Talk about the matchmaking before we talk about balance

General Discussion
42

That’s not how the SRD system works.

In the SRD system lobbies are formed naturally from a narrow SR range. The range is adaptive on various criteria (usually a neighbourhood size that depends on active pop or wait times, and place on SR curve). So you take your chances against a sample for that SR range (rank). The lobby odds are not rigged in advance: you could have 99-1 chance or a 1-99 one. Over time you would avg your samples out to 50-50, but given that player agency isn’t rigged away, you can more quickly discriminate yourself over the label margin. That means gaining SR through direct wins until you converge to proper final rank where you finally tighten up on this 50% winrate (but still have the streaks and stomps because of the unforced odds).

SRD takes your current SR and pulls you to your true SR with a gradient that is based on your ability to win vs. the rank, not a rigged lobby. It’s therefore actual laddering and not a carnival scam.

1 Like
43

Matching people with others nearby them in SR is pretty much what the current system does.

Making a lobby out of smaller ranges was an optimisation for the poor matchmakers, so they didn’t have to handle as many people in one go.

I’m not sure it is a better way of going matchmaking, since… You end up with a bunch of breakpoints where the game is really easy (when you are at the top of a range) and REALLY hard (when you are at the bottom of the new one).

It would be like the curve people first hit when they get on the silver / gold border, but over and over again.

I mean, from a “at least our matchmaker doesn’t have to do a lot of work” point of view there is a lot of goodness there.

44

off topic, but I think I’ve queued with you before. We almost have the same overall character exp, and I think I’ve seen your name come up a few times before in recent memory. Perhaps we are in the same elo.

EDIT: nevermind we are not close on the overall exp. I’m just too tired. evidenced by everything I mentioned in this entire post. Still queueing.

another edit:

a quote from my other post

"one last thing I want to say on the subject is:

another big reason toxicity exists and overall players just being dissatisfied with the game is the fact that my example is one of many, and the potential for something like this to happen is another reason why everyone is deathly afraid of getting even one loss. One simple loss can turn into a catastrophic turn of events just like this. My one loss should have quickly recovered. However, it didn’t. Instead of taking it out on players, I am venting everywhere else and in every other kind of way. I’m also channeling this energy into playing even harder than I have before with much better personal gameplay results in each match. If they would eliminate this kind of stuff from happening, players won’t be so frustrated each time they get a loss, and they won’t be so hard on other players for their shortcomings."

45

I disagree. It differs functionally but more importantly is a different philosophy.

First off, you have all the gaming hate towards SBMM because the effects are extremely noticeable (and when they’re not, they’re blamable which is a minus for your brand). There is also the ethical debate about agency in sports, natural competition, and hopefully legal stance that says you can’t rig the prize machinery or contest odds. Finally there are all the math reasons like being more robust to noise better breakdown and ofc explainability since no data-invasive entirely spoofable classifiers are used.

To me the biggest difference is that MMR rigs the lobby (stochastic) outcomes instead of letting the metric (SR) be self-concordant. When you over-tune locally you lose tuning globally (NFL theorem). SRD constantly maps players onto itself, MMR does not. So with SRD you overall get a ladder progression (SR labelling) that better matches up with the lobby difficulty.

Correct. If the ranges are static and the population is divided into pools (rank ranges) then with enough wins or sr gain you suddenly “jump” up into matches where the centroid is the middle of that pool. For small enough ranges this wouldn’t be an issue until you’re nearing that final deserved rank.

Classical ranking systems and tournament sorting algs solved this with adaptive neighbourhood distance. So it basically pulls k=12 nearest like a clustering alg but instead of handicapping with knapsack balance you just shoot direct and take those odds. Sometimes the top6 and bot6 of that select 12 are on either team and it’s a stomp, but that shouldn’t happen often (6 heads vs. 6 tails odds).

This leads to a much faster cycle time (queue reduction) and gets people to where they belong without data (privacy) and without rigging (fairness). The percolation rate (transition from current to destination rank) is robust to noise for and against (any corruption provably balances out without mmr vs. sr lock-in and achor-stat effects viz. no-reset epochs). When you simulate it out it’s a random walk biased by skill gradient over the rank and hindered by noise (but everyone is hindered equally no discrim to “likeable” types).

My $0.02 for now looks like my 18minute queue just popped.

1 Like
46

Mmr = people with the same sr, some permutations of the teams would be thrown out due to hidden factors that the matchmaking thinks <40 >60.

Mean while lobbies of the same sr all combinations should be playable. Right? If not then SR has a different meaning than what the literal words mean.

I say/ask this so far no1 rebukes it.

3 Likes
47

liked your comment. it helps to have a refreshing review of the overall perspective

1 Like
48

IDK about comp, but QP def has an egregious matchmaking issue. It gets really bad when the server population count seems to dwindle during off hours. What I don’t understand is as to why devs have such a hard time mixing and matching so much so that I end up in matches in which I have to hard carry doing 30+% of the team’s damage and lose. Multiple times in a row. My guess is that the average IQ of the guys working on QP mm is in the double digits. The mm is so broken that those with relatively high stats are essentially placed in unwinnable games during off-hours. I can’t imagine many of these guys have a proper statistics education.

If this is their approach of “anti-smurfing,” I mean, I am lost for words. Do they even have a proper data scientist on their team? My guess is no, because Blizzard has been historically a very stingy company (just look at former employees recounting how they had been paid so little that they had to room-share to survive in Irvine/Anaheim), and I doubt they can afford to keep one on their team.

2 Likes
50

Ah, yes… This user was crying about match making on the Competitive forums not long ago.

As you once requested/advised me (not long ago), don’t feed the bridge troll.

51

you’re too stupid to see that you are the bridge troll, and intentionally, brazenly, and obviously to everyone reading… off base. Stay ignorant

1 Like
52

Coming from a player blaming a software utility for their inability to be successful in a video game?

Go off Champ. You deserve a participation trophy.

53

Small enough it isn’t a problem and it is small enough that blow out your queue times to eye watering lengths.

This leads to a much faster cycle time (queue reduction)

It really doesn’t. If you don’t have a tank in your “SR range which you have queued” you get no game.

In the current system you get a game, because you get tanks outside of your queue range.

Not really (I think you are making an argument against rolequeue here)

Say you have 6 mercy only players, they are all good at what they do, but as a combination it won’t work for you.

You could make a game where all combo were equally playable, by making all the heroes practically identical, but OW isn’t that kind of game.

That would be fair, except I don’t think they are trolling.

because of this…

Which is not someone trying to troll, but an legit call for help.

Oh, RobotWizard is one of the names given by the system if you don’t put in a name, so there are literally 1000s of us. like Sneakypants, etc.

You will see the same names show up over and over again, they are ones where the system as allocated you a name, rather then you have chosen it.

I mean, if you are out in the OCE, maybe we have played together.

1 Like
54

Missing the point. In the current MMR system you ALSO have to wait for the rigging.

In SRD system the “SR range” is adaptive based on location on pdf and active pop (adjusts for sparsity). Again, with MMR you have this additional layer of constraint-satisfaction w.r.t. rigging the outcomes. With SRD you simply random 12 neighbours and ship. This is best-possible in terms of queues. And when you adjust the SR for those results you are self-conforming the SR metric.

2 Likes
55

I think this needs to be a new thread, can you make one?

56

there are 10s of threads on why mmr is just plain anti-math
and until i see some layoffs - updated linkedin profiles blizzard entertainment xyz-2022 gg, then there isn’t much point. they’re going ahead with the same trash we’ve had for years.

2 Likes
57

How long.

I’ve been waiting 6 years for a rigged match.

58

your comment is like someone saying “I’ve been waiting to see economic problems.” “I’ve been waiting to see racism anywhere” “I’ve been waiting to see …” fill in the blank. How can we converse with these kinds of replies?

We can also get the opinions of all of the people who don’t play anymore, and in my case, are baffled that anyone even wants to play this game. Oh, and the people who played this game for maybe one day and said “no”. I think we need to come to understand that there are real problems that lie here beyond just simple differences of opinions and preferences of gamers, or that those that disagree with the forum dwellers here are all just simply “too stupid to understand truth and superiority of all of those who inhabit the sanctity of these forums”.

3 Likes
59

*with an assumption that whoever the player picks has a value monotonically increasing function that takes the players quality of inputs as an argument and that function’s derivative (with respects to quality of player’s inputs) isn’t too close to 0 for large portions of its domain where most people’s skill may lie.

:nerd_face:

1 Like
60

Yeah, ok, Mercy not withstanding :wink:

1 Like
61

This makes sense. I played Titanfall 2 really really late and all that’s left are people with like thousands of hours one tapping you with a kraber from god knows where while zooming across the map in seconds.
Suffice to say I never played it again.

1 Like
62

Well that’s not the spirit. Is it not possible to factor this in when matching people up? Red flags could be new accounts with crazy stats (or have a yellow flag of new account so place them together with other newer accounts) or sudden SR drops, wildly variant accuracy/win loss ratios/etc :man_shrugging:t2: