Guild Wars 2

Lopez.7369:

It’s interesting to see people say the leaderboard must be a grind because it’s based on points. Isn’t MMR based on points?

No. MMR becomes more stable and accurate with more games, but there can be two players with equal number of games, even an equal number of wins and losses, that have very different MMR.

MarkPhilips.5169:

I’m ready to bet that in the final top100 there will be people with a really bad win ratio and a huge amount of matches played.

I don’t think this will happen, but if it does happen we can change the table so losing always takes away points. It’s only the narrow losses, or going up against a higher MMR and doing OK, that will let you stay neutral after a loss.

Booms.2594:

I am willing to bet you 1000g that I am better than 24/25 of the people on the front page, and another 1000g there are 25 better than me that arent there

I believe you. Seeing as we haven’t officially started the season and 0 people before today were taking it seriously, I think its fair to assume the end of season ladder will not look like this one.

Archaon.9524:

Well no…that might be true for someone sitting around 500 or if you just soloq…if you have a good team and you aim at high ladder 50% ratio is actually pretty bad (That makes you an average team but for sure not a top 100 one)

Not if we’re matching you against other full rosters with similar skill, which is our goal.

Rank 1: 67 wins, 72 losses, total games = 139

higher than

Rank 8: 38 wins, 8 losses, total games = 46

sounds good and fair to you guys…i really dunno what to say anymore

Rank 1 could be solo and have really accurate matchmaking, giving them better quality matches. Rank 8 could be always going with a full roster and face-rolling due to an inaccurate MMR that is still in the process of stablizing.

If you think win/loss ratio is the metric we should be watching, then I know what to say. You’re wrong.

Win/loss ratio is only a valid metric for skill when the number of participants are low, and each participant players ever other participant a fairly equal number of times. This is not the case with nearly every online game, and is not the case with GW2.

Silentshoes.1805:

Is there any significance to the fact that 99% of the time my team is called “Team Silentshoes”?

It used to be random before the patch. Is it ALWAYS your name now?

I believe it takes the player with the highest MMR. Let me ping Evan or Branden. They’re both away from their desks…. so, let’s just go with Justin = correct.

Archaon.9524:

65 wins 75 loses 1st place
i don’t even want to know about others then….rofl

Rank 1: 67 wins, 72 losses, total games = 139
Rank 6: 50 wins, 20 losses, total games = 70
Rank 8: 38 wins, 8 losses, total games = 46

Data is holding to the hypothesis that “more skill equals less grind.”

Also, win/loss ratio shouldn’t mean anything to you. With an ideal perfect matchmaking everyone’d would be at 50%.

crunchyraisin.6054:

I would bet the player in 1st place at that time had a low MMR and queued solo more often than not.

Since there win/loss ratio is under .5, I’d assume their MMR is pretty accurate. I’d think the people with closer to .9 win/loss ratios are ones who might be benefiting from an artificially low MMR and/or are grouping up and facerolling.

Of course, that’s just guessing. I still have to look at the actual data.

Look at the top 2 in NA, both with 58 points as of writing this..

First place:
58 points with 65 wins, 75 loses. That’s ~0.43 points per game.

Second place:
58 points with 63 wins, 51 loses. That’s ~.51 points per game.

That .43 vs .51 is the difference between narrower losses, and winning against tougher opponents. That’s where the skill comes in.

Yes, someone with .5 points per game will be higher than other players with .5 points per game the more they play, but someone with a higher point per game stat will need fewer games to get the same number of points.

Addendum:
Yes number of games also matters. We want the players that play more to be higher on the ladder than similarly skilled players who play less. We also want less skilled players to have a chance as well.

Acandis.3250:

so from what i gather, new ladder is simply about grinding points. so…… most games played = rank 1

More games mean more opportunity for points, but look at the data. There are plenty of players with fewer wins that are higher. There are also players with fewer loses that aren’t as high.

There are people with really high win/loss rations who aren’t even doing as well with 50/50 ratio.

It’s who you play, and how close you come to winning if you lose.

Silentshoes.1805:

Justin ODell.9517:

Silentshoes.1805:

Meaning 4×5 taken into account?

No, but things like 5 vs 1+1+1+1+1

That’s good to know. 1+1+1+1+1 should get a deep adjustment based on experiences in this test period.

Can you adjust the weighting and then run through ALL the games since the patch in a simulation, until the Expected Winner starts hitting at 100%?

Yes, though it will never hit 100%. Sometimes matches are just too close to tell.

tichorum.2415:

So all wins/losses and rank on the new LB will be reset on the 16th?

Yes. We might also reset it before then to test the reset procedures, but we don’t have a firm date for that because we weren’t expecting it to be visible.

Silentshoes.1805:

Meaning 4×5 taken into account?

No, but things like 5 vs 1+1+1+1+1

Silentshoes.1805:

It’s called the Expected Winner calculation. It is determined by the average MMR of each team.

Not just MMR, we also include the maximum roster size on each side.

http://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm#Match_Prediction

Blackmoa.3186:

http://i.imgur.com/Jt0uVNm.png

So how exactly can this happen?
So i was on 8 wins, played 2 games (won both) and now i am on 2 wins?
I think something is not quite working correctly

This could have happened because while testing the ladder we’ve reset the data several times. It wasn’t suppose to be visible to you guys until the 16th.

On the 16th, we will reset the data one more time and then the beta season officially begins.

No, there should be a bullet under that graph saying that winning always awards at least 1 point.

The current leaderboards are no longer being updated, and the new beta ladder will start on the 16th.

Edit: Apparently someone decided to show the new ladder early, so you’re seeing your data. This will still be reset on the official start, the 16th.

Also, if you’re seeing a down arrow, its probably because someone jumped ahead of you.

longbawl.5072:

And there seems to be see-sawing in philosophy of where to take the game. Back to merging queues? Do we remember why the didn’t work in the first place?

When we had them together everyone hated being stacked (which is understandable) and wanted a solo queue. When we had solo queue everyone hated how it was random and the ladder meant practically nothing.

Now we have them together and everyone hates being stacked against (again, understandable.)

The solution isn’t to try the same thing no one liked, the solution is take care of the actual problem: being stacked against. I’m working on that. Looking at the data, the current implementation seems to be helping, but still has a quite ways to go. I have a few ideas I’m going to run by some people on Monday, and I’ll run a lot more simulations before bringing them into the server.

Another part of the problem is the perception players have, which is being fueled by a selection and confirmation biases, and was not helped by unranked and profession ratings starting at the default.

Lewis Burnell.2493:

How are we now supposed to find out our PvP Rank? I queued today for 10 minutes and got no ranked queue pops. I’m Rank 52 and previously rated 320 on the Leaderboards.

Ladder rank? The beta ladder does not start until the 16th.

What time were you queuing?

glock.6590:

Just bring back the old match making system. Is it that hard? The new system is obviously not gonna work if there’s like only 1k players that actively play spvp.

At it’s worst, the new matchmaking works the same as the old. Which is to say, all it does is find players near the same MMR. The new matchmaking attempts to improve matches by holding on to players until there are more to compare, and then adds addition metrics to the mix.

What specifically is bothering you? Is it queue time? Is it merging solo and team arena?

Have you been watching the forums long? Plenty of people called the old matchmaking broken as well. My goal is to figure out which complaints have merit and can be improved.

Sins.4782:

Can you guys write a script to change matchmaking variables at set intervals based on number of players queuing or something?

It isn’t just a matter of how many players are queuing, it’s how many are queuing within your skill range.

Essence Snow.3194:

Ahh …I’m on NA…..45mins now…and afraid it might pop any second and don’t want to restart whole process again

This is exactly what I was afraid of when I put min potentials up higher. Sorry :\

It needs to be made more flexible, or it will just keep outliers out of the game during off-peak hours.

Also interesting… Legancy of the Foefire, Forest of Niflhel, and Battle of Kyhlo have the lowest SD between teams.

Just looking over the data for the last day. With the exception of Courtyard, which seems to get a lot of blowouts, increasing min potentials seems to have a caused decrease in end-game scores between teams. I think that’s a pretty safe metric to gauge the competitiveness of a match, does anyone have any thoughts on that?

I upped it by another 25, but I don’t feel it’s safe to go any further until we get something in to automatically drop it back down during off-peak hours.

Silentshoes.1805:

It starts pretending that some people with a low skill actually have a higher skill than they do. It is called the “padding” variable. It pretends that a weak team is actually not weak, and then matches it with a waiting team of veterans.

Close, but it’s more likely to happen the other way around. We match players by the lower bound of their rating (i.e. rating – deviation), so it’s more often that it starts pretending veterans are weak.

This only kicks in if someone has been waiting in the queue for 4 minutes.

Xista.7391:

0 improvements so far for me in unranked queue.

I only updated ranked. Unranked now has the same settings.

Lordrosicky.5813:

Lopez.7369:

The point is you would probably be waiting forever on some nights since the game’s population is so tiny. You might be okay with that, but a lot of people aren’t.

Yeh you would literally wait from about midnight till 2 pm. 14 hours queue time anyone? :P

EU has a much strong drop off in off-peak hours. It’s like they all go to bed at the same time or something.

Justin ODell.9517:

That would help, but I’m wary of the impact it would have on queue times.

OK, that change is in ranked arena, but if queue times start getting out of hand I’ll drop it back down.

Lordrosicky.5813:

I am looking at the matchmaking wiki. Perhaps it would be worth making the rating padding only kick in at 3 minutes or something instead of 30 seconds? And maybe it is worth increasing the minimum number of rosters that must be looked at from 20 to something like 50? Anyone else think this might help?

The padding could be the issue as its the high rated players experiences most of the problems.

That would help, but I’m wary of the impact it would have on queue times.

Booms.2594:

darek.1836

Interesting. The prediction that you would win each of those matches was 92-99%. I’ll keep looking to see if I can find out why you’re getting paired so far downward.

Booms.2594:

500-0s 3 times in a row just now. Not even close, we could 500-0 them in a 4v5… Please make a DRASTIC change to match making soon because this is horrible.

Are you playing on another account? I’m only seeing hot-joins on your account.

Lexiceta.4156:

Because not accepting is also dishonor it is a catch 22, either way you lose.

Dishonor is disabled right now. Also, the dishonor for not accepting is much lower than the dishonor for leaving a match.

This should be fixed, but there may be a few more people affected until the data is cleared.

Please let me know if you are still receiving the problem over the weekend.

We’ve created and deployed a new configuration option ‘Rating/@profession-ratio’. This new option controls the balance between your core rating and profession rating.

Since we were not previously tracking rating for professions, everyone’s profession rating started at the default level. This turned out to have more of an impact on matchmaking than we anticipated, hence this change.

Until everyone’s profession rating stabilizes, we’re going to set the ratio to ‘0’, meaning profession ratio isn’t being included at all.

We’ve also seeded unranked arena rating data with same data we used to seed ranked arena, so you should see improvements there as well.

Please let us know if you notice any improvement in your match quality.

Thanks

Oops. My bad.

hangs head in shame

Laserbolt.6731:

If the MMRs are not truly a measure of a player’s actual skill, then you would see blowouts, and the expected win prediction will usually be wrong.

Assuming the match was fair, or the blowout was not already predicted.

Laserbolt.6731:

I take it you guys are primarily turning the dials trying to FIRST make the Expected Win prediction truly accurate? Am I right?

I have dual priorities, that do not conflict. 1) Improve match quality, 2) Ensure prediction is accurate.

Laserbolt.6731:

Isn’t an accurate prediction the best proof that the MMRs are accurate?

It depends. We factor more into the prediction than just ratings. If the predictions do not match the trends, then either the prediction algorithm is wrong, or we’re are missing some essential metric that is required to accurately predict, or the, yes, MMR could be not working the way it we think it should.

Sunshine.5014:

Please correct me if I’m wrong, but I don’t think this will reduce the number of duplicate professions. If I understand it correctly, it reduces the MMR of the team with duplicate professions by 500, and increase the MMR for the unique profession by 500. That means a team with duplicate professions will have the benefit of fighting a lower skill team (since they got the reduced MMR). That’s even better for them. For example, a team of 5 engis stacking now will totally farm the newbies 500 MMR lower.

No, the penalty doesn’t reduce MMR. It is in reference to scoring a roster versus all other possible rosters that could be included in a match.

MMR is still in the forefront because we do not even consider a roster as a potential match unless it’s within the MMR range, and it also has a scoring penalty for being further away from the target MMR.

Saiyan.1704:

This may or may not be in the ladder matches, along with skyhammer.

It will not be included during season play.

lordhelmos.7623:

But did that unranked play have to come at the price of subjugating all solo players to months if not an entire year of being curbstomped so you can build metrics?

I’m guessing you’re expressing skepticism here, but its only been 2 days… give me a bit more time. I’ll consider it a personal failing if I can’t resolve the issue after a year.

lordhelmos.7623:

Unranked matches are going to turn into a reward farm for organized teams to climb the R80 ladder. Even fewer still are actually going to play at the level envisioned for ranked team matches.

Unranked will not have a ladder, so maybe you meant ranked? The goal is to make it far more likely that high skill level teams are paired against each other.

lordhelmos.7623:

If you want to drive people towards organized play only, with no step ladder to get there

We choose unranked because it acts as a stepping stone. We expect the more organized players to gravitate toward ranked arena for the additional ladder rewards.

lordhelmos.7623:

Some people, like myself, just want accessible pvp with decent rewards. SoloQ made that happen.

I hope unranked will fill this place for you. I understand what you’re saying, and we are paying attention.

Jak Shadow.2864:

I can understand that there were improvements that could be made to the matching algorithm but did this require also the merging of SoloQ and TeamQ. Not sure how this was supposed to help but it surely hasn’t.

I am sure all the work on the algorithm is worth it just restore SoloQ as an option. Most major issues improved instantly.

The shortest version of that story is that we believe there is more value in having an unranked place for people to play. Part of that choice includes the fact we can now modify the servers much more easily and make changes faster.

That does mean we’re ignoring what people are saying about solo vs premade.

Laserbolt.6731:

Are you guys validating the “expected win algorithm” against actual games, then?

If there is not a very strong correlation, i.e. “80% of the predictions come true”, then there is a real problem.

Right?

Yes. I think its key for ladders to have value. With this release we’re collecting way more metrics than before, and many of us are hyped for a chance to get at that data.

Booms.2594:

wow it sure is promising when this is what our pvp devs know..

Hah You’d be amazed at number of things I don’t know. :P

Deimos Tel Arin.7391:

could they have been disconnected by the server?
or they were just maliciously disconnecting the games, causing 5v4 to occur, since no dishonor in place yet?

I’m not sure, I’ll need to look at the data.

Pandatrice.7183:

well I didn’t get any information from the pictures…yeah I’m very new…is it two players standing on one point? could someone point out for me? thx

All I got from the pictures is that apparently Backpack stands outside the enemy spawn and taunts players looking for a challenge… or something.

That and his team was pretty spread out.

Booms.2594:

is this seriously the reason? So if we want to get legit mmr we should only play 1 class for forever ?

Yes, it’s a temporary situation that is more noticeable to those players who are further away from the default MMR.

No, it’s by design that it lowers your MMR when playing a class you play less frequently. Sure, playing only one class will probably mean your core MMR and profession MMR are in sync, but I wouldn’t say this is preferable as the profession MMRs should catch up quickly.

Some of these are really hard to make tangible in code.

Stand The Wall.6987:

average knowledge of pvp strategies, rotations, and roles

We can sort of make assumptions about this by looking at rank and rating, but it’s not a direct correlation.

Stand The Wall.6987:

the power of 3rd party communications

This isn’t something we can detect. Even if we could, some teams just communicate better. We definitely want to do things that lessen the gap in this regard, so feel free to start a new thread if you have ideas here.

Stand The Wall.6987:

average player skill

MMR is suppose to correlate with this by gauging your performance over time versus other players.

Stand The Wall.6987:

what a 2 player premade can do

We try to pair like-sized rosters together to compensate for things like this.

Stand The Wall.6987:

a timeline (impossible as of now) of larger premades vs smaller ones

We’re recording the result of all games played under the new system. So, hopefully this data will improve our understanding and give us more ways to test the assumptions we are making.

Stand The Wall.6987:

team composition in a pick up group environment

Team composition (as in roles, etc.) would be nice, but people already don’t like not being able to change character… I think they’d like being locked into a build even less.

Stand The Wall.6987:

what all of these things can do for a team when they are in sync

This is a big one. It might be nice to pair players together in the long run by their performance together over time, but we’re not at a point where we want to try this yet.

Backpack.8517:

http://i.imgur.com/AdvKQMr.jpg
http://i.imgur.com/etjI8C1.jpg
pls….
I was rank 2 team Q, rank 30 solo q before the patch..
I had a very high MMR, and I rarely played with or agaisnt bad players….
WHY AM I PLAYING VS THESE PEOPLE (And with)

Btw, we won 500-190…. Someone died to the boss once as well.. Pls anet pls

Just took a look at your account. Your profession MMRs are taking a while to catch up to your account MMR. Not my fault its so dang high. :P

Lexiceta.4156:

idk I thought the idea was to lessen duplicates, maybe I am misunderstanding
the new info.

I’ll look into it tomorrow, but we can’t do anything about people who party with multiple of the same profession. Don’t know if that is what is happening here, but I wanted to make sure that was clear.

Ithilwen.1529:

I think that the major ques are these:

Seeing the same (winning) group of players repeatedly.

Seeing groups entirely or mostly composed of one guild.

Coordination in a team that makes it clear that they are on voice or otherwise highly experienced.

Admittedly; This last cue could represent a group of random high skill players or an organized group. The effect is the same either way.

I think not factoring MMR is as big a mistake as factoring it only. The game should factor MMR and other statistics.

If this is true, it would suggest we increase the scoring penalty for rank. I’ll take another look at the data tomorrow.

Ithilwen.1529:

@Justin ODell: This debate is about perception, not statistics.

This is not a debate over numbers. It is over subjective experience. In my opinion, the experience of PvP has suffered badly because the skill levels are blatantly unbalanced.

This is exactly why I asked the question. I wan’t to know what ques people are picking up on that are leading to this perception. There could be something else going on, maybe if we gave players more information it would help, etc.. There could also be a disconnect between the data, which I’ll need to look at and see if we can make improvements.

Not losing MMR means exactly nothing emotionally when you get steamrollered.

Agreed, which is why we moved away from the old matchmaking that only factored for MMR.

Evan and I were talking about adding roster size logic to the filter part of the algorithm, which would match what you’re saying here. We’d need to expand the filter over time though, because we’re pretty hard set on making sure people get a match.

We’re going to run some simulations and go from there.

Mike.4830:

much worse..like i keep fighting 5 engis that r best engis game… and we all solo q

Um… I just made these changes. Shortest match eva?!