Guild Wars 2

What did the scoreboard say on the League tab?

There is also the party and teammate desertions that have been quite confusing.

WuGzaGenius.5684:

Evan Lesh.3295:

Fluffball.8307:

Just out of curiosity is it a median mode or a mean?

Edit: math terms.

It is the mean. The player base distributes into a bell curve like such:
https://en.wikipedia.org/wiki/Standard_deviation

We place people right in the middle.

Why dont these numbers relate-able to League rank distribution?

There are way to many legends to fairly say That it’s also on a bell curve.

The league is more like half a bell curve because everyone starts at the beginning and don’t all play the entire season. The lack of backwards progress in the divisions also changes the distribution.

Fluffball.8307:

Just out of curiosity is it a median mode or a mean?

Edit: math terms.

It is the mean. The player base distributes into a bell curve like such:
https://en.wikipedia.org/wiki/Standard_deviation

We place people right in the middle.

Why start people at the average MMR? Because that’s where most players belong. It is the most accurate rating for the majority of players. For every player we start with a lower rating that needs it, there is a skilled player that has to fight their way out.

There are a lot of reasons this could have happened, but here are a few:

• You haven’t played in a while, so your rating deviation is high. You won a match, your rating went up a lot, and you’re now adjusting back down.
• Other players in the queue are experiencing the first problem.
• You are playing at a different time of day than you usually do and there is a different set of players in the queue than you normally match up against.

This is a problem of rating confidence. New players start with an MMR very close to the average of the player base, but their rating is much less confident than someone who has average MMR and 1000 games played. There are three avenues I see that can be taken:

• Care more about games played than we already do. This functionality was actually added recently and replaced rank consideration, but it’s not a very accurate indication of skill.
• Care more about rating deviation. This number is meant to track with rating confidence, and is higher for new players. We could use this for matchmaking, but it’s already being considered in the ratings formula…
• Let the rating system handle itself. Rating deviation is meant to move players quickly to a more accurate rating. Since this player had only played 2 games, you can be sure that his rating deviation is high and his rating dropped dramatically after that game. You may never see that player again.

I prefer the last route as we let the Glicko algorithm do what it does best. We can’t avoid every single bad match up. As you said in your post, it could have been an alternate account. There is no way for the system to know how good a player they are unless they play, and that means a some guesses have to be made along the way.

philheat.3956:

Evan Lesh.3295:

5 vs 2-1-1-1 is still allowed. The cutoff for roster size differences is configurable, but it is entirely possible that match quality and queue times would not be acceptable at all ladder positions if the acceptable spread were smaller.

Honestly Evan is not acceptable going against a full premade in 2-1-1-1 or with another unbalanced roster size.

Can you explain why like soloer i should lose 1 pip or interrupt my win streak if this happens?

If 5 people realize they have a long queue, they go in smaller grp size, easy, and if they want to train like full premade they organize games with other full premade, like every other decent team in this game do.

Full premade against a soloer or semi soloer is not training, is pure farming.

At least i would add 2-1-1-1 config with 1-1-1-1-1

with 2-2-1 or 3-1-1 or 3-2 you have a bit better coordination (it’s not like a full premade ofc but is a bit better).

We can increase the size restriction, but queue times are a limiting factor. If the number of premades in the queue is healthy, then we can discuss making more strict premade rules.

The configuration limits the size difference between the largest premades. Configured currently at 3, a 5 vs 2 or 4 vs 1 is the largest gap possible. If we lower the cutoff to 2, then 5v3, 4v2, and 3v1 become the largest gaps possible.

Thiefz.3695:

As was stated in reddit. What happens when the one person in the premade leaves the group while in the match and then DC’s? Is the party roster saved when the match starts or only reviewed when the match ends?

If it checks when the match ends, then leaving the group will save the lost pip.

As Erik mentioned, party groups are remembered throughout the match process even if you leave your party manually. The premade brackets on the scoreboard at the end of the match also reflect this.

Cynz.9437:

Sunshine.5014:

Cynz.9437:

Not a word about obnoxious 50% winrate algorithm that makes this whole league/matchmaking system pointless~

Did you even read? Match making is not based on MMR anymore. There is no more 50% force win rate.

Yeah, there is no MMR anymore, but there is so called predetermined pip range. Also, the problem is queue times. I wouldn’t be surprised if matchmaking would increase that so called range to reduce queue times.

Pip range will be the same as season 1: +/- 15 point maximum. This did greatly impact queue times for people at the top, but now that MMR is removed as a limiting factor and players should be able to climb more reliably, there should be a healthier population to pull from in the queue.

LanfearShadowflame.3189:

Sekai.2987:

NiloyBardhan.9170:

From what I understood, you will lose 1 pip whether you lose 0-500 or 499-500 in a match. In other words, in a PuG party once you’re down by ~100 points its a GG for your team since there would be no benefit of making the match close anymore.

if thats true then its a bad change, last season i often saw in teamchat things like:
“guys even if we lose , lets score as much as possible so maybe we wont lose a pip”

and i think that was a good thing , i remember how when i faced proleagues players and i knew its unlikely that i win, but at least i can try to score as much points as possible to not lose a pip, or maybe even gain one , even if i lose

and if that is not possible anymore then i am sure ALOT of players will just stay in the base once the score hit 300-100, saying things like “well we lost anyway, doesnt matter how many points we score”

I see your point here, and I don’t disagree with it; however my concern is with how that did work in S1. It was entirely dependent upon your personal score whether you gained, lost, or broke even. It tended to encourage a lot more yolo play as opposed to actual teamwork. We need a little bit of a happy medium to encourage teamwork, and encourage the losing team to keep fighting instead of just rolling over.

Maybe I don’t understand your post, but personal score is not a factor in the league system (season 1 or 2).

Mr Godlike.6098:

Guys i feel a bit terrified with some parts of this changes.

At the end of seasson one i was at the end of diamond divison. In matches i was playing almost all the time with ppl from legendary dvision and now i will propably have to fight legends in amber in seasson 2. At that time i had almost all the time +1/-1 win ratio.

What if i will have the same +1/-1 in sapphire seasson 2. What if old legendary divsion players will have the same win ratio or guys who stoped in sapphire? Can be possible that if you don’t extreamly improve from last seasson you will be stuck in 3th dvision even if you are legendary skill material?

I hope somebody who is smarter then me and better undestands these changes wil prove that this situation is not possible. I am begging for answers.

P.S sorry for my English ;/

If you are a diamond-skilled player, you would have been fighting more legendary players in amber during season 1 than you will in season 2. Season 1 gave you opponents based on your MMR so every match would have been tough all the way up the ladder. Season 2 will give you opponents only based on ladder position, so diamond and legendary players will be climbing over the large amount of lower skilled players relatively quickly.

Malediktus.9250:

Does this mean soloQing to legendary will be even harder than in season 1? It does sound like it

It should be easier since your teammates will be of more similar skill instead of the team teams having player skill split between them. Imagine matchmaking as if it were trying to make a bunch of 1v1 matches by picking 5 players as similar as possible to fight another set of 5 similar players.

5 vs 2-1-1-1 is still allowed. The cutoff for roster size differences is configurable, but it is entirely possible that match quality and queue times would not be acceptable at all ladder positions if the acceptable spread were smaller.

The reduction of queue times was only partially related to quality at best. There were some configurations that were adding extra queue times for zero benefit.

Not accepting the queue is only a 3 minute timeout with no prior offenses. It sounds like you’re disconnected multiple times in the last day or two to get a timeout that long.

This is unintentional and I am looking into it.

Looking into this.

Now that the box is fixed for anyone who hasn’t opened the box yet, we can now distribute replacement items for those who have.

Matches for some people will be better after the change. The way timing lined up, players on the edges of the MMR curve would go from no matches to all the (bad) matches in one jump.

We’ve been finding more ways to analyze how matchmaking is behaving in order make improvements to the system. Today you should notice a great improvement in queue times as we’ve identified some configurations that were adding wait times for no tangible benefits. Some players will notice a reduction in queue times of up to 3 minutes!

We are aware of the issue and are looking into a fix.

Was the bouncy-chest empty? Or was the consumable item empty?

Division placement is definitely used in matchmaking. Players are matched with others +/- 15 pips away. This is why you get some spill-over into other divisions at the beginning and end. We are changing matchmaking for the second season, and I think it may help cover what you’re after.

If we just use games played, you don’t get a very accurate representation of skill because there could be a great player who has won every match in a division and has low games played and there could be a not-great player who also has low games played but is stuck at the bottom. Should these two players get matched together?

ZilentNight.5089:

Evan Lesh.3295:

witcher.3197:

Evan Lesh.3295:

These are coming, we’re postponing the distribution while some other guild bugs are fixed.

I just hope you guys turned off the decay so teams don’t drop out while we are waiting xP

Yep!

Can you elaborate on what guild bugs? Also does the guild only get 1 decoration token or 1 per person?

I’m not on top of the guild bugs, so unfortunately I don’t have any more information.

The tokens I believe are per person (can’t double check this second). The trophy decoration is limited to one per guild and will automatically unlock when the first guild team member logs in after rewards are distributed.

We still track profession MMR, but it won’t being used for matchmaking unless we have some form of profession locking.

Allowing people to swap to other professions not yet on the team isn’t something we can support. This becomes ridiculously weird/impossible to support when more than one person tries to do it at a time.

We could possibly matchmaking off of unique builds and still allow more than one profession per team, but that means no one would be allowed to change their build then entire time they’re queued or during warm up. Forgot you were testing that other weapon set? Oops!

It sounds like you want profession-specific MMR? In order to do that, we would have to enforce a profession lock upon queuing. There is another thread advocating this idea with the benefit of removing profession stacking.

witcher.3197:

Evan Lesh.3295:

These are coming, we’re postponing the distribution while some other guild bugs are fixed.

I just hope you guys turned off the decay so teams don’t drop out while we are waiting xP

Yep!

These are coming, we’re postponing the distribution while some other guild bugs are fixed.

This has been coming up a lot recently, so I’ll post the rules all at once:

Red player 1 disconnects for 2 minutes and gets dishonor.
Red player 1 gets a desertion and loses pips no matter what the outcome of the match is.

If red team wins:
Red player 2-5 get a victory and pips.

If red team loses:
Red player 2-5 get a forfeit and do not lose pips. This still shows as a defeat in the UI.
Except:
Any red players (2-5) in the same party as red player 1 get a defeat and still loses pips. This prevents players from disconnecting to save their friends from losing pips.

The reason people still lost pips when teammates disconnected during the end of season 1 was because of a bug in desertion detection. This should not be the case for season 2 and will work as described above.

This counts as a forfeit and not a loss, though it is lumped together with defeats for the statistics UI.

Soul.9280:

Evan, the last night of League play, I saw lots of people swapping classes during the 90 seconds wait period before the match started, and I was playing in the Legendary division against many of the top players. And I did see a surprising number of people who queued as thief or warrior and then switched to something meta.

I fully expect higher-skilled players to swap more, but that is a much smaller portion of the player-base.

An important question i: Would all high-skilled players be OK with profession locking if we could ensure no stacking?

It is correct that allowing profession swapping works against matchmaking. While I don’t have hard metrics readily available, my gut tells me a relatively small percentage of the player base does this. I don’t think it would be worth while to rip out the profession scoring entirely, though we did lower its effectiveness in relation to other scoring factors.

Locking professions upon queuing gets us a couple things: Profession stacking can be completely controlled via the matchmaker, and we can start using profession MMR again which should make much more accurate matches than just using account-wide MMR.

Galandil.9641:

Evan Lesh.3295:

The change wouldn’t have actually made class stacking better unfortunately. What was fixed was it breaking other matchmaking rules.

Can you elaborate on this point?

Before and now we get these same class-stacking results, so exactly what was changed and why?

Sure! To get in the right mindset, it helps to understand the goal of the matchmaker a bit. When comparing two rosters, we first assume they are a perfect match with a score of ‘0’. We then reduce this score for things like how far apart the ratings are, difference in party sizes, etc. This way we can decide if one roster is a better choice than another when filling up teams because one roster will have a higher score.

Before and after the change, we reduce the score of a roster if it contains profession duplicates over the desired(configurable) maximum for a team. However before the change, we were also increasing the score of a roster if it had professions that were not yet on the team. They were both trying to accomplishing the same goal.

The are three problems with the second part; the piece we removed. First, increasing the score of a roster effectively cancels out some of the negative scoring we added for other parameters. For example, a roster could have a poor rating match, but ‘oh well, they have unique professions!’ Second, the score was erroneously given to the first roster added to a team, so each profession was guaranteed to be considered unique and result in a score increase. Thirdly, the matchmaker has no choice in picking unique professions when dealing with large premades, but their rosters were still getting huge score increases because most of the time it had 5 unique professions. All combined, this meant premade rosters were getting a score increase that cancelled out most if not all negative aspects of the match up.

After this change we’ve seen the win rates of larger premades drop a noticeable amount. Now that rating differences aren’t being lost in scoring due to profession bonuses, we can see that the party rating boost was too high (which we’ve now lowered).

I’d like to say that this change was a direct result of a screenshot posted to the forums. While investigating the details of the match, I saw that a premade had a wildly abnormal score in relation to the other rosters. So thank you for the feedback!

The change wouldn’t have actually made class stacking better unfortunately. What was fixed was it breaking other matchmaking rules.

Also keep in mind that players who have not played in unranked for the duration of season 1 will have more volatile ratings. This will result in some volatility at the beginning of the season break.

This is a followup to the matchmaking changes in the recent patch where we fixed a problem regarding parties getting an unintended ‘slack’ in their matchmaking parameters. Previously, parties were getting matched against a wider range of opponents. Now that parties are getting much more accurate opponents, we can see that our rating boost for pre-mades is too high and the smaller roster sizes are winning too often. We are lowering the party rating boost a small amount to account for this.

Thanks for the continued feedback on matchmaking. Your posts lead directly to investigating gaps in our data which in turn lead to improvements of the system.

reapex.8546:

Evan Lesh.3295:

Forfeits are reenabled. Losing a game in which a teammate received dishonor will result in a forfeit. Players partied with the dishonorable player receive a desertion on loss, but they will not get dishonor.

Could you or anyone explain this part to me? It sounds like, if we’re in a match with a player that gets dishonor “during the match” and we lose…that player will get “desertion” but they will not get dishonor. I know dishonor stops people from entering a match but what does “desertion” do to the player?

In this case the desertion flag just means you’ll get a defeat on your record instead of a forfeit which prevents league progression loss. This prevents party members from disconnecting to save their friends from losing points.

Teutos.8620:

Evan Lesh.3295:

Now, two players in unranked that both have 500 unranked games played will score better than a player with 500 and a player with 200.

Can you please elaborate, why you taint your match making algorithm with other variables instead of only using your glicko2-based mmr?

From an outside perspective, it looks like you don’t trust your mmr, that it will provide us with good match ups. What else could be the reason to include games played?

Why should a player with 200 matches not play vs one with 2.000 when the MMR system rates them equally?

It keeps new players without an accurate rating from playing with experienced players who have similar, but accurate ratings. We could use rating deviation as a scoring parameter for this purpose, but it not as fine-grained as games played. This effect has a configurable limit so that the difference in games played don’t matter past a certain point.

Salamander.2504:

Also, for the first point “Changed matchmaking to use games played per arena instead of rank. For example, matchmaking in Unranked Arena will consider total Unranked games played.”

I’m a little confused; I thought the total number of games was already used in the MMR?

Games-played was not including in matchmaking scoring before. Rank was similar, but could have been gained in varying PvP environments (like custom arenas) and has a cap to it. Now, two players in unranked that both have 500 unranked games played will score better than a player with 500 and a player with 200.

I’m happy all these changes are out now! I’m looking forward to watching all the graphs and charts and player perception change :P

Hellbound.5610:

I would still really like to know in which way the website had inaccurate reflections…?

The website was using the beta ladder numbers, didn’t get reset with the season, wasn’t using the same formula as the actual season, and was likely bugged as we stopped maintaining the beta algorithm.

I verified your MMR updated every match. There were no configuration changes to how MMR behaves last week. The only thing that was changing was the acceptable ladder division search range.

Your MMR has updated every match. Can you please elaborate?

Kevin.4209:

Evan Lesh.3295:

The changes for season 2 will make this less of a problem. Each team will have a tight MMR grouping such that each teammate should be very similar in skill to you. This is different from season 1 and unranked where we balance players on each team, i.e, high MMR and both sides and low MMR on both sides.

The goal is to ensure than if the higher skilled team wins, all the higher MMR players go up the ladder. Less unlucky high MMR players will drop on the ladder, and fewer low MMR players will get carried up the ladder.

Am I right that the intention of this system is to enable high MMR players to climb very very quickly? I’m just trying to understand why this would be better than a reset – I’m guessing because given a reset, it will still take a long time for ESL players to climb up just because of the element of randomness that is involved. The downside of the system you are proposing is that during the first couple weeks, there will be hugely one sided matches (I’m guessing you guys see this as a necessary evil – but perhaps it should be publicized before you get huge numbers of people whining) – however, after that the high mmr players will be in high divisions and it won’t be as much of an issue.

That is one goal of the system. With an MMR reset, it would only create instability (ability to climb) for a short while. Part of having the ladder better represent skill is that the beginning of the season will be chaotic as everyone fights out of amber.

Avectius.3271:

Evan Lesh.3295:

Sunshine.5014:

Evan Lesh.3295:

The changes for season 2 will make this less of a problem. Each team will have a tight MMR grouping such that each teammate should be very similar in skill to you. This is different from season 1 and unranked where we balance players on each team, i.e, high MMR and both sides and low MMR on both sides.

The goal is to ensure than if the higher skilled team wins, all the higher MMR players go up the ladder. Less unlucky high MMR players will drop on the ladder, and fewer low MMR players will get carried up the ladder.

So if you’re unlucky and got matched with a low MMR team, you will keep losing and have no way to recover?

You wouldn’t be matched with a low MMR team unless you also have low MMR.

But what about the opposing team? If the system’s going to choose team mates as skilled as us, then what about the enemy players? What if we get paired up against enemy players with a much higher MMR?

Then that would mean either you climbed to high in divisions and you need to drop some, or the other team isn’t high enough yet and they are climbing out.

Xiahou Mao.9701:

I’m assuming people think it happened because the website’s PvP leaderboards are no longer available.

That could be. The website wasn’t accurate and shouldn’t have been visible anyways.

Sunshine.5014:

Evan Lesh.3295:

The changes for season 2 will make this less of a problem. Each team will have a tight MMR grouping such that each teammate should be very similar in skill to you. This is different from season 1 and unranked where we balance players on each team, i.e, high MMR and both sides and low MMR on both sides.

The goal is to ensure than if the higher skilled team wins, all the higher MMR players go up the ladder. Less unlucky high MMR players will drop on the ladder, and fewer low MMR players will get carried up the ladder.

So if you’re unlucky and got matched with a low MMR team, you will keep losing and have no way to recover?

You wouldn’t be matched with a low MMR team unless you also have low MMR.

gacob.1759:

Why not a pause system for the load screen?

I think everyone agrees on how this should be implemented. Either queue is paused during load screens, or you’re kicked from queue on load screens. I agree with this.

What I was mentioning earlier is that there are under-appreciated technical hurdles to implementing this that make it non-trivial.

The changes for season 2 will make this less of a problem. Each team will have a tight MMR grouping such that each teammate should be very similar in skill to you. This is different from season 1 and unranked where we balance players on each team, i.e, high MMR and both sides and low MMR on both sides.

The goal is to ensure than if the higher skilled team wins, all the higher MMR players go up the ladder. Less unlucky high MMR players will drop on the ladder, and fewer low MMR players will get carried up the ladder.