Matchmaking working as intended.

First Riot Post
Comment below rating threshold, click here to show it.

Zileas

VP of Game Design

02-07-2014
2 of 5 Riot Posts

Quote:
Originally Posted by JayArrrr View Post
On AVERAGE, I get 2.54 more kills a game, while dieing 3.7 LESS times per game, and I average 146.88 more CS PER GAME than this team.

If this isn't enough proof of a broken matchmaking system, then you are blind.
You know, you control 20% of the power of your team, and 0% of the power of the enemy team, so if you are above average vs your teammates today, with the way MMR works, you will eventually have teammates of equal skill... That's a mathematical certainty.

Of course, it could be that you are already being paired with equal-skill teammates on average -- it could be that your higher stats vs teammates are just that you are a stronger solo player than them, and a worse team player (who, for example, blames them on the forums in a long thread, or over-prioritizes CS over joining emerging team fights), which averages you out to the same overall ability. The matchmaking system acts on game outcomes, which you have control over.

It could also be that you play a role that tends to get higher CS and KDA (e.g. NOT support, and especially ADC).


Comment below rating threshold, click here to show it.

GladiatorBigwood

Junior Member

02-07-2014

Quote:
worse team player (who, for example, blames them on the forums in a long thread)
Yuuuuuuuuuuuup


Comment below rating threshold, click here to show it.

Super Explosion

Senior Member

02-07-2014

Quote:
Originally Posted by Zileas View Post
you will eventually have teammates of equal skill... That's a mathematical certainty.
Champion variance.

Snowball variance.


Comment below rating threshold, click here to show it.

Zileas

VP of Game Design

02-07-2014
3 of 5 Riot Posts

Quote:
Originally Posted by Super Explosion View Post
Champion variance.

Snowball variance.
Which over several hundred games wash out to a very small effect.


Comment below rating threshold, click here to show it.

Super Explosion

Senior Member

02-07-2014

Quote:
Originally Posted by Zileas View Post
Which over several hundred games wash out to a very small effect.
No, it won't, because you will get players of roughly equivalent skill, not "equal skill".

This means that the average does not describe the potential hills and valleys per game.

And that the true 'quality' of the match is magnified by the game's mechanical balance.

So if Kha'Zix is OP, and I decide to pick Kha'Zix this game, and then I get first blood, well...


Comment below rating threshold, click here to show it.

Zileas

VP of Game Design

02-08-2014
4 of 5 Riot Posts

Quote:
Originally Posted by Super Explosion View Post
No, it won't, because you will get players of roughly equivalent skill, not "equal skill".

This means that the average does not describe the potential hills and valleys per game.

And that the true 'quality' of the match is magnified by the game's mechanical balance.

So if Kha'Zix is OP, and I decide to pick Kha'Zix this game, and then I get first blood, well...
Oh. You are talking about how when LoL is played by players of equal skill against each other, you can still get a stomp or extreme snowball. Sure... that can happen, though, in that case it's not a matchmaking issue, it's just natural variance in execution and circumstances.


Comment below rating threshold, click here to show it.

Super Explosion

Senior Member

02-08-2014

Quote:
Originally Posted by Zileas View Post
Oh. You are talking about how when LoL is played by players of equal skill against each other, you can still get a stomp or extreme snowball. Sure... that can happen, though, in that case it's not a matchmaking issue, it's just natural variance in execution and circumstances.
Certainly not. There are perhaps two dimensions to it:

Players can change their effective MMR per game by playing different champions. This can vary wildly.

A subset of this is the 'one trick pony' effect of someone carrying to higher MMR on a single champion, in which if they ever don't get that champion, or a balance change affects that champion, their 'true skill' is drastically altered.

Will this even out? Only roughly. Stating players are "equal skill" is factually incorrect. There is no supporting metric for exactitude.

So ostensibly, if a player picked alternating champions where that player is highly skilled with one, and not highly skilled with the other, the results would appear as a single average MMR, despite per game the team contribution swinging wildly.

As an extreme example, having your games alternate half and half between shut out wins and stomped losses, you'll be at a fifty percent win rate, but still likely not having a good time.

This can be further compounded by the restrictiveness of the laning structure-- if you set yourself up as the 'master of mid lane', but there's a ton of competition for mid lane, your skill per game might take much more variance.

Secondly, the hinge of mechanical balance:

If your champion picks contribute too largely to the outcome of the game, then MMR is largely testing your skill at picking champions, not at actually playing the game.

If the laning snowball can quickly create an 'unassailable' position by midgame, then your skill perhaps only becomes applicable if you personally enabled that snowball.


Comment below rating threshold, click here to show it.

JayArrrr

Senior Member

02-08-2014

Quote:
Originally Posted by Zileas View Post
Step 1) Identify semi-rare matchmaking edge case (pre-made of people with level distribution from 6 to 30), and critique what appears to be a reasonably good attempt at offsetting it.
Step 2) Become disappointed at lack of community interest in this edge case.
Step 3) Post hourly thread bumps in an attempt to generate interest, and also criticize your teammates.
Step 4) ????
Step 5) Profit
*(Is ???? = "zileas responds"? not sure)

Sorry I could not resist. Given your persistence, I do have a real answer --

Matchmaking depends upon a large pool of players in similar situations to match quickly and effectively. When edge cases like this are introduced, it has less options, unless it wants to leave people hanging for a very long time. We've made the decision to not leave people waiting forever, and to take an "ok" match in favor of shorter times (and supporting people playing with their friends).

While I'm sure we could improve matches like this (we always can improve matchmaking and are doing so continually), I think this one is pretty good in terms of balancing 'not waiting 10 minutes' vs 'plausible match'. Note that if said premade plays together a lot and is actually winning often (not a given), their normal elo will raise very quickly and they'll soon be facing extremely difficult opponents who will beat them.

Sorry this came out of a bad experience you had -- and yes, sometimes the matchmaker will give you a 'meh' match (see below). But I really don't think this is an egregious matchmaking error. For what it's worth, we do want to increase the 4 man vs 4 man matching frequency, but matching a spread-out 4 man like that in a way that is 'obviously correct' is really hard unless you can find a more or less identical 4-man in a short space of time -- which is not a given.

Lastly, as long as we have millions of players playing every day, we will have some number of 'weird' matches like this. I know for a fact the severity and frequency has greatly decreased with time. We do continue to improve, but even if we invested years more into this, you guys would still see results that could be debated pretty frequently, just because even with millions of players, the system is just not going to be able to match edge cases against identical edge cases in a quick time -- we aren't willing to make players wait an hour for a perfect match, and that means every day there will be examples of 'not ideal' matches, though they will constitute a tiny % of matches.
Thank you for the reply. Is there any way that there could be a checkbox somewhere when going to queue, or creating an account, that says we understand the length of time between queues will take longer, but we opt-in to a different queue that will make games more evenly balanced, even if the wait-time is much longer?


Comment below rating threshold, click here to show it.

JayArrrr

Senior Member

02-08-2014

Quote:
Originally Posted by Zileas View Post
You know, you control 20% of the power of your team, and 0% of the power of the enemy team, so if you are above average vs your teammates today, with the way MMR works, you will eventually have teammates of equal skill... That's a mathematical certainty.

Of course, it could be that you are already being paired with equal-skill teammates on average -- it could be that your higher stats vs teammates are just that you are a stronger solo player than them, and a worse team player (who, for example, blames them on the forums in a long thread, or over-prioritizes CS over joining emerging team fights), which averages you out to the same overall ability. The matchmaking system acts on game outcomes, which you have control over.

It could also be that you play a role that tends to get higher CS and KDA (e.g. NOT support, and especially ADC).
Here's the thing with this post though.

Fist off, I believe individuals have more affect on the power percentage of the enemy team then you are letting on. A good player will roam out of his lane early, and often when his enemy is missing, and disrupt other lanes as well. Also, how long does it take before the MMR evens out so you begin getting players of your own skill? Is there anything in the works (especially with the new teambuilder coming soon hopefully) that will include individual statistics in the matchmaking process, such as CS and/or KDA when building teams? This should be easy to implement if people are queueing for specific roles. Have a certain threshold in each category that puts them in a specific MMR. For example, X amount of CS is equivalent to a bronze 5 MMR, and will be matched up with a support who's X amount of assists is equivalent to a bronze V player, ect.

As far as teamfights go, everyone knows that in low elo, regardless of how many times you ask in chat, regardless of how many times you ping the minimap, inevitably, there are a lot of people who are extremely bad att he game and will continue to refuse to group up in teamfights, resulting in losses. This is the main thing I am trying to avoid, players who play decently should NEVER be forced to play with people like that. It's very sad that you have stated that MMR is ONLY based on outcome. Doing research on my fellow teammates and enemies alike shows there is a significant difference in players skill levels that is very obvious when looking at the statistics of a loss or win. For example, the person who loses and goes 8/2/5 with 325 CS in a 43 minute game versus his teammates, who go 2/14/3 with 120 CS. Why do these two players get the same MMR reduction? This isnt' right, and doesn't accurately reflect the skill levels of the players involved, and is largely the reason behind these mismatches, I believe.

Shoutcasters don't talk about statistics in the LCS when showing who the best players of the region are for no reason, it is an accurate reflection of how skilled the player is, and should be taken into affect when trying to match players of equal skill, there is no arguing that point.


Comment below rating threshold, click here to show it.

JayArrrr

Senior Member

02-08-2014

Quote:
Originally Posted by JayArrrr View Post
Here's the thing with this post though.

Fist off, I believe individuals have more affect on the power percentage of the enemy team then you are letting on. A good player will roam out of his lane early, and often when his enemy is missing, and disrupt other lanes as well. Also, how long does it take before the MMR evens out so you begin getting players of your own skill? Is there anything in the works (especially with the new teambuilder coming soon hopefully) that will include individual statistics in the matchmaking process, such as CS and/or KDA when building teams? This should be easy to implement if people are queueing for specific roles. Have a certain threshold in each category that puts them in a specific MMR. For example, X amount of CS is equivalent to a bronze 5 MMR, and will be matched up with a support who's X amount of assists is equivalent to a bronze V player, ect.

As far as teamfights go, everyone knows that in low elo, regardless of how many times you ask in chat, regardless of how many times you ping the minimap, inevitably, there are a lot of people who are extremely bad att he game and will continue to refuse to group up in teamfights, resulting in losses. This is the main thing I am trying to avoid, players who play decently should NEVER be forced to play with people like that. It's very sad that you have stated that MMR is ONLY based on outcome. Doing research on my fellow teammates and enemies alike shows there is a significant difference in players skill levels that is very obvious when looking at the statistics of a loss or win. For example, the person who loses and goes 8/2/5 with 325 CS in a 43 minute game versus his teammates, who go 2/14/3 with 120 CS. Why do these two players get the same MMR reduction? This isnt' right, and doesn't accurately reflect the skill levels of the players involved, and is largely the reason behind these mismatches, I believe.

Shoutcasters don't talk about statistics in the LCS when showing who the best players of the region are for no reason, it is an accurate reflection of how skilled the player is, and should be taken into affect when trying to match players of equal skill, there is no arguing that point.
j