155 replies to this topic

[Horseman]

Curccu, on 08 April 2022 - 08:08 AM, said:

But if it stays 12v12 might be stupidly hard to get games as T1 as I understood from ASHs post few months back that there are very few T1 compared to other tiers.

Easier now that more players are well distributed. I'd rather wait for a few minutes longer for good quality matches than ones padded with players so terrible that I wonder how the **** did they figure out how to log into the game at all.

[Nightbird]

The problem of the MM doesn't have anything to do with the number of players in T1 and T2, it has to do with how it assigns the 24 players in a match to the two teams.

Closing the gate from +2 to +1 tiers wouldn't do anything if the best 12 players of the 24 are on one team.

[Knownswift]

Horseman, on 08 April 2022 - 06:56 AM, said:

The first step to reducing MM issues is clamping down the tier floodgates from +/- 2 to +/- 1 . We had it before, matchmaking was longer but the matches were much tighter as well.

A single tier differential made no difference in match quality at the higher end. The bonus was waiting 5-10 minutes for a match.

Thats a big F for any game, all around.

[Horseman]

Knownswift, on 09 April 2022 - 06:42 AM, said:

A single tier differential made no difference in match quality at the higher end. The bonus was waiting 5-10 minutes for a match.

In my subjective experience, the difference was very noticeable.

[Urbie Connoisseur]

Going on a slight tangent, meanwhile, I feel the skill tiers are too compressed. I've been in tier 5 since day 1, and been from between some distance before the text, to the bar being almost full, and I've had games where I've been pinpoint cored in seemingly an instant with absolutely zero damage in any other part... in assaults. That alone says that only 5 tiers is by far too compressed. There should probably be like 8 or 10 skill tiers with a spread of maybe 2 or 3 tiers.

[pbiggz]

Marida Connoisseur, on 09 April 2022 - 04:49 PM, said:

Going on a slight tangent, meanwhile, I feel the skill tiers are too compressed. I've been in tier 5 since day 1, and been from between some distance before the text, to the bar being almost full, and I've had games where I've been pinpoint cored in seemingly an instant with absolutely zero damage in any other part... in assaults. That alone says that only 5 tiers is by far too compressed. There should probably be like 8 or 10 skill tiers with a spread of maybe 2 or 3 tiers.

That is something that can be explored, however, as is the case with any matchmaker, more granularity doesn't necessarily change outcomes.

If you have 10 tiers instead of five, but your tolerance is +-4, then functionally its identical to the current +-2. If you made the tolerance +-2 in a 10 tier system that's functionally the same as +-1 in a 5 tier system, meaning you get all the same drawbacks, including top-tier players being effectively locked out of matches, and longer wait times across the board.

That's not to say more tiers wouldn't help. I expect it would, but you have to balance granularity and strictness with matchmaking speed. The more stringent you are with match requirements, the longer it will take to form matches. A balance has to be struck and that means concessions must be made. The question is where.

[Der Geisterbaer]

pbiggz, on 10 April 2022 - 09:22 AM, said:

If you have 10 tiers instead of five, but your tolerance is +-4, then functionally its identical to the current +-2.

Minor nitpick: With 10 tiers and ±4 it wouldn't actually be "functionally" identical to the current system because in a 10 tier system "the middle" is split between tiers 5 and 6 and tier 1 would thus only get matched against the upper half of "the middle" (in tier 6) whereas the current 5 tier system matches against the entire "middle" (lower and upper half) in tier 5. => (Almost) Functionally identical would be 9 tiers with ±4 but even then you'd most likely see some variance

Edited by Der Geisterbaer, 10 April 2022 - 09:36 AM.

[DaZur]

pbiggz, on 10 April 2022 - 09:22 AM, said:

That is something that can be explored, however, as is the case with any matchmaker, more granularity doesn't necessarily change outcomes.

If you have 10 tiers instead of five, but your tolerance is +-4, then functionally its identical to the current +-2. If you made the tolerance +-2 in a 10 tier system that's functionally the same as +-1 in a 5 tier system, meaning you get all the same drawbacks, including top-tier players being effectively locked out of matches, and longer wait times across the board.

That's not to say more tiers wouldn't help. I expect it would, but you have to balance granularity and strictness with matchmaking speed. The more stringent you are with match requirements, the longer it will take to form matches. A balance has to be struck and that means concessions must be made. The question is where.

11 tiers... 11 tiers with a +/- 3

This would allow for a softer transition from tier to tier and be an insular for strings of bad beats and and would be a defense against high and low tier leaks during low draw pools...

Edited by DaZur, 10 April 2022 - 09:37 AM.

[pbiggz]

Der Geisterbaer, on 10 April 2022 - 09:34 AM, said:

Minor nitpick: With 10 tiers and ±4 it wouldn't actually be "functionally" identical to the current system because in a 10 tier system "the middle" is split between tiers 5 and 6 and tier 1 would thus only get matched against the upper half of "the middle" (in tier 6) whereas the current 5 tier system matches against the entire "middle" (lower and upper half) in tier 5. => (Almost) Functionally identical would be 9 tiers with ±4 but even then you'd most likely see some variance

Im not a rocket surgeon so, yes, this is correct i think.

DaZur, on 10 April 2022 - 09:34 AM, said:

11 tiers... 11 tiers with a +/- 3

This would allow for a softer transition from tier to tier and be an insular for strings of bad beats and and would be a defense against high and low tier leaks during low draw pools...

It could. I definitely would hope that if PGI was exploring the matchmaker again, they'd explore these solutions, but, I just don't want anyone thinking there's a silver bullet that will solve all the problems. Matchmakers are best-effort machines, they aren't that precise.

[Nightbird]

DaZur, on 10 April 2022 - 09:34 AM, said:

11 tiers... 11 tiers with a +/- 3

This would allow for a softer transition from tier to tier and be an insular for strings of bad beats and and would be a defense against high and low tier leaks during low draw pools...

More Tiers wouldn't help because Match score, which is used to move up or down in tiers/PSR, is not a good indicator of skill.

Other than throwing away Match Score, PSR, and Tiers altogether, and using WLR instead, there's no simple way to fix the match maker.

[DaZur]

Nightbird, on 10 April 2022 - 12:06 PM, said:

More Tiers wouldn't help because Match score, which is used to move up or down in tiers/PSR, is not a good indicator of skill.

Other than throwing away Match Score, PSR, and Tiers altogether, and using WLR instead, there's no simple way to fix the match maker.

It doesn't "fix" anything per say... What is would do is lessen the transitional impact of moving up or down in the tier ladder as well as further filter the high and low bleed through during non peek play times when the draw pool is lowest.

I agree that the whole idea of determining an "individuals" potential skill value from a composite derived from a team outcome should have their head examined. That said, it's what we have to work with... I don't think PGI is collectively prepared or staffed to scrap the existing system and start fresh.

That said, I think it's high time to drop the vitriol over the existing system and rather spend a little more time focusing on how small changes in the existing system can be done to improve what we have without reinventing the wheel...

Edited by DaZur, 10 April 2022 - 09:16 PM.

[pbiggz]

Nightbird, on 10 April 2022 - 12:06 PM, said:

More Tiers wouldn't help because Match score, which is used to move up or down in tiers/PSR, is not a good indicator of skill.

Ok, what is? At some point you're going too have to create some abstraction for the matchmaker to work with. Do you advocate for use of WLR as a metric for skill instead? Because WLR itself is an abstraction because you could just get placed on good teams.

In the current system more tiers, done carefully, could help. I don't see what you contribute to this conversation by simply taking a dump on every single thought anyone has because it isn't sufficiently critical of the existing matchmaker.

Nightbird, on 10 April 2022 - 12:06 PM, said:

Other than throwing away Match Score, PSR, and Tiers altogether, and using WLR instead, there's no simple way to fix the match maker.

That's true, but given the confidence and bluster with which you talk about how bad the current matchmaker is, and given you don't actually know how it works because nobody outside PGI gets to know that, I'm not convinced anyone should take you seriously on it at all.


We're having a conversation about how we can piece together something better out of the tools we have available. We are not having a conversation about how PGI sucks. That is certainly a conversation that can be had. I certainly have my opinions about their many missteps. We can have it in a different thread. Contribute something other than vitriol to the conversation.

[Nightbird]

pbiggz, on 10 April 2022 - 07:11 PM, said:

you don't actually know how it works because nobody outside PGI gets to know that, I'm not convinced anyone should take you seriously on it at all.

I knew well enough about how the MM works that I was able to predict how the last matchmaker update was going to perform before it even released - terribly - so I know more than PGI on how the MM works. At least, I hope they didn't know, most people don't intentionally point a gun at their foot and pull the trigger.

https://mwomercs.com...cy-with-graphs/

https://mwomercs.com...th-in-business/

Edited by Nightbird, 10 April 2022 - 08:25 PM.

[pbiggz]

Nightbird, on 10 April 2022 - 08:24 PM, said:

I knew well enough about how the MM works that I was able to predict how the last matchmaker update was going to perform before it even released - terribly - so I know more than PGI on how the MM works. At least, I hope they didn't know, most people don't intentionally point a gun at their foot and pull the trigger.

https://mwomercs.com...cy-with-graphs/

https://mwomercs.com...th-in-business/

Have you seen the source code?

If not, then you don't know better about how the matchmaker works.

[Nightbird]

pbiggz, on 11 April 2022 - 05:56 AM, said:

Have you seen the source code?

If not, then you don't know better about how the matchmaker works.

What's more important, knowing that the MM does or exactly how it works? PGI might know the code, but they don't know what that code does.

It's the difference between knowing where the brakes are on a train, and knowing the bridge is out. PGI only knows the former.

[pbiggz]

I think knowing exactly how the matchmaker works is important when you sell yourself on knowing why its bad.

[Nightbird]

pbiggz, on 11 April 2022 - 09:32 AM, said:

I think knowing exactly how the matchmaker works is important when you sell yourself on knowing why its bad.

I disagree, knowing the results is more important than the process. That is what is more important for a company at a high level.

Keep the ship from sinking is more important than having engineers figure out why it sunk.

[TheCaptainJZ]

The MM has adjustable valves. As far as I know, no one has touched them in a long time. Since that's already built into the system, PGI should be exploring those options for tweaking the MM first, but they need a good way to assess the change and whether it's better or worse.

[Nightbird]

TheCaptainJZ, on 11 April 2022 - 10:21 AM, said:

The MM has adjustable valves. As far as I know, no one has touched them in a long time. Since that's already built into the system, PGI should be exploring those options for tweaking the MM first, but they need a good way to assess the change and whether it's better or worse.

Those valves only allow more or less tiers into the same lobby for a match or limit the tonnage difference between teams. None of the valves balance the skill of players on either team, because the PSR doesn't provide that information accurately. If PSR is wrong, match making will be bad regardless because it's a fundamental metric.

[SFC174]

Ultimately WLR is the only way that we can fix the MM given the limited resources. Yes, it'll take 100+ games for you to get sorted properly, but over a large enough sample size WLR is the best predictor of future success.

The current MM is broken. It was bad before the queue merge and it got worse afterward. I used to be in favor of a hybrid WLR/matchscore system, but Nightbird has convinced me over the years with the math and with some very accurate predictive modeling of the current matchmaker.

This won't make the middle tiers dramatically better since you'll still be mixing high and low skill players due to population, but at least you'll have the same amount of each skill on either side of the equation.