Matchmaker Pool Feedback Thread

Wasn't very productive I'm afraid. Instead of gaining an appreciation for your position I feel sickened by your entitlement instead. Maybe I'll have to reread your arguments when I'm feeling more charitable, but the long wait times and gatekeeping still seem like too high costs to me.

With the system I proposed, 2 very high or low rated players could maybe find a 4v4 match if 12 other players were looking, whereas what you propose might require the player pool to reach into the 50s before they get a match. That cuts the time they could use tmm from a few hours a day to a few hours each weekend.

And this would affect all players queueing, regardless how picky they were about the balance they got

Just watch the goddamn replay.

@xiaomao I will once I get home.

I watched the replay. In case I missed the point you were trying to make, here's what I noticed: the lane balance was really weird and could have been better (opti fail? dunno) . Ruler (1000 4v4 rating) made a fatboy on beach, that's weird, right? Surely I missed a lot of things.

Since FTX never bothered to give more details I'll just give my thoughts on both possible scenarios:

If this was a tmm game, opti needs to be more like autobalance at the very least. I think if the lane balance was different the game might have been better. I'm not sure why opti decided 1200 vs 1800 on air was good, or why 1000 vs 2300 in north pond was good, or any other matchup, in fact. Not as good as if everyone had the same rating, obviously. That would be ideal. I just think it's too much to wish for when the player pool is so small and enforcing that too strictly could lead to players in the tails of the rating distribution finding way fewer games, especially 3v3 / 4v4. If this is one of Raider's first tmm games and the rating system is using his global rating as a temporary stand-in for his 4v4 rating, it's understandable why the matchups were so garbage, but that seems like an edge case on top of another edge case: first game of an abuser. Doesn't really scream "typical tmm experience" to me and even with ftx's proposed change such edge cases will exist.

If this was a custom game, why exactly should it not be rated? Everyone participating decided it was a fair matchup by joining and x'ing. The quality was 89%. The ratings are definitely not representative of skill, and 3300 is ridiculous. Ruler's rating is in fact around 1000 on all ladders except global, where they got that high by spamming astro exclusively. The solution is to make ladder/tmm more popular than custom games and bury global rating in a deep UI hole so that players actually have to get good to maximize their status signifier which should be their league badges, not global rating. UI changes to the client and game are needed for that, at the very least, but anything that can increase the popularity of ladder/tmm for the average casual player will help achieve this goal. Banning "all welcome" might help with cases such as Ruler but is throwing out too much baby with the bathwater and the league solution is better and at least 60% of the way there in terms of implementation, meaning most of the back-end stuff is done and some of the front-end as well.

It's not even about the 3.3k dude. The game is over at minute 10 cuz the low rated dude got put on carry slot vs someone competent. And there's nothing you can do about that.

@xiaomao is the carry slot not rock? isn't that the 3300's slot? I'm confused. Anyway, opti balance could use some improvement since autobalance proves the task can be done better. Why not use that for tmm instead? As I said, ideally every game has every player of equal rating and the balancer is moot, but the more restrictions, the more players need to queue before games can be found. It would be great if someday it could serve us all perfectly balanced games with little hassle but the priority should be growing it's popularity because that's the only way it can be achieved. Until then you can host custom and enforce rating restrictions with not that much hassle.

None of it explains why custom "all welcome" lobbies need to be unrated.

There was literally 10 minute strat bomber that got ctrlk'ed to just not end the game at that point...

Also, it's a perfect example of game that should of have never been rated.

Looked to me like the better team won in that replay ftx posted, but it also looked like the game never should have had 89% quality as everyone on south team was doing significantly worse than their counterparts and their 4v4 ratings compared to north team don't justify 89%. In that sense I'd say the rating system worked but a bit suboptimally, and I suspect some goofy abused global rating was substituted for 4v4 rating so there was bad data to start with. The rating system adjusted player ratings in the right direction. The lane balance was shit but that can be better and the code already exists, courtesy of Penguin. Why do you think it shouldn't have been rated @Xiaomao? I agree it's not a good match for the matchmaker to put together, probably happened because of bad input data, but the rating system responded correctly, more or less, and adjusted the ratings in the right directions.

#20443714 can you tell me why this should be unrated? I didn't cherry pick it, it's literally the first 3v3 tmm replay i stumbled upon with 2 high and 1 low rated member per team and it's a solution i think would work decently well. Why is this not good enough?

G-great. You picked a game with underrated Voodo stack, full on voice and playing same squad every fucking day be it 4v4 or 3v3.
The 500 rated player you see there is actually 1.1k in 4v4 with his Voodo stack.

You better watch this https://replay.faforever.com/19738422 and tell me this is fine.

Also, this is just sad:

so voodoo's "500" buddy should uhh, grind alone until his rating reflects his skill? that will surely fix the issue right ?

Yes. Just like Grimplex shouldn't be able to play with all these 100-500 rated players in TMM.

Watch the replay and say it straight to my face that it was a fine game of FAF. And that we should encourage more of such game to be had in a ranked enviroment.

will watch, currently ingame though. I just don't understand how underrated players grinding at their current rating fixes the issue. you still get bad matchups until the rating system zeroes in on their true skill and it's no faster if they do so with the same nr of players. My choice of replay might have confused the matter since this is a different issue of inaccurate ratings

@xiaomao i watched the replay. It's not fine. I agree a problem is there. Why are you fishing with this "tell me that's fine" bullshit when that's not what I'm arguing?

That doesn't automatically mean the scorched earth solution is good. assuming the player ratings are representative (since the opposite is just a red herring) my suggestion is to modify the matchmaker to look for teams with similar rating spreads. They're easier to come by than full 2k 4v4 teams anyway since some people don't queue in a group and would affect waiting times far less.

A fix to the lane balance could make that scenario better than it is now.

There isn't a fix because someone semi competent at the game can utterly throttle someone with no idea all the time, forever. The reason the guy is 3300 is because he has some level of understanding of how to win the game and plays dudes with zero clue. For it to "fix" itself everyone on FAF at 1800+ is going to gain like 600-800 rating because that's what they would be capable of gaining by just sitting in all welcome games farming 1v0 games. But people don't do that, because it isn't fun to gank bad players for most of the people of the game. People play relatively close players in TS because those lead to the games which are close and you can actually think about what to do.

This is a problem that exists in every rating system, these distribution algorithms are not good at the utter tails where the difference in 100 rating is a 99.9997% win rate or a 99.9999997% win rate.

"Fixing lane balance" also sucks as a solution because you force boring maps where every slot is facing an equal counterbalance of itself. The fun games are where slots aren't mirrored and one slot has 15 mexes vs another than has 10. It introduces gameplay elements beyond "well my mirror did X so now I do X because we always have the same resources available to us at all times."

Ftx, that was in response to the ridiculous replay you posted with the global rating abuser and the stupid air balance on display. there was a 1400 player on the opposite team that should have been vs white and you can't possibly be arguing that the 1200 vs 1800 garbage that the opti balancer came up with for air was "interesting". Given a bunch of players that don't queue together, autobalance does a better job of arranging a decent game than opti, not least because it also takes uncertainty into account. You also say that some noobs complain when pitted against much stronger opponents by opti balance (presumably a bad thing, since you seemed to think this argument supports your claims), then you go back and defend that as being "interesting". Funnily enough, it's an interesting thing you want gone. Take your pick, you can't have it both ways.

Autobalance doesn't currently support pre-made teams but it's a feature Penguin had already prototyped but got cut, I'm sure it could be brought back.

You want the tails of the distribution to only play among themselves? good luck finding a rated game at any other time than saturday night CET. The tails are too thin with the current player count for that to work otherwise.

@ftxcommando said in Matchmaker Pool Feedback Thread:

The logic of building up to map gen is that lower skill brackets need to "learn the lessons" of some maps in a more controlled environment.

I'd totally agree with you here, except for the fact that many of the 'low elo' maps actually teach completely different lessons than those required for top-level play.
Our discussion of williamson's bridge, and its lack of 'going around defenses' tactics, are important here. Many of the low-ELo ladder maps actually teach the wrong lessons - while random maps would teach better lessons.
(And, also, random maps would remove the advantage given to players than have just played a lot of games, too)

The rest of this discussion (carry slots, etc) is totally over my head, I'm afraid! At the very least I'll offer the data point that I've been given 'air; in every team game I've queued for in the matchmaker (I know this because at the start of the each team game I ask what I'm supposed to be doing!).

The only other data-point I feel might be useful is that a team-game 1000-rating player seems roughly equivalent to a 400 1v1 player.

Again, this is a newcomer's perspective. x

@xiaomao said in Matchmaker Pool Feedback Thread:

G-great. You picked a game with underrated Voodo stack, full on voice and playing same squad every fucking day be it 4v4 or 3v3.
The 500 rated player you see there is actually 1.1k in 4v4 with his Voodo stack.

You better watch this https://replay.faforever.com/19738422 and tell me this is fine.

Also, this is just sad:

XD

You misunderstand my statement about tails. I’m talking about the individual result of a 500 beating a 1900 or whatever equivalent. Is this a 99.99997% end result or a 99.9999997% end result? Impossible for the system to ever know because nobody is simulating 3000 of these games to normalize the ratings across this disparate of a rating difference.

Though this is mainly a problem when combined with the reality of teamgames basically always being possible to balance at 85% or better and therefore allow this level of farming. Because again, you’re pretty much not facing any equivalent opponent if you’re like 2000 on sentons air with 500 allies but the whole enemy team consists of 1000s.

My proposal of allowing 1000 rating of breathing room effectively means everyone has access to the mean population group, so the issue of games doesn’t really matter. Except if you’re like -500 or 2500, but these dudes should still not really be playing with 1000s honestly.

@spikeynoob you guys did a sweaty t2 air all in vs 1ks smh xD

good evidence for ftx though