Allow us to pick rating brackets for TMM

Give me something to work with instead of just ranting on the forum Endranii. As a 1.6k player, what is the smallest rating you are willing to accept?

And can we please in general discuss with a bit more detail? I think everybody is on the same page that we want to have wait times as small as possible and reasonably well balanced games. If the games are shit this will kill the matchmaker in no time. I believe the audience of "give me any game, no matter who is in it" is rather small. If we deem any wait time increase unacceptable then this whole discussion is pointless. And it would be pretty arbitrary too. The wait time is already greater than zero, so why is that the absolute maximum of acceptably wait time?

I think it is acceptable to have wait times for new people as long as we communicate that properly. I see two problems with the current UX: 1. It's very hard to gauge the general queue activity, because people get matched and are then basically invisible. 2. The queue can reach over 8 people without matching and the reason is not apparent.
Increasing game quality and thus wait time could partially fix this, as the queue will basically constantly be over 8 people during active hours, so it will be more obvious that this is intentional. The general activity could then be gauged by the queue size.
However, this is still a very crude fix as the reason for this is still not obvious and it will even more regularly than now lead to people not getting matched repeatedly while seeing other people getting matched.
Any ideas how to improve this?

I don't think a user selectable switch like that is viable, because we can't guarantee that it will actually do anything. If seven people in queue selected they want a fast game, but the rest has not, then we still can't match them without dragging someone in that potentially bad game that has specifically selected that they don't want an unbalanced game.

Another thing that we should probably do regardless of touching the main parameters, is to find a solution for the people at the outer ends of the bell curve. The wait times increase drastically. For the left side is probably acceptable to match a bit more leniently, as there will be many people with high deviation that we can't accurately match anyway. It would also automatically improve wait times for new people, as new people are low rated.
An open question is: Do we also want that for the high rated end?

And a little bit more technical question. When determining where the top of the bell curve is and how wide it is, do we deem it acceptable to just hardcode a value, or do we try to recalculate the bell curve regularly. If yes, based on what data do we want to do that?

@blackyps In terms of the switch option, I'd expect it to be relevant if at least 8 people in the queue hadn't checked the option to restrict games they join to those within a smaller rating range, so it'd be limited in effect, but still relevant (whether or not it'd be enough of an impact to justify the additional time to create I don't know though). If only 7 people had chosen 'unbalanced game > no game' then it'd have no impact (since as you say the alternative would mean forcing someone who would rather have no game to play).

In terms of UI improvements regarding the wait, one option could be an 'average wait time' figure, although quite how you would come up with it I dont know. E.g. if it's possible (and not lots of extra work) you'd track the average time each person that joined the last 5 games had to wait before the game was formed, and then present that as the average figure (with an average of more than a certain amount, say 10m, just displayed as '10m+' (to avoid discouraging people too much if no games have formed for a long time due to it being an off-peak time). An alternative to coming up with the average wait time could be to track the average time to form a game based on the number of players in the queue (and potentially the rating range of those players) in aggregate (e.g. over a month) and use this to come up with a formula to approximate the number.

Another UI improvement could be to highlight the rating range currently being considered for you to be matched into a game, since I understand this increases with each queue cycle where you aren't matched (although I dont know if it is quite that simple). This could also be used as an alternative to a 'checkbox' selection approach, in that it would allow people to leave the queue and rejoin it to reset the rating range being considered for them to be matched into a game (thus achieving a similar effect to the 'checkbox' option without any extra coding being required). It'd also help newer players since they might think 'ah there are 8 players in the queue but I have a rating range so those other players can't be in a similar enough range to me, and my rating range has now increased so if I hang around a few more minutes maybe I'll get a game'.

However I wouldn't favour actually displaying the rating ranges of the other people in the queue since this risks disincentivising people from joining or staying in the queue if they see that all the people in the queue are a different range from them (as it is, I find if a few people join the queue there might quickly be an influx of more people who think the chances of a game have increased, which then becomes a self fulfilling prophecy)

For the distribution I would advise just taking n-standard deviations away to make it robust to change in the future. Hard coding just will cause problems going forward. Also considering the fact that it would technically be different for each rating.

@sheikah said in Allow us to pick rating brackets for TMM:

For the distribution I would advise just taking n-standard deviations away to make it robust to change in the future.

I don't understand what this is supposed to mean

@blackyps said in Allow us to pick rating brackets for TMM:

@sheikah said in Allow us to pick rating brackets for TMM:

For the distribution I would advise just taking n-standard deviations away to make it robust to change in the future.

I don't understand what this is supposed to mean

We should calculate the mean and standard deviation for each rating and then set the threshold so that the top and bottom x% of people are treated differently if you want to try something to squash the tail wait times.

Anything else will just be a headache to maintain.

We probably should strive for making the Matchmaker match new people as fast as possible while not sacrificing the smaller but still important segment of 1.2k+. For that, you need to decrease the max deviation and imbalance in the 1k-1.8k range but then increase that again above 1.8k since at that point people are so far apart that otherwise it's impossible to get games and it's commonly accepted that high-rated lobbies usually start at around 1.6k-1.8k and 2k+ ones are a rarity

Could you simply display the league divisions that are currently searching if you hover over the In Queue? You can make that information specific in terms of Diamond I/Diamond III or it could just be
Silver: 1
Gold: 4
Diamond: 2

The main issue is that solo queueing as an 1800+ is just absurdly risky because if 6 Masters/GMs are in queue solo, it is still highly possible the system wants to make a game with only 2 of those depending on the rest of the queue. But at least seeing who is in queue might encourage some loose players at those ratings to queue up if they know a decent game is conceivable.

You can even combine that with new IRC chat rooms that autojoin you based on matchmaker ratings that you can opt-out of through the autojoin setting in settings. Or people could actually starting using the Discord rating channels 4head.

@wikingest said in Allow us to pick rating brackets for TMM:

@casternumerouno said in Allow us to pick rating brackets for TMM:

I can as well get by hosting all welcome

You can host whatever rating bracket or balance you want in custom games.

Then what the fuck is the point of TMM? Isn't it supposed to make a balanced game? If it ends up making games that are no different from random ass all welcome lobbies then what is the point of it? Just make the whole Q unranked and call it a day then.

@blackyps said in Allow us to pick rating brackets for TMM:

Give me something to work with instead of just ranting on the forum Endranii. As a 1.6k player, what is the smallest rating you are willing to accept?

It might have been lost in the discussion but I already said that we should most likely try by limiting it by brackets.
For example Gold V can match with Silver V and Diamond V which looks decent at first sight.

As for the bigger outliers like masters and wood V you could give them 1.5 bracket leeway to make it easier for them to get games without making the games too imbalanced? Though looking at it now the 1.5 bracket might still be too high for the GM as D3 is around 1.4k rating while Blodir is 2.3k or so.

Also moving to brackets/divisions could be a positive move as it's way easier to understand what D4 stands for rather than 1.3k True skill rating.

@blackyps said in Allow us to pick rating brackets for TMM:

I think it is acceptable to have wait times for new people as long as we communicate that properly. I see two problems with the current UX: 1. It's very hard to gauge the general queue activity, because people get matched and are then basically invisible. 2. The queue can reach over 8 people without matching and the reason is not apparent.
Increasing game quality and thus wait time could partially fix this, as the queue will basically constantly be over 8 people during active hours, so it will be more obvious that this is intentional. The general activity could then be gauged by the queue size.
However, this is still a very crude fix as the reason for this is still not obvious and it will even more regularly than now lead to people not getting matched repeatedly while seeing other people getting matched.
Any ideas how to improve this?

https://forum.faforever.com/topic/3561/show-amount-of-players-in-game-in-the-tmm-play-tab?_=1665430587911

Could also add additional split of players in Q by their rating bracket, near the people searching number, same as it is done in world of tanks matchmaking where you can/could see the amount of players in the Q per tier/tank type.

@blackyps said in Allow us to pick rating brackets for TMM:

I think it is acceptable to have wait times for new people as long as we communicate that properly. I see two problems with the current UX: 1. It's very hard to gauge the general queue activity, because people get matched and are then basically invisible. 2. The queue can reach over 8 people without matching and the reason is not apparent.
Increasing game quality and thus wait time could partially fix this, as the queue will basically constantly be over 8 people during active hours, so it will be more obvious that this is intentional. The general activity could then be gauged by the queue size.
However, this is still a very crude fix as the reason for this is still not obvious and it will even more regularly than now lead to people not getting matched repeatedly while seeing other people getting matched.
Any ideas how to improve this?

For each queue, you could list the number of matches played (from that queue) in the last hour (preferably only counting matches that last >5 min or smth).

Also, in addition to displaying the total number of users in queue, you could also display the number of users in queue that the player is within a compatible rating range with, ie:
15 in queue (10 compatible)
or
15 in queue (10 near your rating)
or
15 in queue (10 you can match with)

I will reply to other things mentioned in the thread later.

I like the idea of displaying the division brackets of other players, but it might be technically complicated to implement, because the league service and the rest of the server is very separated. They don't even use the same database.

We can't accurately display how many people are compatible to match due to how the matchmaker works. The rating difference is not the only metric, wait time and how balanced the resulting game will be are also factors. And even if you have sufficient people to match, the matchmaker might choose a different game that doesn't have you in it.

Restricting the match range to +- one brackets is not feasible I think. That is 340 rating. There is already a noticable uptick in wait times to restrict games to about 700 rating difference as you can see in the graphs.

At risk of rocking the boat too much, I think the matchmaking algorithm is more or less fine. In my experience reading discord and aeolus, there are a pretty similar number of people complaining about the balance of matches as there are complaining about the matchmaking times.

@swkoll said in Allow us to pick rating brackets for TMM:

At risk of rocking the boat too much, I think the matchmaking algorithm is more or less fine. In my experience reading discord and aeolus, there are a pretty similar number of people complaining about the balance of matches as there are complaining about the matchmaking times.

Different people. 1500s are complaining about balance and 2000s complaining about matchmaking times (or more realistically just sucking it up and hosting custom, completely ignoring the existence of tmm).

@casternumerouno said in Allow us to pick rating brackets for TMM:

Then what the fuck is the point of TMM? Isn't it supposed to make a balanced game? If it ends up making games that are no different from random ass all welcome lobbies then what is the point of it? Just make the whole Q unranked and call it a day then.

All those games were balanced. Quite well balanced. Whoever made the algorithm, did a great job balancing games. Just look at the small rating difference between affronting teams.

The discussion is about one(two) person who do not like playing with lower rated players. Not about bad balance. While there is probably many players who like playing with lower rated players, and who could leave the matchmaker when things get changed. But it seems that we do not have exact statistic about that at this moment. And it is in the point where "who is more toxic and screams stronger gets heard".

@casternumerouno said in Allow us to pick rating brackets for TMM:

Then what the fuck is the point of TMM?

The point of TMM, is that people who do not have friendlist in FAF can just join games, and play with random people(without getting thrown out, insulted etc.).

switch option

Let me put it this way: This feature would be complicated to implement and I personally don't think it would be too useful, so I won't work on this. Maybe you can find somebody else that is willing to do it, but I don't think this will be implemented in the foreseeable future.

@ftxcommando said in Allow us to pick rating brackets for TMM:

The main issue is that solo queueing as an 1800+ is just absurdly risky because if 6 Masters/GMs are in queue solo, it is still highly possible the system wants to make a game with only 2 of those depending on the rest of the queue.

Uhh, so what is the absurd risk here? Not getting matched? That doesn't sound too severe. Am I missing something?

match time

I thought about this some more and I don't think we need to be as fast as possible. We just need to be faster or about as fast as custom lobbies filling up. Does anyone know how long e.g. astro or an all welcome map gen game takes to fill? I don't play custom games.

Everybody knows this community is not huge, so people should be able to understand that one minute match times are not realistic. A system that throughs you in games fast is appealing at first, but if it turns out that these games are too unbalanced (and balanced in this context also means that the rating ranges of the players are reasonably close) to be enjoyable then people will stop queueing after some games. I think it is much better for long term player retention in the matchmaker if you can get nice games with wait times that are in the range of reasonably fast filling custom lobbies.

@blackyps said in Allow us to pick rating brackets for TMM:

@ftxcommando said in Allow us to pick rating brackets for TMM:

The main issue is that solo queueing as an 1800+ is just absurdly risky because if 6 Masters/GMs are in queue solo, it is still highly possible the system wants to make a game with only 2 of those depending on the rest of the queue.

Uhh, so what is the absurd risk here? Not getting matched? That doesn't sound too severe. Am I missing something?

By absurd risk I mean that I personally would rather have no game over getting dumped into a game where I am 1800 with an 1800 enemy and then everyone else in the game is 1000-1200. Some people are fine with it, others aren’t. But that possibility is what stops me from queueing alone. I want a game where I feel like I can actually coordinate with my allies to accomplish something.

I played around a bit more and gave people at the edge of the distribution an additional bonus. With this we can improve the situation at the very high level spectrum, by getting more high level games and we can drastically improve the situation in the mid-range where we have the playerbase to be more strict with matching.
I looked up the actual queue activity and noticed that I estimated it way too low in my previous runs. So I include the graphs for the current situation again.

Let me say that yes, the wait times still increase a bit, but if you look at the absolute numbers in the bottom left, you notice that the situation is not too bad. At the moment almost half the people get matched instantly. By increasing the average wait time by less than even one queue pop, we can drastically improve the game quality.
Waiting three minutes more to get an actual high quality game seems like a good tradeoff to me.

Current:

New:

The new system might be worthwhile for 1k+ players, but I suggest you do something closer to the current system's wait times for new players (who are generally <1k).

Please elaborate why you think that. And don't just look at the wait time charts. The current matchmaker is able to create games with 1800 rating difference. What do you think about that?