We recently just rolled out a beta test of a new matchmaking change that is live in NA only. When we make sure it has the desired effects, we'll turn it on globally.
--
The matchmaker will now take into account the [Number of Wins] a player has in every matchmade queue (Normal SR, Normal TT, Ranked Solo/Duo, etc).
This means that for the first few minutes, the matchmaker will attempt to find teammates and opponents that have a relatively similar [Number of Wins] as you in addition to trying to find teammates and opponents that also have a similar [Elo]. However, this does not mean it's impossible to get matched into a game with a large difference in [Number of Wins]--it's just more rare.
Happy to discuss the matchmaking change here or any other matchmaking questions players might have. For reference, I've provided some older matchmaking discussions we've had in the past:
but number of wins doesnt directly mean a more skilled player... that applies especially to ranked. you can have plat level players who have <10 ranked wins in the same game with honestly bad players who have 200 wins, but still hover around 1200 elo.
have you tested how this will affect matchmaking speed? you said itll be trying to do this for the first few minutes... my queues are sub 20 seconds usually. does that mean that my queue time will be pushed to a few minutes? just so i can be matched with people with similar pointless stats?
We agree, [Number of Wins] doesn't directly correlate with higher skill; however, we've done some initial analyses and there's some tangible advantages to adding this parameter to the matchmaker.
We do know that adding this parameter may increase queue times for a number of players which is why we are doing a small beta test in NA first to determine the actual impact and whether we are ready to roll out the change to every territory.
Actually, Yegg will talk about something similar to this very soon. In the past, we knew there was a problem with Ranked players who played 90% Ranked and 10% Normal Blind Pick. Sometimes, these players were high-skill Ranked players that would join a Blind Pick game and crush players because their Blind Pick 'hidden' Elo was quite low.
Yegg's new matchmaking feature will fix this but I'll let him explain it in the future.
Turned on since midnight? Are you sure?... I played a few ranked, and was matched with some 900+ win/lose players. And the queue was like 40 secs long on average.
We're playing with the ramp up times; basically, for each second in the queue, how wide of a range in number of wins is the matchmaker OK with?
Right now it's pretty aggressive to make sure that players with an extreme low or high number of wins don't sit in queues for an arbitrarily long time.
So what problem is this supposed to solve exactly?
People complaining that the low win person lost them the game will continue to complain. The tangible effects on matchmaking that I could see would be longer queue times and a higher number of games before the matchmaking can be certain of your rating. This higher number of games would be caused by the system having to relearn skill levels as players leave the fish bowl of low wins and enter the "real world" of league. Furthermore I think it would actually increase the variance in skill levels for games made up of low win players. Not knowing more than league uses a modified Elo rating system I understand my conclusions might be off base, but in chess this seems like a very bad way to do matchmaking. You want to pair new people with the player you are certain is average over pairing the new people together and learning nothing about them relative to the rest of the population.
There's one slight nuance to your analysis: League of Legends has a level and unlock system that isn't true in Chess. As a player plays more games they get more IP which is then used to unlock Champions or Runes.
Secondly, we're not replacing Elo or the various other skill metrics in the matchmaker. We're simply adding a new parameter that the matchmaker will begin taking into account.
So when you say the matchmaker makes predictions on the outcome of the game, does that mean that it purposely favors one team or the other from the matching? If so, that'd be pretty outrageous. Especially in ranked.
No.
Every matchmaker in every game (even Chess!) actually makes predictions about the match (after it the teams have been made). This is one way scientists or designers can measure the efficiency and accuracy of their system. The subtle goal of every matchmaker is actually to be able to predict the outcome of any match with perfect accuracy.
For example, a perfect matchmaker might be able to predict that when you put a 1800 rated player against a 1500 rated player that the 1800 rated player will win 75 out of 100 games. If we force these players to play 100 games and the 1800 rated player actually wins 75, then the matchmaker is extremely accurate.
However, we wouldn't actually ask the matchmaker to compose these matches.
Lyte, this is change i have been begging after 500 games won at 1200 and being matched with 5 win newbies... however isn't amount of matches reset with preseason? That would make collecting data and testig this feature in this particular moment a bit hard wouldn't it?
We haven't reset yet, so this is the perfect time to collect data on a 'mature' set of player records before we enter Season 3 after a reset.
Adding [Number of Wins] to the matchmaking parameters
--
The matchmaker will now take into account the [Number of Wins] a player has in every matchmade queue (Normal SR, Normal TT, Ranked Solo/Duo, etc).
This means that for the first few minutes, the matchmaker will attempt to find teammates and opponents that have a relatively similar [Number of Wins] as you in addition to trying to find teammates and opponents that also have a similar [Elo]. However, this does not mean it's impossible to get matched into a game with a large difference in [Number of Wins]--it's just more rare.
Happy to discuss the matchmaking change here or any other matchmaking questions players might have. For reference, I've provided some older matchmaking discussions we've had in the past:
http://na.leagueoflegends.com/board/....php?t=1827767
http://na.leagueoflegends.com/board/....php?t=1976101
PS - RiotKiddington was the dev that did most of the work.
have you tested how this will affect matchmaking speed? you said itll be trying to do this for the first few minutes... my queues are sub 20 seconds usually. does that mean that my queue time will be pushed to a few minutes? just so i can be matched with people with similar pointless stats?
We do know that adding this parameter may increase queue times for a number of players which is why we are doing a small beta test in NA first to determine the actual impact and whether we are ready to roll out the change to every territory.
Yegg's new matchmaking feature will fix this but I'll let him explain it in the future.
If not, when?
SB
2) When the matchmaker predicts the outcome of a match, does the accuracy of the prediction match the actual outcome?
Right now it's pretty aggressive to make sure that players with an extreme low or high number of wins don't sit in queues for an arbitrarily long time.
People complaining that the low win person lost them the game will continue to complain. The tangible effects on matchmaking that I could see would be longer queue times and a higher number of games before the matchmaking can be certain of your rating. This higher number of games would be caused by the system having to relearn skill levels as players leave the fish bowl of low wins and enter the "real world" of league. Furthermore I think it would actually increase the variance in skill levels for games made up of low win players. Not knowing more than league uses a modified Elo rating system I understand my conclusions might be off base, but in chess this seems like a very bad way to do matchmaking. You want to pair new people with the player you are certain is average over pairing the new people together and learning nothing about them relative to the rest of the population.
Secondly, we're not replacing Elo or the various other skill metrics in the matchmaker. We're simply adding a new parameter that the matchmaker will begin taking into account.
Every matchmaker in every game (even Chess!) actually makes predictions about the match (after it the teams have been made). This is one way scientists or designers can measure the efficiency and accuracy of their system. The subtle goal of every matchmaker is actually to be able to predict the outcome of any match with perfect accuracy.
For example, a perfect matchmaker might be able to predict that when you put a 1800 rated player against a 1500 rated player that the 1800 rated player will win 75 out of 100 games. If we force these players to play 100 games and the 1800 rated player actually wins 75, then the matchmaker is extremely accurate.
However, we wouldn't actually ask the matchmaker to compose these matches.