Welcome to the Forum Archive!

Years of conversation fill a ton of digital pages, and we've kept all of it accessible to browse or copy over. Whether you're looking for reveal articles for older champions, or the first time that Rammus rolled into an "OK" thread, or anything in between, you can find it here. When you're finished, check out the boards to join in the latest League of Legends discussions.

GO TO BOARDS


LOL Matchmaking Explained

Comment below rating threshold, click here to show it.

Frag

Recruiter

09-15-2009

Good read, I love you Zileas; oh, and btw i'm pregnant with your child and will take riot point in loo of child support.


Comment below rating threshold, click here to show it.

Rooks

Member

09-15-2009

Hm makes sense. I really like to play competitively (as you might recall =] ), and am just making sure if I chose to solo queue I will stand as good as chance as anyone else, be it pre-mades or not, at being ranked #1.


Comment below rating threshold, click here to show it.

LoveBuzz

Member

09-15-2009

Quote:
Zileas:
Yes, because that's a bug in the present system where it considers partial premades way stronger than they actually are.


Ahhh... well at least my profanity-laced tirades are actually founded in reality.


Comment below rating threshold, click here to show it.

Teirdome

Recruiter

09-15-2009

How leavers are handled is pretty interesting, but I'm not sure that I'm a fan of it for ranked matches. By the same token, I'm not sure there's a better way to handle it. If you were to make it so that the leaving player always took a loss, you would have to decrease the amount of Elo lost for the remaining team members to prevent deflation. The other option is to increase the amount of points won for the non-leavers, but then you're creating greater inaccuracies for the non-leaver's Elo. I think it's the right solution in the end, but I wish there was a bit more punishment for leavers.


Comment below rating threshold, click here to show it.

Depravity

Junior Member

09-15-2009

would you guys ever factor in the champion picked into the rating? i know with the current set up champions are picked after the team is assembled but certain players are noticably better with specific champions. and i think picking someone you dont usually play or suck with would give a drastic disadvantage, just a thought.


Comment below rating threshold, click here to show it.

Demandred

Senior Member

09-15-2009

Quote:
Depravity:
would you guys ever factor in the champion picked into the rating? i know with the current set up champions are picked after the team is assembled but certain players are noticably better with specific champions. and i think picking someone you dont usually play or suck with would give a drastic disadvantage, just a thought.


No. I am not part of riot staff but I can say: No. It'd get into "Oh which champ is better." If uyouare speaking as I've never played him before, still no.


Comment below rating threshold, click here to show it.

Savage

Senior Member

09-15-2009

Short version: Outside of the edge cases where the current system falls apart, if I solo queue for all of my games, my W/L ratio will approach 50% as games go to infinity, provided I'm matched against solo queues. If I only queue in 5 man premades, my W/L ratio will approach a natural equilibrium percentage that correlates positively with my rating. If I solo queue against a mix of solo queues and 5 man premades, my W/L ratio will be somewhere between 50% and the equilibrium value in proportion to the density of premades in the system.

I consider this a problem.


Quote:
Zileas:
That is the case for VERY VERY high rated matches more often than not (for solo queues vs solo queues, which is again rare at the very highest levels)


Well your model pushes all of the variance in win percentages out of solo queues, who get their natural win percentages tanked by the matchmaker, and into a combination of 5 man premades and the edge cases. As the two overlap it isn't all that surprising that things start to blow up there.

Note that this will make the 'solo queuing will tank your rating and win percentage' effect stronger for top players, since they're going to be sucking up an inflated share of the top end of the variance through 5 man premades that will tend to get sucked back out if you solo queue.


Quote:
Zileas:
and we are running analysis to figure out if this is inherently more disadvantageous than 5 "average" players at those levels.


I'm much less concerned about whether or not it's advantageous or not to be on the team with the larger skill disparity; I'm more concerned about how much less *fun* it is to be on the team with the much larger skill disparity.

I'm also much more concerned about the perception of matchmaking when it is understood that the matchmaker purposefully sabotages your team if you're highly rated - sure, it's a minor effect for most players, and doesn't make a big difference, but it's still coded into the system - if you are a good player, the system sabotages your team if you solo queue.

This almost certainly isn't going to affect me personally, as I simply don't have the time to make a run at the top in this game. But having played at that level before, it really, really bugs me to see that in the matchmaker.


Quote:
Zileas:
But if you arent in the top .2 or .3% of players, this effect is very small -- if you are a 1600 player, you still get matched with people "around" 1600, so you will only have a little bit of the whole "weaker player on your team" effect.


Right, of course, the closer you are to the mode the denser the population and thus the smaller the range you need to scan to find comparable players; players right at the mode will have a very low tank effect, and it will increase as you get further away up or down the distribution to the thin regions.

I guess there's some consolation to the 1250 ranked player that the amount they get screwed by the matchmaker when solo queuing won't be perceivable, but still.


Comment below rating threshold, click here to show it.

Zileas

VP of Game Design

09-15-2009

Quote:
Savage:
Short version: Outside of the edge cases where the current system falls apart, if I solo queue for all of my games, my W/L ratio will approach 50% as games go to infinity. If I only queue in 5 man premades, my W/L ratio will approach a natural equilibrium percentage that correlates positively with my rating.

I think that most systems "fall apart" on the edge, because there just arent reasonable matches to be made in the time constraint of not wanting to wait several hours. This as you know is also an n-p complete problem, which means that the optimal queue strategy is fairly elusive.

Quote:

Well your model pushes all of the variance in win percentages out of solo queues, who get their natural win percentages tanked by the matchmaker, and into a combination of 5 man premades and the edge cases. As the two overlap it isn't all that surprising that things start to blow up there.

That's not the case. Statistical regressions show that we are adjusting 5 man strength correctly except for elo ratings> 2000. At that point, we just dont know because we have so much heterskedasticity that we have no idea what is happening -- no one solos in that range, so it's impossible to know if solo matching is fair there or not. However, the offset is large enough that its virtually impossible for those teams to do anything BUT play 5 man premades. Therefore, we aren't "stomping" solos with premades. The matching math adds up correctly and gives the proper 50/50 matchings, and doesnt confer "rating advantage" on someone whos in 5 mans a lot in the mid-tie.rs



Quote:

I'm much less concerned about whether or not it's advantageous or not to be on the team with the larger skill disparity; I'm more concerned about how much less *fun* it is to be on the team with the much larger skill disparity.

It's rare now by the ratings, and it will be FAR rarer in release. Right now, you see a lot of new players which are being introduced into the general player pool incorrectly -- that creates most of the perceived skill disparity problems. The particular skill disparity problem that will occur is with people in the top .1 to .2% and there is just straight up no good way to match these people unless they want to wait 30 min to an hour for a match. It would not work by hand either, the best you could do to do better than on the fly matching by this method is organize tournaments with premades at given times.


Quote:

I'm also much more concerned about the perception of matchmaking when it is understood that the matchmaker purposefully sabotages your team if you're highly rated - sure, it's a minor effect for most players, and doesn't make a big difference, but it's still coded into the system - if you are a good player, the system sabotages your team if you solo queue.

This applies in a "material" (i.e. noticable) way to at most .2% of players, less at release, and this 'sabotaging' is generally getting matched with guys like Cefx, Korean, etc who are top 200 players and in the top .5%. We are not assigning you newbies.


Do you have a better idea? Because as we add players, this covers 99.9% of our players very well, and has edge cases that don't function well in ANY matching system just because you cant match when there are no players.


Comment below rating threshold, click here to show it.

Ambi

Member

09-15-2009

I'm sitting on a 110 game 6:5 win:lose ratio right now but I've lost so many games recently from low skilled players. Every single game I've gone into are with low levels such as level 10s in which I end up carrying, how is this possible through this matchmaking system, do I somehow have a really low rating despite a positive win ratio on top of being level 30?.


Comment below rating threshold, click here to show it.

Savage

Senior Member

09-15-2009

Quote:
Zileas:
That's not the case. Statistical regressions show that we are adjusting 5 man strength correctly except for elo ratings> 2000.


Are you using asymmetrical comparable ranges? I.E., if your adjusted ELO rating is 1600, you could be paired against players that are, say, -70 to +100 rating difference from you?

I cannot think of any other way that you could be properly adjusting for 5 man strength and still delivering 50% win rates.


Quote:
Zileas:
It's rare now by the ratings, and it will be FAR rarer in release.


Well depends on what you mean by rare. It will affect about the same number of players, but they'll be a much smaller percentage of the player base.


Quote:
Zileas:
This applies in a "material" (i.e. noticable) way to at most .2% of players, less at release, and this 'sabotaging' is generally getting matched with guys like Cefx, Korean, etc who are top 200 players and in the top .5%. We are not assigning you newbies.


I love you so much right now you don't even understand. I haven't laughed this hard in a long time. You win.


Quote:
Zileas:
Do you have a better idea?


Well I'm curious why you think that having a tanking effect for the better players is preferable to just scrambling the teams randomly, especially when the magnitude of the effect is typically very small.