Guild Wars 2

someone.7485:

@Justin: I am glad you keep trying to improve matchmaking and everything. However, a quick forum post before you do it would be nice. Just tell us what disruption you are expecting. I do not even care if you tell us what you are up to at that time, just nice to know what to expect.

I was a conscious choice to not leave a post before the changes (there were multiple) to try and gauge community reaction without bias.

Quite a few people here tend to exaggerate, and it makes my job harder.

Solstice.1097:

P.S. – I’d share my match history if it was available to me…. but if you could see it, it would show I am more often than not expected to lose not because the enemy is high MMR but because my team is very low.

64 Ladder Points, 58 Wins, 69 Losses, 1 Desertion, 5 Buys, 9 Forfeits

I did a partial-reset of ratings (meaning deviation only) the other day, and you’re ranked rating is still adjusting, so you have a wider than normal ratings range.

The partial-reset had an effect on everyone, so that’s why matchmaking has been a bit off. It should continue to settle over the next week.

I also turned profession ratings on at 0.25 percent, so if you switch professions a lot, those deviations will settle slower.

I’ll have a forum post with more details soon.

Vieux P.1238:

(4 vs 5) match-up as default Should not be counted in lost or wins on the leader board. It’s just frustrating to loose cuz either 1 player or more is missing for what ever reason. If a player leaves in mid game or so, he till should be punished.

They don’t count toward points, or affect your MMR. However we do role defeats, forfeits, and desertions into the loss column on the leaderboard.

When (time/timezone) and where did this happen (EU/NA, unranked)?

I’m doing some unannounced traffic shaping experiments.

Hopefully this will get match queuing in sync with the natural rhythm of players entering and leaving matches and improve match quality some.

Solstice.1097:

I JUST NOW 10:46pm CST was end of the match, had a different bug happen.

My teammate appeared to D/C from the match – he was gone from top left corner as well as the ‘B’ screen. However he came back shortly after when looking at the ‘B’ screen he was there again, but for like 2 minutes he still never showed up in the map or even in the top left, only on the ‘B’ screen. At near the end of the match he actually loaded into the match.

He was gone for at least 2 minutes but he wasn’t missing from the ‘B’ screen for a full minute. It seems to be a scenario that it should be treated as a desertion but probably wasn’t.

I don’t see a second match. That doesn’t mean there isn’t a bug, it’s probably just related to how the details of the single match are sent or processed.

Still looking, thanks!

Do you happen to remember the time? In general, if you can tell me that, the timezone, the account (if not the one you’re posting from), and the datacenter (NA/EU) it goes a long way toward helping me find issues faster.

I know you said you ‘just’ had the bug, but the relative time the forum uses, and however long it took you to post, makes it harder for me to narrow down a game when there can be any numbers of games played in that time window.

Shockwave.1230:

Justin you should target your query to matches where there are MMR outliers of some kind. 5 v 4+1 in the top quartile of players for example.

I enjoy advice like this. I wrote a ladder sim a week or so ago that basically replays the season with different rules. I’ve also found it a very handy, and faster, way to get stats out of the data. Been adding all sorts of reports to it as I get time.

Mrbig.8019:

This is actually proving my point.

I wouldn’t necessarly jump to that conclusion. The data could also be explained by the unique lord mechanic which can add 150 to your team’s score. Unfortunately I have no data on how often that happens.

ArthurDent.9538:

Those are actually really small differences and I am pleasantly surprised at how close games are on average apparently. It does show that courtyard, and to a lesser extent foefire are outliers compared to the other maps. Also why no data on Nifhel?

The data is relative to Forest, and not actual numbers. We can be picky about which data should be shared, so I thought relative data would be safer to share w/o asking first. :P

Sins.4782:

I’m kinda curious, how similar are the numbers from unranked and the numbers from ranked?

The same but with a little variance. If you include hot-join the numbers become very different, but the order is almost the same.

Just dropping off some data.

Maps in order of mean difference between team scores in the last week of unranked arena.

1. Forest of Niflhel
2. Battle of Kyhlo +36
3. Spirit Watch +40
4. Temple of the Silent Storm +42
5. Skyhammer +44
6. Legacy of the Foefire +72
7. Courtyard +124

insanemaniac.2456:

no, you haf to offer rewords for the test season cuz udderwize ppl no tryhard

bad data

We aren’t doing anything with the data collected during the off-season, other than to maybe short-term spot checking of changes we may include in the next test season.

Maugetarr.6823:

That’s an interesting way to display the differences. At first I thought the MAMM was dragging the team up too much before realizing the scales on the y and x axis were slightly different to emphasize the curves of the lines. It’s cool that the geometric mean has a little more volitility when all of players are around the same skill levels while the MAMM has more volitility in the higher differences.

Could you maybe apply the geometric one to people queuing solo to limit their highest value while applying the MAMM to premades to account for the communication between higher and lower mmr people in the premade?

They shouldn’t be at different scales. Here is one with them overlayed.

Arietta The Broken.1875:

day no. 28 of horrible match ups vs full teams. \o/

In the last 7 days you’ve had 175 games played in unranked arena. That’s made up of 82 victories, 13 buys, 73 defeats, 6 forfeits, and 1 desertion… that’s 95 wins to 80 losses. Of the 175 games only 4 times were you stacked against pre-mades and did not have an equivalent pre-made on your team.

Your parties, their parties, prediction, result, your score, their score, rating diff, rank diff.
1. 1112 5 60.1600 Victory 500 352 124.0 0.0
2. 1112 14 52.5000 Victory 500 171 35.0 23.0
3. 11111 14 50.0000 Desertion 176 501 70.0 -10.0
4. 11111 5 54.6900 Defeat 353 500 103.0 8.0

In #1 and #2, you had the MMR advantage and won. In #3 you deserted and your team lost. In #4 you lost but had the rating and rank advantage.

Rising Dusk.2408:

Justin ODell.9517:

Hey…. you had a 5% chance of winning. Don’t blame me if you couldn’t pull it off.

Any game where a team has a less than 30% chance of winning shouldn’t even be allowed to happen. I get that all of the other details of matchmaking are important, but match-ups like that are really the ones that result in the worst play experiences.

I literally just had one myself, so yeah, I feel you BLU.

I also agree, but I also have to balance it with queue times. In all likelihood, the other team probably spent around 6-8 minutes in the queue before it would even begin to consider making a match like that.

I am tracking these sort of matches, and am working to figure out exactly what we can do to improve the situation w/o bloating queue times to absurd levels.

I threw together these three plots of the three most interesting ways of getting the mean of a team. The geometric mean, the (regular) arithmetic mean, and the mass average.

Y is the adjusted mean of the team, X is the 5th member on a range from 1000 to 3000.

Hopefully this gives you a better idea how each type works.

BLUna.7928:

Game ended at approximately 9:18 PM Eastern, and was played on this account.

I will admit before going on that matchmaking has been better since the patch (been keeping a spreadsheet of match data) but I just queued into something that confuses me on how it was even able to pop. Details below:

• Full premade vs full solo players.
• Premade has decent comp.
• Solos (my team) have 2 mesmers and 2 thieves.
• 4v5, our team’s fifth never even loaded into the map.

Putting the 4v5 aside for the time being, why was a comp with 2 thieves and 2 mesmers even allowed to be put against a full premade? Ill provide a screenshot of the aftergame screen as soon as my stream catches up to it on the vod. Was just a bit upsetting. The worst part was everyone in the match unanimously agreed there was no point in even attempting the match. I ended up AFKing to grab a drink. :/

Hey…. you had a 5% chance of winning. Don’t blame me if you couldn’t pull it off.

In all seriousness, it probably happened due to limited availability at that end of the distribution.

As for the profession duplicates, that’s because I had to weaken that scoring because the profession dailies were stressing the system and causing bad matches when it came to MMR and party size.

Archaon.9524:

Justin…please…

Archaon…please…explain. I’m busy, it helps if you point me in the right direction.

I wouldn’t think meta-builds are something you should expect in unranked. It’s more casual, so I would assume more people are willing to play whatever they want and experiment, instead of going with what is most optimal.

TPMN.1483:

The solo q needs a different MMR than when running as a team MMR or 2-3 premade MMR.

I was considering this a while back. You could do a linear interpolation between the solo MMR and team MMR based on the number of players in the party.

I haven’t thought about it since then, so it would be good to get people’s impression.

Maugetarr.6823:

I didn’t run the numbers before, but that first team has a weight average of 2066, while the other team has a weight average of 1518. This produces a PDI of 1.3 and 1 respectively, showing that even though they have the same number average, the first is being lowered by having 2 low mmr players.

OK, I think I understand the concept.

I think what you’re suggesting is using the mass average molecular mass formula.
http://en.wikipedia.org/wiki/Molar_mass_distribution#Mass_average_molar_mass

MAMM = (a^2+b^2+c^2+d^2+e^2) / (a+b+c+d+e)
Team 1: 2068
Team 2: 1520

I’m not sure how we would incorporate the PDI.

Maugetarr.6823:

Have you thought about using a (molecular) weight versus (molecular) number average like in polymer science? You could use the equivalent of the polydispersity for the system to see how the team values are skewed to the upper or lower mmr values.

Simplistic overview of weight versus number average :
http://www.pslc.ws/macrog/weight.htm

Polydispersity Overview:

http://en.m.wikipedia.org/wiki/Dispersity

Taking a bit to absorb the concept, in the mean time could you tell us what this would look like for the example?

Macmoarning.3208:

Grow up, you read one line and immediately came to a conclusion. I have two accounts, I didn’t know that was important from which one I post my opinion

Actually he reached to conclusion : you have another acount OR you’re a liar. You linked it.

Can we calm this down. I was not trying to call out rainbow.4695.

We’re actually doing a decent job matching by party size, so Iwas curious if their experience could be explained by a known issue, or if it was something new.

Aphrodite.6851:

Today our team played 5 pvp ranked games .
and 5 win.
But LB show 4 win 1 lose.

Looking at your game history, your team definitely had 5 victories on Jan 19th, but someone had a desertion for the second game, so the entire party was also marked as desertion which is included in the loss column.

To get marked as a deserter you must have been disconnected from the game for a cumulative 60 seconds.

We mark the entire party as deserters to prevent exploits where one member of the party will desert to invalidate the result for the rest of their party. We will be changing this soon so that the other party members do not get marked as deserters if their team wins, but the player that was disconnected for a full minute will always get deserter standing.

Laserbolt.6731:

So desertions should not end the discussion.

I agree, and I did not mean to imply that. I was merely pointing out that these sort of examples are far more likely to be an issue when dealing with pre-mades, and not a matchmaking flaw.

Solstice.1097:

methinks math should not allow 1 team to have both extremely high and extremely low MMR players.

Well, Laserbolt.6731’s example is pretty extreme. The matchmaker shouldn’t ever produce the example in this thread with all solos.

The reason for the original example (in another) thread was because it was a pre-made team that contained 4 higher MMR players, and 1 player with an artificially low MMR (due to desertions).

Adding the math for a few of the suggestions I’ve seen.

Geometric Mean = (a*b*c*d*e)^(1/5)
Team 1: 1273
Team 2: 1517

Center = (min(a,b,c,d,e)+max(a,b,c,d,e))/2
Team 1: 1450
Team 2: 1525

Median = c
Team 1: 2200
Team 2: 1500

Before saying one is superior, try finding contradicting examples.

For example, I don’t like median for instances like this:
(2500, 2500, 1500, 1500, 1500) vs (1500, 1500, 1500, 1500, 1500)
Both have a median of 1500, but clearly team 1 would have the advantage.

Chukree.1756:

Your leader board is ridiculous and undermines any form of competitive plays. It’s a farm board, where loses means kitten even. Take a look at my current rank. Only guy in the top with a win % at 88. My wins are far more superior than any current player on the leaderboard, and my loses are very few, meaning that my rank points shouldn’t have been deducted in any way. People with the same wins and a shocking multiplication of 5 times in losses, have, for some ridiculous reason, more points than I do. I have even experienced people getting points for their loses on the leaderboard. It’s a joke. Not feeling a need to be rude, I just think that is inevitable.

You’re certainly entitled to your opinion, and we are aware of this criticism.

We’re currently simulating various rule changes by replaying data from the test season, so if you have any ideas on how to improve things please feel free to share them.

We’ll also be creating a postmortem thread for the test season soon, but we’ve been really busy and haven’t had time to prepare for it yet.

insanemaniac.2456:

do you hold on to historical mmr records from like a month or 2 ago?

We have logs of every game and every time your rating changes, but those are not stored with the database that contains the current player data.

If you’re thinking we could roll back to a specific point in time, it’s possible but probably not worth doing. It would be easier to just use team arena rating, which is what we used when we imported player data from the old system (which happened on and after Dec 2nd).

Tao.5096:

Matchmaking isn’t that bad actually.
If I’ll solo or duo queue, final score is around 470:500.
Obviously it happens, when we get 500:120, but it’s rare occurrence since last update of the matchmaking.

Justin, it’s possible to make size team matchmaking and add a rule to fit the rating.

For example, could make like this:

• each team is made with rating in certain range(+/- 100)
• teams would be grouped with similar rating value (+/- 100)

That system could avoid going full premades who are veteran players with High rating vs premades that are rather casual and their rating is somewhere in the middle.

The only downside I can think of, is increased waiting time for system to match teams properly.

The system already does this to a degree, though not in discrete windows. The thing that can throw it off are the fail-safes in place to prevent players from being a queue for too long.

insanemaniac.2456:

rollback?

reset seems… idk i dont think itd be effective?

Well, there are other options.

• We could reset just the deviation
• We could offset the rating so that the new rating minus a reset deviation of 350 is equivalent to the old rating minus deviation.
• We could copy team arena, solo arena ratings, or an average of the two.

Etc.. If you have a good idea, let me know.

Solstice.1097:

I didn’t think desertion lowers MMR just LB points???

It doesn’t as of last Tuesday (which changed this logic and added a grace period), but most desertions were from before last Tuesday anyways.

In retrospect this was a mistake. I don’t remember the exact reasoning for it, but I believe it was related to the original MMR based leaderboard.

Solstice.1097:

dont tell me i lost a point for that match >.<

OK, I won’t tell you.

Solstice.1097:

i just played 5 pugs vs a premade that included toker nos supcutie and 2 others

and the next game i won by ~350+ points

i have generally not had very long queues

maybe it shouldnt resort to the extremes so quickly.

I can mostly explain the first match. They had a very low rated player on their team, which skewed the average much closer to your team.

I don’t think it was intentional though, as the low rated player had a lot of desertions that probably explain why they were so low.

I’m wondering if we need to do a blanket ratings reset for anyone with a significant number of desertions.

hackks.3687:

You can’t put experienced players in the same match as your veteran players. It ruins the experience for both parties. I don’t want to play against players who are just learning their skills for the first time.

I get what you are saying. Putting players together based on skill is the central point of the matchmaking system. It may not always succeed, but it always tries to do just that.

hackks.3687:

The player experience at all levels is suffering. This is not a new issue but it’s become far more stark very recently.

I don’t agree with this, but I understand that people must put their own experiences first. The vast majority of players are getting good matches, though I am aware of some areas (the outer edges of the spectrum, for example) that are having a lower quality experience.

hackks.3687:

Food for thought: Perhaps the learning curve of your game doesn’t match the scale of your ratings system? Perhaps consider redefining your scale in order to create values that can’t be reached by either extreme of the player or party size spectrums?

The rating system scales to our players, we don’t provide a scale. It spreads and contracts as a result of the games people play. The normal distribution it produces is more a result of the players themselves, not the system.

Khal Drogo.9631:

In case anyone is interested, I am not mad because I lost. In fact the score need not even be close as long as my team knows the basics of how to play their class and the PVP map mechanics they are on.

You have 11 desertions in unranked arena. If those desertions happened before last Tuesday then they probably lowered your rating quite a bit. This would explain why you are getting matched with players with a skill level lower than you expect.

Desertions no longer modify MMR, so this won’t happen any more, but for some players the damage is already done. Keep playing and it will gradually move you upward.

hackks.3687:

The problem with matchmaking seems to be fundamental – the new player rating is simply too high. There’s no reason a player with thousands of matches played should be able to share an MMR with a player playing their very first match.

Moving the starting rating only moves the center. Also, starting players are matched by their rating – their deviation, and we prefer to make matches the same way, so I don’t believe that is much of an issue. This doesn’t mean we couldn’t try to find a better placement for players that have PvP experience but don’t yet have any games played under a rating type.

hackks.3687:

The worst part is this is just one of a variety of examples that prove that the MMR system is not OK. It is logically and fundamentally flawed. Whether it be by adding hard filters, or tiers, or adding barriers to entry something needs to be done to segregate players based on experience because the game is not played the same at every level. And that is what creates the negative experiences that hamper PvP’s growth.

I think you are conflating MMR and matchmaking, they are not the same. Matchmaking rating (MMR) is only one metric that is part (althought a core part) of matchmaking algorithm. There are lots of variables that can influence matchmaking to create a bad match. E.g. duplicate professions, party size, queue population, how long you are in the queue, how long others are in the queue, etc..

truthishly.9513:

Just a brief update, since going 28-44 over the weekend, I’ve been 21-12 since, so maybe it was just a really (statistically improbable) unlucky streak? Or maybe I should just stay out of solo queue on weekends…

It could have also been that your rating has settled closer to a proper level. Please let me know if you run into any other odd streaks or games.

LoopySnoopy.7923:

There’s 2 of us me + 1 other we seem to get a decent match evenly spread both sides are equal in ratings.

We add a 3rd player to our team another guildy the balance then shifts we seem to get an imbalanced opposing team they’re either weaker then us or they are hugely stronger then us.

We add a full guild group to the mix and the opposing team all seem to have a larger skill base then us,higher mmr,all r80.

Unfortunately this fits with what I posted earlier. Since we are weighing party size so heavily, you’re getting matched other parties of the same size (or close to it) even though you may not be a good match according to MMR. Again, this suggest we could reduce that scoring some, or increase the scoring for rating.

We basically need to find a balance between two competing goals. On one end we want to match based on rating (which is a stand in for skill), and on the other we want to match based on party size (since premade parties have an advantage.)

Last Tuesday we introduced a change that adjusts a party’s rating based on the number of players in that party. The intent is to match pre-mades with and against higher rated players, assuming that matching a premade against higher skilled solos is less offensive than a premade with equally skilled solos. However, I never went back and reduced the scoring for party size.

Blair.3796:

rainbow.4695:

Yes I have two accounts. Does it really matter ?

He wants to see your match data for two reasons. The first is to analyze exactly how your matchups went, in case there are more problems than the devs realize with the current mm system.

The second is to make sure you actually did experience these situations, and aren’t simply misunderstanding, overreacting or intentionally spreading misinformation. Which sounds kind of overly suspicious and a little Big Brother-y in my opinion, but if I were a dev I’d probably want to be able to make sure of this as well.

It’s mostly the former, as I take complaints very seriously. The latter is just a neat side-effect that is sometimes amusing.

There is also a third reason. I’ve found helping others understand how matchmaking works and what’s going on usually calms things down a bit. I believe a good deal of the frustration involved is just having a lot of the information invisible to you, preventing you from reasoning about your experience.

rainbow.4695:

Is as simple as that….for solo players like myself is very frustrating to go against premade teams , why you ever though that this would be fun or fair…..

I couldn’t find any game data for you in the last week. Do you play on another account?

Morwath.9817:

Justin ODell.9517:

Were you expecting it to not show a loss?

Exacly. Since if it doesn’t counts, then it gives false image.

It’s buy design that forfeits still show as losses in the win/loss columns. Though I suppose I could have worded the patch note to make that more clear.

Morwath.9817:

Today at morning for example I’ve played Ranger, lost 4v5 and gained loss (guy was DCed early and never came back). Was unable to monitor if I’ve lost points for it, because I’ve played more than one game at that time. However loss is displayed.

I guess it gave me -points, same as I got +points for 5v4 (I think you can see it in my first 5 ranked games after preseason lanuch).

I found the game, and it’s showing it as a forfeit and had no ladder point change associated with it.

Forfeits will still show in the loss column, even though they do not affect your point standing. Were you expecting it to not show a loss?

Marcos.3690:

If you weigh party size too heavily you have this kind of threads
If you don’t you’ll have premade vs soloplayers complains

I don’t think there is a way via matchmaking to fix a low playerbase

I agree with you in general, but I may have been a bit over-zealous with the scoring changes I made. It’s at least worth looking into.

Archaon.9524:

U freaking serious anet? 5 pugs vs 4que gg

You got rekt. :o

I’m looking to figuring out exactly how that happened. Not only did it put you against a 4-player premade, you team also had a much lower rating. The rating of the other team isn’t even so high that it couldn’t find better matches.

Just pulled up your game history. I believe this is caused by us weighing party size so heavily to prevent stacking solos against pre-mades.

Since you seem to play near exclusively with a full party, we’ve preferred matching you against other full parties, of which there are fewer queued at any given time. This reduced selection limits match quality, which is why you get the occasional uneven matchup.

I have some changes coming next Tuesday that may help, and I’ll see about easing up the party weighting some now that we had some changes that came in last Tuesday that also helped.

Morwath.9817:

Losses will not count toward the leaderboard standings if a teammate outside your party has deserted. - Patchnotes

Thats not true.

Do you have a specific game date and time in mind where you believe that was not true for you? I’m querying your game history now, and that info would help me find it quicker.

MrKrataus.6420:

Did you tweaked something ? Because I see 6ppl in the top 25 lb with more lose than win, and when I see the first Guy with 80 points (80w/40l), I wonder if losing games matter now since I get only 1pts per win with 0lose

Can’t we reach the top without being able to spam games all the day ?

We’re in an off season, there have been no rule changes, and there will be no rewards. I suggest you only follow the leaderboard as a way of gauging your personal progress.

We may also change the off-season rules at any time to experiment with scoring changes ahead of the next test season.

Vatras.2806:

As I could not find a dev post about my particular question:

Now that the ladder has been reset and the rewards for the top 250 players have been sent out can we expect to get the same rewards (glorious hero armour pieces, mini llamas) again once this ladder will be reset?

I heard the phrase “off season” going around in this forum, but I could not find any information about possible rewards of the end of this ladder cycle.

Random Weird Guy.3528 is correct. This is an off-season, there will be no rewards. We enabled it just so people could monitor how they perform, if that is something they wish to track.

Lordrosicky.5813:

Justin is my MMR on this account terrible or do I just keep getting paired down when I duo/trio/etc queue?

Seems like I never play with good players….

You’re currently 23 and 3 for the off-season. You must carry so hard. :P

Your ranked rating is a bit higher than average, but lower than your unranked rating. This suggest to me that your desertions, of which there are more in ranked, may have lowered your rating and it’s still adjusting upward.