Cal Cohen.3527
-
Posts
89 -
Joined
-
Last visited
Content Type
Forums
Events
Articles
Gallery
Posts posted by Cal Cohen.3527
-
-
yes
-
> @"Kahrgan.7401" said:
I took a look at this match and interestingly enough the team with no duos won 503-498, which sounds like a great match. There are going to be some combinations of ratings where it makes sense for multiple duos/trios to be on the same team. For example, if the 3 highest rated players queue together and 2 lowest players queue together (with the other 5 all solo players), it probably makes sense for those 5 players to be on the same team. We could choose to always separate multi-man rosters when possible, but it would only result in worse matches.
-
We have a fix for this internally that will ship with a hotfix today
-
> @"Malediktus.9250" said:
> Just lost another close game because ranger longbow made me stuck in wall and I had to waste 60 seconds doing nothing again.
Where did this happen?
-
I'll try to respond to some of the questions and concerns here.
This change did not affect queue times at all. The matchmaker still picks the players for the match in the same way, but there's now a second layer that re-evaluates the match to make sure the most even combination of players is chosen for each team.
Matchmaking isn't an absolute science. Given a set of matches that have a perfect rating distribution (say all 10 players are 1500 rating), they won't all have the same result. Some matches could be blowouts, some could be 500-499, and some will be in between. There are a ton of factors that play into the result of a match, many of which are not things the matcher has control over. Players are human, which means that they're inconsistent. A 1500 rated player doesn't always play at a 1500 level. Sometimes they're playing out of their mind and spike up to an 1800 level, or sometimes they're on tilt and drop to a 1200 level. These numbers are pretty arbitrary but I think it gets the idea across. Some players are better at certain aspects of the game and weaker at others. A match where they only encounter their strengths will allow them to perform better than their actual rating while a match where they run into their weaknesses will cause them to play worse. As others have mentioned some team compositions are stronger than others and can tip an otherwise balanced match in favor of one team.
Essentially our goal is to make the best matches possible based on rating, which should leave the outcome of the match up to the team that plays better. We want to avoid cases where a team is clearly favored and make the distributions as even as possible, and the new matcher is doing this pretty well as the data shows.
Unranked matches will always have some additional volatility as players are more likely to be trying out new builds or professions that they're unfamiliar with, and with larger roster sizes making it more difficult to make balanced teams. I'm pretty excited to see what impact the new matcher has on ranked next season as it should be operating under more ideal conditions, especially at higher rating.
-
We collected some data from matches with the new matchmaker. I also grabbed the match data from the same chunk of time last week (under the old matchmaker) to compare to. Below are a few comparisons that I thought were interesting. I also did the same comparisons on the subset of data in which at least one of the teams had an average rating of 1500 or above. All of the data is from unranked matches. The data looks good, and we'll likely be turning the new matchmaker on in ranked sometime next week and for the seasons moving forward. We'll continue to keep an eye on things but I thought I'd share some of the data for those who find it interesting.
**Overall**
Old
* Average rating difference between teams: 37.29
* Average standard deviation difference between teams: 35.04
* Percent of games with average rating difference over 50: 24.5%
New
* Average rating difference between teams: 18.02
* Average standard deviation difference between teams: 23.9
* Percent of games with average rating difference over 50: 4.7%
**Matches where at least one team has average rating >=1500**
Old
* Average rating difference between teams: 56.87
* Average standard deviation difference between teams: 37.72
* Percent of games with average rating difference over 50: 41.3%
New
* Average rating difference between teams: 23.67
* Average standard deviation difference between teams: 23.28
* Percent of games with average rating difference over 50: 8.8%
-
Can confirm 120. The games threshold increases each week on tuesday but the season ends on monday.
-
> @"Anklot.3274" said:
> Can you guys test it a little more before implementing into ranked?
We’ll definitely be turning it back on in unranked to collect more data after we get the bug sorted out
-
This month didnt work out too great as I’m currently hanging out with my family for the holidays. I should be able to throw something together for january though so stay tuned
-
Hey everyone. We disabled the new matchmaker last night on the suspicion that it was causing some matches to not start up correctly. On a positive note, I was looking at some match data before leaving the office and it looked like there were improvements with the new matcher in terms of both average rating difference between the teams as well as match score differential. As soon as we get the bug resolved we'll take a closer look at the data and decide if we want to activate the new matchmaker for ranked next season.
Thanks
-
Hey guys. We've made an adjustment that should (hopefully) fix this issue that you've all been running into. If anyone is still experiencing this problem please post in this thread again and we'll continue to investigate.
Thanks!
-
Hey guys. We've made an adjustment that should (hopefully) fix this issue that you've all been running into. If anyone is still experiencing this problem please post in this thread again and we'll continue to investigate.
Thanks!
-
I can see some unranked matches starting up fine.
Are you queuing solo or in a party, and how large of a party if so?
-
> @"Rufo.3716" said:
> Keep getting Q pops, everybody accepts, then it just sits there, doesn't go to map selection screen just sits there until it puts you back in the Q. Rinse and repeat, can't get a game to start.
1) Is this is ranked or unranked?
2) How do you know everyone accepted? Are all 10 boxes checked before it kicks you back to queue?
Thanks for the info
-
> @"Toron.4856" said:
> 21 games 18 wins so far in unranked. your new mmr is working for me, thx :)
>
>
> also, in the past 15 games enemies didnt score 300 points a single time lol
I took a look at the account in the picture and most of the games were played before we shipped today's fix, so the old matcher was still active. Certainly matchmaking will be slightly less accurate with a fresh account of an experienced player as it tries to determine your true rating again, but it should continue to improve as you play more matches.
-
We've identified what's causing the problem and are looking into a fix now.
-
-
> @Saiyan.1704 said:
> Please please please fix the in-game map icons. It's annoying realizing it was a 2v1 on point because of a class icon covering another.
>
> Last I heard, Anet wanted to fix this issue as well. That was 2 years ago.
This is definitely annoying. I'll add it to Ben's list of things to follow up on with programmers.
-
> @"Honest John.4673" said:
> For the Timeline feature at the end of a match I'd like to be able to zoom in to a more granular level and hover over to see who died and when. With it being so small (it doesn't auto-resize, it's always based off 15 mins) clumps of deaths and caps/objectives sprout up and it's difficult to make sense of what exactly happened and where, negating the helpfulness of the feature.
Resizing to the actual length of the game and allowing a more targeted zoom-in are both on our list of timeline polish items.
-
> @brannigan.9831 said:
> Team based MMR averages for matchmaking is a terrible system. I know you responded to questions already about possible changes but it needs to be done. Let me give you an example and explain what happens with your system sometimes. Team One:1700, 1600, 1550, 1450, 1400 average MMR 1540. Team two 1600,1600,1550 1500, 1500 = MMR 1550. You matcher will say this is a great match. Im here to tell you that the later team will win easily the vast majority of the time because 1450 and 1400 MMR players will lose any fight where they are together unless they get extremely lucky. There is a weakest link element in this game that can't be overlooked. What ever team has the most good players even if the other team has the highest MMR player or two in the batch will win most of the time.
The data that we've been looking at suggests that larger advantages in average rating of a team result in a higher win percentage. The closer the average rating of each team, the closer to a 50/50 result. Obviously average rating is not an absolute measurement for matchmaking as you pointed out, but we can further tiebreak by comparing the standard deviation of the ratings on each team to look for a more even distribution of players. In your example here's a match with the same average ratings but much more even distributions:
Team One
1700
1550
1550
1500
1450
Team Two
1600
1600
1600
1500
1400
This is just one combination I came up with by eyeballing it; the matchmaker will consider all possible combinations of players and find the best one. So while it would consider your example a "good" match, it won't be the final selection as there are better ones.
One thing to keep in mind as the numbers get less pretty is that there will be buckets of average rating difference that are considered as "equal enough" to then compare standard deviations as a tiebreak. For example if we had one set of pairings that resulted in an average rating difference of 9.9 but had a huge discrepancy in standard deviations, this would be considered a worse match than one with an average rating difference of 10.1 and a much smaller deviation.
>Not to mention the idea of whether its fair or not to have people forced to group with people far below there skill level and I would argue anything over about 150 MMR difference is pretty big difference in skill level if we are talking about true established true MMRs.
As both Ben and I have said earlier in the thread, we're also going to be doing some testing on reducing the maximum rating range of the matchmaker and seeing what impact this has on queue times. I think it's fair to say that the range will be reduced by some amount, we just don't know by how much yet. We definitely agree that the smaller the range the better, but at some point the edges of the rating curve won't get matches in a reasonable amount of time, which is something we want to avoid.
-
> @Ario.8964 said:
> > @"Cal Cohen.3527" said:
> > > @Ario.8964 said:
> > > I guess my first question would be: Is there a way to tweak the current matchmaker to balance teams so that each individual player in a match is within 50 rating of the others? Cause the current one does some weird things to average out the mmr (Iasked a group I was playing with for ratings in one match just to check behavior and we had at the highest 1758 and at the lowest 1400 which I thought was incredibly bizarre).
> >
> > We've talked internally about reducing the maximum rating range of the matchmaker, but the tradeoff is that queue times will increase. We will likely do some testing in unranked to see how much of an impact certain ranges have, and look to make some adjustment for ranked in the future. 50 rating might be an unrealistic goal, but I do think there's some adjustment that can be made here without blowing out queue times.
> >
> > > Now as far as matchmaking goes on alt classes, is it a possible idea to try and create a system with class specific mmr (or at least a grace period of 10-15 matches where you are matched at lower and slowly increasing mmr until you are back where your actual mmr is) to encourage players to try out new classes? (this can go in unranked if needed but at least from my experience, trying new classes is incredibly frustrating because you get placed in matched where you'd want to play your main and end up getting farmed pretty hard by a good deal of people in the match.)
> >
> > > And not to be "that guy" and bring this up but is there a possibility of a class lock upon entering a match and removal of class stacking from all pvp modes? I just know myself and many others are curious as far as if there was any discussion around the implementation of such a system or if other things were a priority thus leading to this not really being considered yet as matchmaking would be seriously impacted by a change like this.
> >
> > These two ideas are tied together. We do actually track profession specific mmr, but it isn't currently used **because** we don't lock professions in queue. Without the character lock, players could queue on a profession with lower mmr, then swap to their main profession after the queue pops for easier wins. Some time ago we held a poll asking players if they would prefer using profession mmr and locking characters on queue, but it did not pass.
> >
>
> Yeah my first number was kind of a spitball/hopeful number but If population can't support it that could be widened to a certain degree if it works with the matchmaker.
>
> Correct me if I'm wrong, but I thought the original poll was only asking players about class locks in matches. The main reason many said no to it (myself included) would've been fear of having stacked classes such as thieves making it harder for your team or stacked scourges on the other team making your life harder. If the entirety of the idea were pitched it may have seen at the time, or likely would now see, different results.
This is likely the reason why the poll didn't pass. Players would rather have the option of improving their team composition during the warmup period. The matchmaker already makes some considerations to distribute professions evenly between teams, but it won't make a "bad" match for the sake of balancing out the professions. When these situations occur, I think it's a good thing for players who can to be able to swap professions, and this wouldn't be possible with profession locking. These same cases could still happen, but players wouldn't be able to do anything about it.
-
> @witcher.3197 said:
> **Special tournaments**
>
> How far can you push the boundaries on this one? GW2 has several excellent PvP minigames/ activities such as Snowball Mayhem and Keg Brawl that have so much potential. In my opinion these would be a lot of fun and a nice change of pace for organized teams, but even after 5 years we can't play them with friends.
>
> Would it be feasible to set up some Snowball Mayhem tournaments during the ~1 month Wintersday event?
We've definitely talked about setting up activity tournaments, but there's some work that needs to be done with the activities before they can be used in the tournament system. It's not a high priority at the moment, but it's possible somewhere down the line. Snowball Mayhem won't be tournament-ready for Wintersday this year.
-
> @bluri.2653 said:
> I was mainly thinking the monthly tournament since it's where the rewards are at
Streaming the monthly should be doable. Maybe we wouldn't have a fancy stream setup in the office but I can do it from my apartment. Just gotta convince Ben to stream the other region and we're set.
-
> @Salvo.8290 said:
> Swiss style tournaments sounds like a good idea. The one thing that might need to be adjusted if you move to swiss style is actually the rewards. If you have people spending significantly more time doing this event, as it sounds like would happen, you would have to bump up the rewards to compensate for this.
This is a good point, and we're discussing increasing the per-match rewards when we move to the swiss system.
>One thing you did not mention in your post was about observing ATs. I understand that partners can do this, but is there plans to allow for viewing for all players? If you could go to the tournament tab and watch replays of the previous tournament it would be a welcome addition also. Thank you for the addition of another daily tournament as some people can now get in one a day, and others get two.
I touched on this in a previous post, but we currently aren't planning to open tournament spectating to all players.
> One final question for monthly tournaments. This month the tournament is held on the 23rd, and I know it is scheduled for the 4th Saturday of every month. Many people will be with their family at that time. Is there going to be a way in the future to adjust this earlier/later in months around holidays so more people can participate (and adjust required QP if earlier)?
This is something that is always going to be a problem in December. We currently don't have any plans to change the schedule for the monthlies.
please qualifying match FIX NOW!!!!
in Player vs. Player
Posted
Disconnects are handled the same in placements as they are everywhere else. The placement UI still shows a loss because the game counts toward your game total, but rating, deviation, and volatility are all unchanged.