How come you don't play ladder?
-
We could explain how the matchmaker works in a wiki article, but I don't think it will improve the situation. I wrote the matchmaking algorithm and even I can't predict if a match will happen the next cycle. The algorithm is complex. To know what it will do you have to know all the inputs, but the client only shows the number of people in queue. You could suggest to add more info, but let me demonstrate what questions you need to answer to make an informed decision:
Are there any premades? Who has less than 10 games? How long have they been queueing already? What ratings does each search parties have? What was the average rating of the people in the queue during the last day? What are the config settings for the matchmaker?
All of this to answer the only question that I as a player are really interested in: Will I get matched soon?When you argue for quicker matches you are in the bad game vs. no game discussion again as FtX said and we as a community will never reach a consensus that makes everybody happy in that regard.
-
I still think we should have an in-client slider option:
get a match faster <---------------------------> get a match with better team balancepfp credit to gieb
-
@penguin_ said in How come you don't play ladder?:
I still think we should have an in-client slider option:
get a match faster <---------------------------> get a match with better team balancefinding an algorithm to take everyone's preferences into account in a sensible way sounds very hard for such a slider
-
@penguin_ said in How come you don't play ladder?:
I still think we should have an in-client slider option:
get a match faster <---------------------------> get a match with better team balanceI still think this is technically unfeasible
-
It’s also a recipe for total disaster.
“I have it on get match faster and I got no game WTF!!!!”
Yeah well, the other 12 people in queue have it on “getting a better team” whatever that means.
Or the reverse of 7 people having it on matching faster along with a guy saying he wants a better game. There is just too much obfuscation with a matchmaking system and pretending like a button gives much control over it will just make people more bothered.
-
Maybe we should add a nonfunctioning slider just for the placebo effect xd
-
I remember joining SupCom at GPGnet times and i didn't know English enough to communicate with anyone. So rarely played teamgames. GPGnet didn't had any other rating than ladder rating so balancing custom games was tricky.
Actually remember thinking that ladder is awesome for new players cause it gives you an opponent of your skill level -
I don't see any reason it would be technically unfeasible.
You could give everyone a slider-based matching bonus value that is proportional to their selection on the slider (with the better team balance end of the slider giving a bonus of 0, or a negative amount, depending on how you want to set it up).
Example A:
Then, if you only want to put people in matches of equal or better game balance than their individual selection on the slider, you could just cap that bonus value for a given combination of players for a potential match based on the minimum of their individual bonus values (so, if you're matching 4v4's with 20 people in queue, the bonus value used for a given potential match would be the lowest bonus value of the particular 8 players in the potential match that is being considered, and a different value could be used when evaluating a different set of 8 players within the same matching cycle).Example B:
Alternatively, if you want to not restrict in that way and make matches more likely if anyone wants faster matching, then you could do the same thing but not restrict the bonus value to that of the minimum, but instead make it equal to that of the average of the set of players being considered for a potential match. That could put people in matches of better, equal, or worse game balance than their individual selection on the slider, but it would factor everyone's individual preference in with equal weight.I think either of those systems could be better than what we have now.
-
@Penguin_
Is there a way to implement some slider that behaves mostly consistently with the aggregate preferences of the people in queue? yes.
Is there a way to implement such a slider so that it doesn't just cause annoyance to everyone using it without providing any real benefit? Not that I'm aware offThe main problem is that for such a slider to make sense, you'd essentially split the queue into two sub-queues (with some blurriness inbetween). One queue that wants to wait longer for a more balanced game and another that just wants to play whatever.
If we had queue sizes of like 100+ players, you could do that, because two queues of 50 players each, with some overlap from time to time when the "fast" queue players get matched into a "balanced" queue game, still works quite well.
In the reality of faf where queue sizes for tmm are like 20 players, this doesn't work because two queues of 10 players are entirely non functional where a single 20 player queue is not
-
No professional matchmaking system works like this, even ones with hundreds or thousands of players to parse through.
-
@cheeseberry You don't have to do a queue split - for example a simple implementation although it'd inevitably leave some people unhappy (but fewer than it would presumably make happy) would be to have a 'preference' system - you indicate if you would prefer to find a game faster, slower, or no different to default.
Then the algorithm calculates the most popular choice (you could even do a threshold, so it would only go with something other than default if more than 50% of players want the adjustment), and weights things based on thatSo, if 8 people queue, 7 indicate they want a game sooner and 1 indicates they would prefer to wait for a better balanced game, the algorithm will try and find a game sooner.
If 10 people queue, 4 want a game sooner, 4 want a game slower with lower rating difference, and 2 indicate no preference, then the algorithm will work as it currently does.
As a result, more people benefit than suffer, although it's not without flaws (e.g. 7 high rated players all indicate they want a lower rating difference, and 8 lower rated players all indicate they want a game sooner than later, resulting in some of the higher rated players being matched in a poorly balanced lobby, although I expect some tweaks could reduce the likelihood of this happening)M27AI and M28AI developer; Devlogs and more general AI development guide:
https://forum.faforever.com/topic/2373/ai-development-guide-and-m27ai-v71-devlog
https://forum.faforever.com/topic/5331/m28ai-devlog-v130 -
Why have a timer until matching attempt occurs at all? It would make sense like @Gibsaw already pointed out to try to match as soon as somebody new joins the queue for example. If nobody new joins and nobody leaves the queue the only thing changing is the duration the people are waiting, and even then it doesn't make sense (except from easier algorithm implementation standpoint) to have regular matching intervals, but rather it would make more sense to have some threshold values. For example, if people are more than 2 minutes in queue and gap between rating is somewhat big, but not too big, match.
You said:
"Having these regular cycles makes it easier to track how long people have been waiting in the queue than if we matched at irregular intervals."Could you elaborate what do you mean by that? You could just have a timer that tracks how long you've been in the queue alredy, not a timer that tracks how long until matching attempt. That brings me to the 2nd point: Even if you keep the current matchmaking algorithm it would make more sense from user experience standpoint to only show duration you've alredy been in queue. Why expose one detail of how matchmaking works (by showing it only matches on fixed interval) if you don't expose any other details and the queuer still doesn't have a good way of tracking how long they've already been in the queue? Just my opinion of course.
-
@maudlin27 said in How come you don't play ladder?:
@cheeseberry You don't have to do a queue split - for example a simple implementation although it'd inevitably leave some people unhappy (but fewer than it would presumably make happy) would be to have a 'preference' system - you indicate if you would prefer to find a game faster, slower, or no different to default.
Then the algorithm calculates the most popular choice (you could even do a threshold, so it would only go with something other than default if more than 50% of players want the adjustment), and weights things based on thatSo, if 8 people queue, 7 indicate they want a game sooner and 1 indicates they would prefer to wait for a better balanced game, the algorithm will try and find a game sooner.
If 10 people queue, 4 want a game sooner, 4 want a game slower with lower rating difference, and 2 indicate no preference, then the algorithm will work as it currently does.
As a result, more people benefit than suffer, although it's not without flaws (e.g. 7 high rated players all indicate they want a lower rating difference, and 8 lower rated players all indicate they want a game sooner than later, resulting in some of the higher rated players being matched in a poorly balanced lobby, although I expect some tweaks could reduce the likelihood of this happening)This sounds even worse than the current system and I’d probably never queue again if I would be getting forced into a 1300 average rating game because some dudes wanna play. I already basically only queue in a party since that allows me to artificially control what “average level” of game I want.
-
I think you might've misunderstood as I don't think the queue would split like you imagine with example A or example B. With example A, fast-matching players would still match with better-balance players when the balance would be good enough, and if the better-balance players had already been in the queue long enough, they would presumably have a bonus due to wait time like they do now afaik. With example B, they would be even more mixed.
-
@Penguin_ @maudlin27
I probably phrased my explanation poorly: The queue doesn't need to be literally split in two for very similar behavior to take place as if it were split in too. In the end, you are further dividing an already small userbase along some preference.
In an ideal world, you could do this perfectly where you could input e.g. "I want a game with exactly 327 or less rating deviation" and then exactly get a game with such a deviation as fast as possible.
This is obviously impractical for many reasons and "get me a balanced game" or "get me any game asap" are preferences you can state, but are only fulfillable with everyone being satisfied if there are enough people on close enough to your end of the spectrum for a game, specifically a balanced game, to ever be found.
This is not the case for faf.
If we implemented such a slider, what will predictably happen is that either
- You have enough "fast game" players that they can find unbalanced games every couple minutes and the dozen or so "balanced game" players will literally never find a game, or
- You put players from the "balanced game" preference into unbalanced games, who will then probably just leave the game because they got literally the opposite of what they asked for.
Effectively you have split the queue in two even though, of course, a slider is technically gradual.
Not only is this all of this undesirable, this then gives rise to quite a few perverse incentives, because the slider states your preferences relative to other players: If you know that you will never get a game on the 100% "balanced" side of the slider, you want to be just slightly more on the unbalanced side than the other players looking for a "balanced" game, or you will predictably be excluded out of even most "balanced" games that you'd otherwise like to have played.
Such a slider would effectively be a crude voting system where strategic voting is the optimal choice, i.e. not stating your true preferences is optimal to achieve your prefered outcome.
Listen, I don't want to read 5 page long guides about the state of the "slider meta", the normal faf meta is already enough
-
As FTX mentioned there is a reason that basically no professional matchmaking system works like this, even for games with a much larger playerbase (bandwagon fallacy notwithstanding)
Of course, it could be possible that such a better matchmaking system does exist. One where not only strategic voting is not a thing, but that is also implementable for a large range of players in the queue and that works with the psychology of real people taken into account.
If you can concretely define how this would work, or can link me the implementation of some other game that does this, I'd love to hear about it
-
I get your point, but I think you still misunderstand. The impact of the preference slider wouldn't be so large that it would outweigh everything else combined or whatever. It would have a sensible weight that makes it one factor out of several. The largest factor should continue to be potential game balance. AFAIK, we already factor in things like a new player matching bonus, a high-rating matching bonus, a waiting time matching bonus, etc. The faster-game-preference matching bonus would just be one more factor on the list that would influence things, but it wouldn't be dominant or anything. It basically would just shift people a bit on the spectrum of faster matching vs better balance, but it wouldn't be the end-all-be-all in deciding who gets matched when.
-
@Penguin_
The resulting problems and incentives would still be the same, just less pronounced.
How do you get from:
Slider doesn't matter -> it doesn't matter
Slider matters a lot -> It makes the queue worseto
Slider matters some magic amount -> It makes the queue better?
That being said, just extend the current algorithm to take such a slider preference into account and prove me wrong. I wouldn't know how to do that, but that doesn't mean it can't be done!
-
Every single factor you listed has zero player input. Yours doesn’t. This is the big difference everything posted is about.
Saying “it will just kinda matter” is a problem when the generic end user hits a button and expects that button press to do what the button press says. You cannot guarantee that in a system with 8 independent button presses without making it even more of a PITA to get a game. Ironically the thing this was supposed to solve.
-
A ton of things are on a spectrum where you don't want too little or too much. People can die from having too little water or too much water, too much heat or too little heat, etc. A lot of people like games that are within a certain difficulty range rather than being too easy or too hard. Different people can have different preferences. I don't know what specific slider weight would be optimal, but I imagine it's potentially somewhere in the ballpark of the weights that we give the other factors.
@FtXCommando
The slider would still have an impact. Being able to set one's preferance for that would be a nice QoL improvement that would allow people to choose their preference on the spectrum of better game quality vs faster matching. We already have people being confused and frustrated about not getting a match. This wouldn't create that problem. That problem exists either way. A little info blurb stating that using the slider makes your preference more likely but not guaranteed, could be added if people really want...I'm not claiming that example A or example B are perfect, just that I think the system could be better with either of them than it is without them as it is now. There are pros and cons for each of them compared with each other and compared with the current system. I am inclined to think the pros outweigh the cons for either of them.
