Guild Wars 2

Boggs.6482:

I just joined my first competitive team this week, and we’ve been testing comps and learning each others playstyles in ranked. Don’t quote me on this, but this week I believe we’ve only face 2 premades running the same guild tag, a few premade pugs (lol), and the rest have been pug groups.

I actually just posted on the forums looking for teams to scrim, because with such inconsistent matchmaking, we have no way to judge or skill level and improve. We don’t want to fight pugs. We’re queued kitten men on voice com, and we would like to fight another 5 man on voice com without having to spend hours on the forums :’c

(Please don’t shoot meh down, Justin. Don’t tell everybody how noobish our team is and how perfect your algorithm is.)

Sorry, but I’ve got to shot yeh down, Boggs. Since we can’t detect voice coms, I can’t include it as a factor in matchmaking.

We do have some knobs that we can turn to increase competition for teams, but we have to be careful because a premade team does not always imply more skillful play.

Flumek.9043:

Josh (Davis aka Grouch) is the messiah of pvp.

Here is Justin (ODell) , the wizard who programs your bad loot & and matchmaking. He can also cast a curse on you which does +400 ping.

Hey now… you can blame me for matchmaking, but I had nothing to do with loot.

Trevor Boyer.6524:

After hundreds and thousands of matches, it is very visible that your MMR system functions as this:

• Ten players que at the same time
• High MMRs are 10, low MMRs are 1
• The ten players qued look like this: 10, 6, 5, 4, 4, 3, 3, 2, 2, 1
• System tries to balance two teams from these MMRs
• Result looks like this -> RED: 10, 4, 3, 2, 1 --- BLUE: 6, 5, 4, 3, 2

This is not how the system works. The old solo queue worked like this, sort of.

You can find more details on matchmaking here:
http://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

The current system is more likely to produce a result like this…
Red: 10, 5, 4, 3, 2
Blue: 6, 4, 3, 2, 1

Which could be argued is less fair, but really because MMR isn’t the only factor this is an oversimplification.

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

Evan Lesh.3295:

insanemaniac.2456:

oh hey evan… question: patch notes or whatever said you guys changed the scoring matrix a little, but ive never seen… say… an updated version of the table that was in the blog. how was it updated?

Still need to update the wiki.

Updated.

wwwes.1398:

Another 6 losses in a row throughout the entire day across multiple classes. Maybe 7, I lose track.

Please take a look, Justin, it’s starting to just not be fun anymore.

I can’t do anything about games like the last one you played. The final score was 473 (you) to 502, which is a close match, and I’d just as well call that a matchmaking success.

Other than Courtyard and that one Spirit Watch, most your matches are close.

uberkingkong.8041:

Where are all the good players at yo?

Define good? Your rating is average, so you should be getting matched with the average player.

On Friday I lowered the inflated MMR parties receive from +5% per member to +2% per member. I believe the initial value of +5% was too high, and we were aiming to overshoot due to the bad perception of solo vs pre-made. The drop to 2% was based on data from the beta ladder.

Could that may be what you noticed this weekend?

Thanks for all the reports!

I’m going to lock this thread for a bit while I catch up.

Should be fixed now.

Sorry, this should be fixed now.

I just checked the numbers for unranked arena.

Skyhammer has 7.5% of the votes, and is the map used in 7.2% of games.

I know quite a few people that enjoy the map, so I wouldn’t just assume only trolls vote for it.

Johje Holan.4607:

Justin ODell.9517:

leman.7682:

think we’re going 4-0 for me

Can confirm, this is true for the last week.

Sounds like the devs encourage flame wars…

Evan made me do it!

Archaon.9524:

what is top 5%? u.u

He’s being modest, his MMR is actually in the top 2%.

leman.7682:

think we’re going 4-0 for me

Can confirm, this is true for the last week.

Your game history shows you won 5 of the last 10 games you played, and have 302 wins and 270 losses on the off-season ladder. Your rating is slightly above average. Nothing I’m seeing suggests a problem.

We’ve had an increase in players recently, maybe you’re feeling their presence?

Another shot in the dark… We changed balance recently, maybe your build needs a little tweaking.

kitten.1685:

You misunderstood. Click where it says “1 minute ago” and it will change to “16:19 today”
sorry typod click to clue

TIL

kitten.1685:

1. If you clue “About an hour ago” it will show you the exact time
2. How did you write my name without it getting censored??/?

1. Do you know how many games take place in an hour? Actually I have to check a two hour window because of rounding.
2. I’m a kittening server programmer, we can do anything /s

SobeSoul.6910:

The problem with player ratings system and glicko is there is zero chance of ever arriving at a statistical model that accurately measures a person’s worth in a team game.

The unobserved data isn’t like models in econometrics. A lot of unobserved data in matches like little things that turn team fights around need to be accounted for to ever make a truly make a player ratings model work.

I have to disagree with this, but if you can come up with a better idea I’m interested.

kitten.1685:

and just got 500-60d by a 4-premade (all decent players with mmr around mine) in a full soloq team, all of which werent even r80 (definitely low mmr players)

Moobs! You’re killing me here. The forums displays post date in relative time and isn’t accurate enough to be useful. Be kind, post time.

rotten.9753:

…

Correct. I’ll update the original post.

uberkingkong.8041:

Also, your we rate people on wins. Like many have said, people get carried. That’s a stupid way to rate people. You need to use a player rating system.

I believe TrueSkill uses player ranking in the game when updating ratings, but I’m not sure that would work as well for us. I’d love to use TrueSkill, but it has a patent and I hear Microsoft doesn’t license it out to competitors. Also, our point structure isn’t designed for this, and would probably need a bit of an overhaul.

People being carried only matters if they are constantly carried by the same people, but it is true that this causes their ratings to take longer to reach the same accuracy.

I’ve been toying around with my own rating system in my spare time, but it’s not well tested or verified, never mind production ready.

uberkingkong.8041:

Me and another good player with 3 newbs vs 3 good players and 2 newbs is what I usually get.
I’d rather have a me and 4 good players vs a premade.

It works on both sides. The players added to your team will be at the inflated rating. I.e. if your regular rating is 1500 but after party size inflation it becomes 1700, the missing spots will be populated with players as close to 1700 as possible.

uberkingkong.8041:

There system of detecting good players is bad, they need a player ranking system to tell whose good and not, so whenever I do go against premade its not me and 2 good players and 2 newbs vs premade (hence we lose, 2 weak links).

Not sure your logic follows given the misunderstanding above. To put it as simply as possible in this context, the rating system doesn’t care who is good and who is bad, it only cares who wins and against whom they were playing.

oneply.9586:

i have a feeling points farmers get marked by the system as good/average players, when really they are bad and cause their team to have to overcome their selfishness to win the match.

By point in “point farmers” do you mean ladder points or individual points? Ladder points work off of your team’s performance, so there should be no conflict of interests there.

Meri.7512:

…

Do you live in EU and play in NA, or vice versa?

Cindas.6149:

Second, if “being matched against a premade when you’re solo is rare”, why have it be a possibility at all? If it is rare, what does it give to the players and the system? It appears it is only reducing the queue time for “high skilled players”.

We also allow it because a lot of the time it doesn’t matter, especially now that we inflate party MMR by the number of members to compensate for the advantage.

Math: party.rating += party.rating * config.party.inflation * pow(party.size – 1, config.party.curve)

5 players with a mean of 1500 becomes 1845 with inflation at 5% and the curve set to 1.1.

There is still that occasional match though that gets totally ruined by the fail-safes. I’m not certain waiting longer for those players would improve quality for them or just reduce the number of games they get to play.

Edit: Correcting algorithm to use party.size – 1, which matches the LIVE code.
Edit 2: Fixed another issue in the way the algorithm was presented. Thanks, rotten.9753.

aussieheals.6843:

This is bullkitten – you’re oversimplifying a complex situation to the point of no longer adding anything useful to the discussion. <3 from reddit :P

Eh? I was asking for more clarification because the most common complaint didn’t fit with the data.

amery.6417:

fact is the patch changed it so I now have a chance to face an unfair match.

You’re entitled to your opinion, but the fact is this has always been true.

The old system would produce unfair matches all the time. It did not factor in PvP rank or professions at all. Even in regard to the one thing it did look at, rating, it could still produce unfair matches based on availability. And this was just in solo arena. In team arena there was a whole other set of problems.

Right now being matched against a premade when you’re solo is rare. There are some exceptions to this, i.e. high skilled players, which face a greater availability issue. Still, when it does happen you are more likely to have an edge when it comes to rating. This is a new addition though.

When did this happen? I’m not seeing it in your game history.

leman.7682:

Have been playing since betas, used to play top100 soloQ, 6000 PvP games…

What happens? 500 AP (indicating a new player), ‘first time PvP’ players on my teams. They die to Svanir, they triple cap points, they don’t read the chat, they don’t look at the map. Absolutely horrid experience. SoloQ was okay at times, in the current system it would be more than fine.

But no… we have to get this…

Justin, I’m begging you – do something about it. I don’t care about ratings, esports and rewards. I just want to have fun after coming back from uni and the office…

All your recent solo games have had players near the same rank as you. Is there a specific game you have in mind?

Gabriell.4856:

Anyway to allow us to know our chance of winning before the match starts?

I would like to quit while I’m behind and get it over with as quickly as possible.

For ranked arena during a ladder, we plan to make the information available. I wouldn’t recommend leaving in one of those matches though.

The hot-join/practice game choosing logic is pretty simple. By simple I mean it doesn’t do anything except some bare minimum checks.

I would suggest using the game browser if you want better control over which custom arena you join.

You could also give unranked arena a try.

Gabriell.4856:

2015-01-30 19:08:08.000 Ranked 1112 5 10.7800 Defeat 132 500 -226.0 8.0

10.7% chance of winning…wtf? Why even start the match.

These sort of matches happen most often due to fail-safes we have in place to keep queue times low. I would guess the premade team had to wait around 8 minutes to get that match, and they didn’t like it either.

I’ll continue to work to make these sort of matches happen less often. I don’t think we’ll find a comprehensive fix without sacrificing queue times though.

Hey guys and gals,

In an effort to streamline how this information gets communicated, I’ve decided it would be better to start a thread so we can keep it in one place.

These are the sort of bad matches I’m looking for:

• All solos vs full premade.
• One side wins by a large margin (250+), excluding courtyard.
• One team has killer hares and another has sickly dragons.

What I’m not looking for:

• Profession stacking caused by PvP profession dailies.
• Matches lost due to players leaving.
• Information about winning or losing streaks.
• General complaints about other players.
• Specific complaints about specific players.
• Any complaints about professions or balance.
• Screenshots.

Please post the time, timezone, the map, which arena (ranked/unranked), the datacenter (NA/EU), which account you were playing on, and a brief description of why you think the game was bad.

If you don’t give me this information I won’t be able to find it.

As this is only a tracking thread I won’t be replying, and would appreciate it if we could keep discussion to a minimum. If you would like some info about the game just include that in the post and I’ll try to get back to you.

holodoc.5748:

Justin ODell.9517:

GeneralJimmy.2186:

Every single game is horrible
Keep getting unfairgames vs 3/5 man

I’m not 100% sure what you’re referencing here. Are you being stacked against pre-mades, or are people on your team leaving?

Looked up your game history for the last week and you’ve had 5 out of 67 games that had the pre-mades stacked against you.

2015-01-24 20:20:50.000 Ranked 212 5 46.0900 Victory 500 371 39.0 -4.0
2015-01-24 23:14:16.000 Ranked 11111 41 54.0600 Victory 500 437 73.0 13.0
2015-01-29 16:31:56.000 Ranked 11111 311 50.3100 Desertion 475 500 41.0 0.0
2015-01-29 18:26:21.000 Ranked 11111 5 34.6900 Defeat 228 502 -25.0 7.0
2015-01-30 19:08:08.000 Ranked 1112 5 10.7800 Defeat 132 500 -226.0 8.0

The last two are the only ones that stand out as particularly bad, but could be explained by your deviation still settling (hopefully, will need to dig deeper).

Excuse me but how exactly can you distinguish between PUGs and premades based only on logs and game statistics?

Sorry, I forgot to include the headers.
Date, Time, Arena, Team Parties, Enemy Parties, Odds of Victory, Result, Team Score, Enemy Score, Rating Diff, Rank Diff

212 vs 5 = 2 parties of 2 + 1 solo vs a party of 5
11111 vs 41 = 5 solos vs a party of 4 + 1 solo
etc..

jportell.2197:

My only question is why was soloque removed? … So my big question is why were the ques remerged? What happened, and what is the reasoning behind remerging the ques?

I’ve answered this before, but with the forum search being what it is, I’m not surprised this one keeps coming up.

The basic answer is this: We felt there was more value in having an unranked and ranked arena more than value we had in ranked solo and ranked team arenas.

We believe it is better in the long run for many reasons, but the two I remember most are: 1) Players want to play with their friends in a structured way. 2) Many players avoid ranked play for any number of valid reasons.

1 and 2 could both be done through hot-join, but that isn’t the sort of experience we’d prefer players to have. 1 could also be done through team arena but that doesn’t solve 2.

There are other reasons, but I don’t remember them all right now and I’m not at work so I can’t just go look it up.

I’ll just leave these here…

• Anchoring or focalism The tendency to rely too heavily, or “anchor,” on one trait or piece of information when making decisions (usually the first piece of information that we acquire on that subject)
• Availability heuristic The tendency to overestimate the likelihood of events with greater “availability” in memory, which can be influenced by how recent the memories are or how unusual or emotionally charged they may be.
• Availability cascade A self-reinforcing process in which a collective belief gains more and more plausibility through its increasing repetition in public discourse (or “repeat something long enough and it will become true”).
• Base rate fallacy or base rate neglect The tendency to ignore base rate information (generic, general information) and focus on specific information (information only pertaining to a certain case).
• Clustering illusion The tendency to overestimate the importance of small runs, streaks, or clusters in large samples of random data (that is, seeing phantom patterns).
• Confirmation bias The tendency to search for, interpret, focus on and remember information in a way that confirms one’s preconceptions.
• Conjunction fallacy The tendency to assume that specific conditions are more probable than general ones.
• Conservatism or regressive bias A certain state of mind wherein high values and high likelihoods are overestimated while low values and low likelihoods are underestimated.
• Conservatism (Bayesian) The tendency to revise one’s belief insufficiently when presented with new evidence.
• Curse of knowledge When better-informed people find it extremely difficult to think about problems from the perspective of lesser-informed people.
• Focusing effect The tendency to place too much importance on one aspect of an event.
• Negativity bias Psychological phenomenon by which humans have a greater recall of unpleasant memories compared with positive memories.

One’s I’ve noticed myself being guilty of from time to time, probably not a complete list…

• Clustering illusion The tendency to overestimate the importance of small runs, streaks, or clusters in large samples of random data (that is, seeing phantom patterns).
• Confirmation bias The tendency to search for, interpret, focus on and remember information in a way that confirms one’s preconceptions.
• Curse of knowledge When better-informed people find it extremely difficult to think about problems from the perspective of lesser-informed people.
• Empathy gap The tendency to underestimate the influence or strength of feelings, in either oneself or others.
• Hindsight bias Sometimes called the “I-knew-it-all-along” effect, the tendency to see past events as being predictable at the time those events happened.
• Identifiable victim effect The tendency to respond more strongly to a single identified person at risk than to a large group of people at risk.
• Illusion of validity Belief that furtherly acquired information generates additional relevant data for predictions, even when it evidently does not.
• Information bias The tendency to seek information even when it cannot affect action.
• Optimism bias The tendency to be over-optimistic, overestimating favorable and pleasing outcomes (see also wishful thinking, valence effect, positive outcome bias).

uberkingkong.8041:

See picture,
All those people queing up WvW.

Do you think PvP is ever like this, where over 100 people queing up (Waiting in Que, due to servers full) at sametime, and cant enter into it because all servers are full?

Heh. We get way more than 100 people queuing in both unranked and ranked arena. To be fair, we’re much more picky about letting people in.

ronpierce.2760:

But think about the potential income of PVPers! We could take part of the population from WoW’s arena base who would love this type of freedom and skipping the leveling process. And bring back old GW2 players and convert at least some PVErs. I thinks a bit short sighted to just think it’d be fracturing the player base, when in reality you could just be growing it. Bringing in more players and ex players could do wonders for both arena AND conquest populations at once.

It isn’t off the table, and we have most of the tech for it. I expect at some point we will do it, just not now. I don’t know when either.

Edit: This isn’t a promise we’ll do these type of arenas, just my opinion based off of previous discussions with the team. We may never do them, or we may do something like them. Nothing is planned at this point.

We’re aware of the community desire for these sort of arenas, but at this time we do not want to fragment the player base by having more than two arenas at a time.

GeneralJimmy.2186:

Every single game is horrible
Keep getting unfairgames vs 3/5 man

I’m not 100% sure what you’re referencing here. Are you being stacked against pre-mades, or are people on your team leaving?

Looked up your game history for the last week and you’ve had 5 out of 67 games that had the pre-mades stacked against you.

2015-01-24 20:20:50.000 Ranked 212 5 46.0900 Victory 500 371 39.0 -4.0
2015-01-24 23:14:16.000 Ranked 11111 41 54.0600 Victory 500 437 73.0 13.0
2015-01-29 16:31:56.000 Ranked 11111 311 50.3100 Desertion 475 500 41.0 0.0
2015-01-29 18:26:21.000 Ranked 11111 5 34.6900 Defeat 228 502 -25.0 7.0
2015-01-30 19:08:08.000 Ranked 1112 5 10.7800 Defeat 132 500 -226.0 8.0

The last two are the only ones that stand out as particularly bad, but could be explained by your deviation still settling (hopefully, will need to dig deeper).

William C.6195:

Bumping, if bumping is allowed.

Bumping isn’t allowed.

We should probably create a thread for people to post bad match instances instead of the various threads.

You can also just PM if you don’t want any data discussed to be public.

Solstice.1097:

they are pretty rare as a whole but for some people like high MMR solo queue’ers they are not nearly as rare. we are being punished it feels like even if not true.

This is true, but you aren’t in this category. I don’t mean that in an offensive way, just that if you’re still getting issues over the next week it may be something else.

Cynz.9437:

yeah except those occasions where you get pug team (with not so great players) vs premade of top ESL players (sup ROM) few times in a row _

That’s a totally different issue, and these changes won’t help. I’ll continue to let the team know about the communities desire for a comprehensive fix (i.e. hard filters and/or bad match abortion). Right now it happens pretty rarely so we aren’t quite ready to sacrifice queue times.

Of course now the more you hit them the more the ratings will change, so that could solve itself soon enough, queue time fail-safes withstanding.

Also, let me know if starts happening less often. You look like one of the unlucky ones, and your rating is settling downward toward where you were in solo and team arena.

Solstice.1097:

WE WANT MOAR DATA! even just a raw data API the community would do wonders with it.

It’s definitely on the list, but we’ve got a few things at a higher priority right now. I don’t mean to tease (or do I?), but I think people will be very happy with what we’re going to do.

SobeSoul.6910:

…Allowing us to manually reset are MMR once in awhile. Perhaps once a month or maybe every other month.

With the new back-end servers we can do this, but I’m unsure if or when we’d be want to do this. I’ve mentioned it to the team on several occasions, but I don’t recall what the designers said, and I know it just isn’t as high a priority as other work right now.

SobeSoul.6910:

I am finding myself in matches that are too easy to me because before the switch to the new leaderboards I multiclassed a lot in solo queue and accumulated losses on some of my off classes. I have no idea what else would be possible to help players to play at their skill levels if they’re stuck in a weird MMR spectrum.

Then again I have played against some players like Ostrich Eggs and others who have high MMR in the past month.

These kind of matches can be caused by a number of things, but rating is probably the least responsible. Nearly everytime I’ve looked up a bad match, rating was usually the strongest factor trying to prevent the match.

SobeSoul.6910:

Justin: Do you think MMR sometimes is overvalued metric for skill level considering the game’s design?

No, I don’t think that.

I’m aware there are some critical combinations of things (team comp, # of people queued at your rating, rank, etc..) that can be very detrimental to match quality. Those things will drop quality regardless of rating accuracy or emphasis.

That’s not to say rating is perfect. I’ve got it on my todo list to explain in more detail in a forum post, but here is the gist of two problems that are now fixed.

1. Early on after the Dec 2nd build there was a bug that caused deviation to stay at the max. This bug was caused by a mismatch between how I was calculating the amount of time that elapsed since the last time you played. One section of code worked in milliseconds and the other in microseconds. I had poor test coverage of this code which is why it was missed.
2. Due to some improvements we made intended to more accurately reflect a player’s impact on a team when updating ratings, the rating changed much slower. Unfortunately deviation was still changing at the normal rate which mean people zero’d in sooner than they should have relative to where their rating was. I’m still not 100% certain how much an effect this had on match quality, but I’m optimistic the fix will improve things.

These two impacted some players (especially those that had desertions early on before #1 was fixed) more than others causing them to have wildly inaccurate ratings that took far too long to adjust to where they should be.

These problems are both fixed now, and we did the partial-reset so that anyone affected will stabilize in their proper location much sooner. We could have done a full reset, but it was a minority of players that were hugely affected.

Everyone will benefit from the fix to #1, but you may not notice the impact of the fix because it will still take time for the population to spread out and stabilize.

Notice that both these problems aren’t with Glicko2, but how I set up the servers to use it. Mae Culpa.

Even with these problems, for players unaffected by the bugs the ratings are still mostly accurate. E.g. the top players are still the top players.

Solstice.1097:

so it suggests i usually lose by less than 100 points? or does it suggest im usually favored to lose?

All of the above are possible.

Solstice.1097:

i was hoping you have the data that differentiated between hypotheses.

The data exists, I just need to wrangle it into a report.

Solstice.1097:

because I personally have witnessed a huge drop in quality of allies, which is seemingly corroborated by many posts on these forums in the last 2 weeks. there is something there, and it needs to be found in data’s terms.

Give it a least week before assuming this. We’ve got an influx of new players and they, as well as everyone else, are still settling.

BrunoBRS.5178:

justin, while you’re here, could you check if my MMR.

Please PM me. I don’t have time to do it right now so I probably won’t remember which thread the request is in and forget about it.

truthishly.9513:

I have roughly the same number of lb pts as wins too, but with 23 more losses than wins as of the last hourly lb update, what I infer from this is that I’m ‘expected’ to lose the vast majority of matches that I have lost, or the -points would have dragged my total pts far below my win total. I think the only thing you can reasonably infer from lb pts is whether or not your teams have been performing ‘as predicted’. I don’t think you can read anything regarding match balance into this.

I don’t think that is necessarily the case. Since we allow you to get 0+ points for a loss, you could simply be doing well for most losses with any unbalanced game gains and losses evening out over time.

Unfortunately as long as a loss isn’t necessarily a loss of ladder points, you can’t really make any strong assumptions either way without looking at actual game history.

The best you can do is compare points to wins (victories + byes), but even that is pretty fuzzy. I’ll see about getting metrics for game results per prediction tier.

Solstice.1097:

Justin ODell.9517:

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

Also, the fact that I have more points than wins…

58 wins + 5 byes = 62 points

and still a losing record (when I’ve never been under 50% for a reallly long time) shows that I am more often getting used as an equalizer for unbalanced matches than I am getting actually placed against similarly skilled opponents. Doesn’kitten That’s my main gripe.

Since you’re only 2 points away from your ladder score it could suggest that you mostly get even matches and do at least well enough to not loses point most of the time. I’m not confident any assumptions can me made from that data alone.

If I get time I can run another report for you.

Harbinger.8637:

I know what desertions are, but what exactly are buys and forfeits? I’m guessing that has to do with 4v5 or something of the sort.

A buy means the other team lost but had deserters.
A forfeit means your team lost and had deserters.