Been wondering if that was the case. It would explain the inconsistency in MM since I might perform better/worse on certain maps (due to hero pool etc.).
Also does MMR include whether I started on the Attacking or Defending Team for each map if correct?
Next question, how does the MM decides which Map I’m playing on next?
If my MMR is partially based on the map I’m playing on, maybe availability (due to balanced match ups)?
Your MMR (Matchmaking Rank) is just a single number. It’s basically equal to your SR, assuming you aren’t Diamond+ ranked and suffering from SR decay.
What map it is, what map type, whether you start attacking or defending, all those things are irrelevant. Which does, yes, mean that you’re expected to perform at the same level on Payload as on Control, even though you might actually be better at one or the other. You also might play heroes who are better suited to certain maps and perform better or worse depending on that, but that’s all the same to the Matchmaker.
Which map you play is mostly random, maybe completely random. It has been suggested that the matchmaker tries not to put you on the same map twice in a row, which would be reasonably easy to do (take a list of all maps, remove the last played map of each player from that list, pick randomly from the rest). It has also been suggested that the matchmaker tries to make sure you see all maps and gives a map priority if you haven’t seen it in a long time. I guess that’s not impossible, you could look for the map not seen for the longest time among all 12 players, and pick that one or pick randomly among any ties.
Anyway, matchmaking does not take into account really anything other than MMR, group sizes, and how long you’ve been in the queue. It tries to find 12 people with similar MMR and similar group sizes, but it gets less strict the longer you’re in the queue.
There’s all kinds of details collected here (on SR/MMR, not on map selection): How SR works (S16)
Why wouldn’t they incorporate that information into your MMR making it an array of numbers rather than a single number?
Each reflecting basically your win-rate yet on map/Att or Def scope.
In that case, the MM possibly wouldn’t allow you to climb at some point without proving that you are able to win on maps you perform worse. This could lead to the perception of “getting bad teammates on purpose” when in fact you perform worse in general in those specific constellations (due to the heroes you usually play).
Did some digging and found this in an old post
src.: Overwatch Forums
It could also mean that they simply have set win expectations for certain maps (mostly hybrid, escort) regardless on who is playing on them.
But wouldn’t it be better to incorporate that information into the players MMR in that case. If certain players perform better on certain maps due to the heroes they are able to and usually play (pure assumption that this is the case).
However, it might inflate queue times unreasonable high which would be a downside to it.
Because it would take much, much longer to place players accurately if you had an array of 17 ratings instead of 1 rating. 17 times longer. By map type rather than specific map would only be 4 times longer, of course. But do you want to play even 40 placement matches, never mind 170?
It also shouldn’t matter all that much which map you’re playing. I’m sure it’s a factor in the ‘bad teammates’ problem, but not nearly as much of a factor as say, a Soldier76 main playing Reinhardt. Or vice versa. MMR is naturally going to primarily reflect your skill on your most-played heroes and so playing an off-role is a huge issue.
That post is talking about quickplay MMR. Since you only attack or defend, it is important to consider which map you’re on and which side you have. And yes, they have per-map win expectations for the average team. In competitive, you have a symmetrical match so you don’t have different chances of winning or losing because you play both sides. I don’t think per-map MMR adjustements are used in competitive. Maybe as weighting factor so that maps that take longer on average are worth more, but I don’t think so.
QP explanation makes sense.
But that placement argument, meh. In the end, you’ll just increase your rank accuracy with every additional match. Offsetting the values over time, it’s basically improving over time and with each match you have on a specific map.
I don’t see how’d require more placements matches.
I don’t get that part, each of those values would still reflect your ability on a specific map in a specific situation yet it remains win based. About Q time, since it doesn’t reduce the overall availability in your play pool as you do for role Q, you probably wouldn’t increase the time that much.
It depends on how you layered it in the end.
If you have per-map MMR, then if you did 10 placement matches you wouldn’t even have data for 7 maps. If you do 17 placements, you have just 1 point of data for each map.
Edit: Is it 17 maps…? Hnnh. Well, I have to go. But anyway, it’s somewhere around that number, the point doesn’t really change much on that detail.
The number of placements is at 10 so that you have a good number of games to average to figure out your place. If you have per map SR, the correct number of placements might not be 10 per map, but it’s certainly some number that involves playing each map more than once.
Even if you cut off placements at 10, and just leave people at the default start of 2250SR on unplayed maps to let regular play sort it out, you’re going to have misplaced people for a long time. In the long run your per-map SSR would be more accurate, but that long run is 17 times longer than what it takes now for someone to reach their proper, stable rank.
It’s really just a tradeoff - do you want people to eventually be extremely accurately placed, or do you want people to quickly be roughly correctly placed?
Personally, I think per-map performance difference is vanishingly small compared to per-hero or per-role SR performance. I really don’t think it’s worth implementing a slow, complex system to account for it. I would guess that pretty much everyone would have all their per-map SR within a couple hundred point range anyway. 100 or so SR one way or the other just isn’t that big a difference to be worried about in terms of match quality.
Think of it more as a progressing adjustment to the current version.
Since right now (we assume) it’s not being used you basically adjust the same value for every map at the same time. With increasing exposure to each map, you’d be able to improve your predictions instead of continuously holding on to the same value.
That’s something you’d had to look into by analyzing the data.
Imagine players returning after a long break playing on maps they haven’t played before or players using a rather small portion of the hero pool.
Since certain heroes do better on certain maps it might have an impact I couldn’t possibly assume without having a great database for each match played.
A per hero SR is impossible to implement since you can’t tell which hero is someone going to play. That’s why Role SR improves your accuracy so drastically as it prevents players from picking heroes outside of that role and limit their options.
Elo ranking systems are heavily based on your ability to predict the outcome of a match, eliminating random variables thus allows for improved MM.
In the end, it probably boils down to queue times which is affected by how you’d implement it as mentioned above.