changing, but may not be relevant if we move to a ladder system.

#36 - April 18, 2014, 1:46 p.m.

vorpal.1497:

However! Unskilled players are not a problem. It is only when one team gets populated by one or more unskilled players, meanwhile the other team gets all average/good players, which is the problem. So by hiding rank now, they hide this broken situation of a team hamstrung by an unskilled player while paired with average/good players.

In solo arena, the system works by sorting players by their MMR – deviation descending and them placing them onto the team that has the lower running total MMR. This means that if one side has more new players then it is just random chance. It also means the players on other team have a lower than average MMR becaue they were comparable to a new player with the default MMR – a very wide deviation. None of this gets reflected by ranks.

In team arena, players are put onto rosters based on their MMR – deviation, and then the matched against another roster by matching based on average MMR.

In both cases, the algorithm will very gradually increase the range of MMRs it will search over time to give players on the edges of the curve the opportunity to play.

vorpal.1497:

Take home point: ArenaNet is aware of these issues with mismatched games, and have made the conscious decision to not fix it, due to lack of time or simply the architecture of the code doesn’t allow them to. Thus they hide rank so we will stop “proving” via screenshots that there is a problem with mismatched teams.

We are aware of the issue, and we have made the conscious choice to fix them, but those changes are not yet ready. If you’ve followed any of the many matchmaking threads, you should have a good idea where we are going. If not, we will give out more information as we get closer to the release of those changes.

#67 - April 30, 2014, 4:07 p.m.

eleshazar.6902:

So say example A of your algorithm failing:

List at start for a 3v3 (20, 5, 4, 4, 2, 1)

According to your algorithm the team comp would look like this at the end:

Team A: (20, 2 ,1) = 23
Team B: (5, 4, 4) = 13

Seems fair right? There has to be more to your algorithm for Solo Q because this one seem awfully barbaric…

That’s the team shuffling algorithm. Which happens after 10 players are gathered using the matchmaking algorithm.

Obviously matchmaking should try to prevent that scenario from happening at all but if for some reason it does fails so bad as to create your scenario, do you see any other combination that would be more fair?

#69 - April 30, 2014, 4:16 p.m.

Solstice.1097:

yea in this scenario the only remotely fair setup would be a 2v4 >.<

Is this something the community would like to see? I could definitely do that, but I’m worried how it would be perceived by those actually playing the game.

Also, in that scenario, even if we stacked everyone against the 20 the numbers would still be on the 20’s side, even though the game mechanics would pretty much guarantee a loss.

#72 - April 30, 2014, 5:01 p.m.

eleshazar.6902:

Yes there is a fair solution: Throw them back into the queue to have a new grouping chosen. A match like that should never take place, because as you alluded to, it will never be fair.

Currently, the only way it is possible is for one or more players to be waiting in the queue for a long time. The alternative is for them to just never get a match. Acandis’ poll is a good start for that discussion.

eleshazar.6902:

Personally I don’t understand how the MMR is failing as badly as it is. True you guys have hidden ranks (which I agree after a certain point, say may 30, are pointless), but it isn’t hard to tell by actions (such as 3 people on your team running home and staying there for cap) or stomps (someone using a deer finisher), or titles (someone wearing Gladiator) that the skill levels are not being matched up fairly in the least.

Excluding the initial period when MMR is “warming up”, every single instance I’ve looked at has been a fair pairing. The downside of not showing the data is that many people will assume the worst.

The experience gap, which MMR does not look at, is one of the first things we’ll likely be changing in the future.

eleshazar.6902:

Maybe you aren’t putting in enough of a time wait before expanding the MMR range of a player. I recall an Anet person (Josh?) saying that the longer you are in the queue the wider the flexibility of your MMR range. In other words if I start being able to accept others on my team between 10-12 it eventually can expand to 1-21.

It might have been me, I’ve said it more than a few times. We could expand the time it takes, and as I mentioned above, Acandis’ poll is a good start.

eleshazar.6902:

Or maybe it is caused by the last minute grabs. I know that when someone leaves a roster, or leaves a game right before it starts, the system grabs someone fast to try to not have a 4v5, well I have a feeling it grabs pretty quickly and doesn’t wait at all which would lead to high variance in what is actually grabbed.

Unfortunately the system does not do that. While you’re right that it could cause problems, I think could solve more than it creates.

eleshazar.6902:

Something your post just sparked in my mind. Are you guys tallying team MMR using mean or median? I think that could be a huge flaw if you guys use mean. An outlier can pull that number to look falsely high when (as you use in your example) just having a 20 on a team doesn’t make up for having a 2 or a 1. In order to properly rank the team’s MMR I definitely think you should be using the medium MMR and not the average since the outliers can totally mess that up. Not sure which you use…

Both mean and median have their edge cases where they can cause problems. When we make adjustments to the system, we generally try to simulate the changes with realistic data and make choices with the most common outcome in mind.

#73 - April 30, 2014, 5:03 p.m.

Kharr.5746:

Instead, have you considered offering a handicap in terms of points per objective and player kill? … snip …

A handicap isn’t a bad idea either, and would be a good topic for another thread.