What is ELO hell?

Comment below rating threshold, click here to show it.

Panzerfaust

Emissary of the League

01-13-2012

It's been explained at least piece-wise, but in simple terms, assuming you belong above your current Elo:

Since you are better than the average player at your Elo, an essentially random team selection of those who are deemed to be, at least currently, at the same Elo as you, will result in two teams that if you were judging on Elo alone, would be quite even. Since you are actually stronger than your Elo, you give your team an edge to win; being the better player doesn't always mean you will win. Players have good days and bad days, players get flat out unlucky when an essentially random choice leads to their downfall, and players overestimate their ability more often than not.

From wiki, the law of large numbers "...is a theorem that describes the result of performing the same experiment a large number of times. According to the law, the average of the results obtained from a large number of trials should be close to the expected value, and will tend to become closer as more trials are performed."

This means that when you flip a coin once, you're not going to expect a 50/50 result. Twice, while you could get a 50/50 result, it's not necessarily likely. Flip a coin 100 times, and you would expect close to 50 heads. Flip a coin 100000 times, and you would expect the percentage to get even closer to 50%. It's impossible to accurately guess a single coin flip, but when you're dealing with exceptionally large numbers, you can be reasonably sure that you can predict, within a minor margin of error, the ratio. Similarly, when dealing with Elo, you will never find your 'proper' Elo after only a game or two, but as you play more games your Elo gets more accurate. Of course Elo calculations are more complicated than flipping a coin; even ignoring the other players, your skill level (~chance of flipping heads) is constantly fluctuating.

In short, it's pretty much the best solution to an impossible problem. See the NP-Complete problem set if you want to see some even more annoying problems.


Comment below rating threshold, click here to show it.

TwistWrist

Senior Member

01-13-2012

Oi. Talking about NP-Complete problems. Going to fry someone's mind.


Comment below rating threshold, click here to show it.

Critkeeper

Senior Member

01-20-2012

I am an undergraduate physics student attending a small college in Tennessee. Part of our studies involve analyzing the effect of random variables in a system with well defined deterministic behavior, and is called a stochastic model. The league of legends system can be interpreted as a stochastic model, in which the following deterministic rules are present:

1. For each game there are a finite number of players, selected from one of many finite pools of players, at random within the pool, but not a random pool.
2. Each game results in every player on a team moving to another pool.

Here pool isn't synonymous with ELO, but its close. All players in a team are supposed to have comparable elo, and whatever range the server interprets as "comparable" (say from 1000 to 1200 elo) has associated with it a finite number of players possessing that ELO. This is a pool of players, and in any game the system won't select players from two different pools. Of course, the anecdotal static range (1000 to 1200) is not how pools are actually represented to the system, but the actual range of what is "comparable ELO" varies over time and changes with the community, however such a thing as a "pool" does exist, otherwise players would be selected at random from all over the ENTIRE ELO range, rather than a subset of that range.

As previously stated in other posts, a bellcurve will represent the random distribution of player skill across the entire system. This means that pool size must be larger in the hump of the bell, where more players exists with comparable elo; also, pool size must be smaller on the fringes of the bell curve, where exceptionally bad players are rare and exceptionally good players are rare. Pool size isn't naturally limited to a specific maximum number or minimum number of players, or else there would not be many pools above 1600 elo, and thus the chance that a few of your teamates with, say 300 ELO lower than you and 140 ELO lower than you (with 1900) would be so great as to make your games unpleasant. While the system does pair higher and lower ELO players together in order to balance out the team, it only selects randomly from players that are waiting in que, from players that are online, from players in the current pool. Contrary, however, if the pool size were large then would always be players to be paired with, but potentially at the player's expense (if the game is lost due to inexperience/ lack of skill of the one or two with comparably low ELO). This means that pool size isn't just a static value; it varies depending on the number of players online and waiting in que. Very long que's (5 or 6 minutes) are just RIOT's way of ensuring the game you play will be with players of similar ELO. Players with extremely low or extremely high ELO will have the longest que.
Sometimes the system will abandon rule (1.) and select players from different pools if the que time would be far too long (8 or 9 minutes).

Rule (2.) above is certainly valid for a player who is new to ranked matches, because the ELO of that player is so drastically modified match after match. However, eventually the ELO of the player begins to stabilize and cannot be drastically modified after just a single match. This is the key that will cause the stochastic behavior of the system to tend towards something called Brownian Motion, which colloquially is just ELO hovering over a final score that most people interepret as their "true ELO", sometimes going over, sometimes going under, but always returning to the range in which the motion is predominant.

Now i can explain why and how ELO hell exists.

Assume we have a player new to ranked matches, called player A. Player A begins his ranked match career and loses his first game. Little does he know that his first few games in ranked matching are far more important than any of his others because of the details covered in the above with respect to diminishing changes in ELO over time. So player A doesn't take his ranked matches seriously and loses a few more, dropping his ELO to about 800. At this point, one would think that the pool size is smaller and the player will not be forced to play over such a wide range of elo, and that hopefully he can play with players that have much more comparable elo (the pool size is smaller because the more by which his ELO is different from the median 1200 ELO, the less players there are with comparable ELO). BUT now assume that he blows his chances and plays terribly, mostly his fault. Player A isn't a bad player, just a very inconsistent one, he could probably do better, perhaps he has 12-17 long winning streaks from time to time, and 12-17 long losing streaks from time to time. After player A reaches the point at which there are so few available players in pool that players from beneath his pool must be selected and put on his team, or else wait 10 minutes in que for other players of comparable ELO to click play, he begins his journey into ELO hell. At this point the pool size continues to get smaller because there are fewer people to select from, and therefore rule (1.) has to be broken. What this does is introduce more randomness into the system. Combine with diminishing elo changes over time and you have brownian motion.


Comment below rating threshold, click here to show it.

TwistWrist

Senior Member

01-20-2012

Cute explanation. But largely meaningless and inaccurate.

Some massive assumptions though.

Quote:
After player A reaches the point at which there are so few available players in pool that players from beneath his pool must be selected and put on his team, or else wait 10 minutes in que for other players of comparable ELO to click play, he begins his journey into ELO hell.
This statement basically assumes that at some Elo, the number of players are so small that large Elo difference have to be used in order to make teams at all. While this statement is probably true at "some" Elo, it is irrelevant to most players. Why? Because most players are not in an extreme end of the Elo. They lumped in what they consider "Elo hell" with plenty of other players. 75% percent of the play population is under 1250 Elo. You'll probably have plenty of company where ever you are.

However, let's go with his assumption that player A in this extremel land of giant Elo gaps. Let's go with the next assumption that makes absolutely no sense.

Quote:
At this point the pool size continues to get smaller because there are fewer people to select from, and therefore rule (1.) has to be broken. What this does is introduce more randomness into the system.
Sure, if matchmaking merely selected 10 people from a pool and made teams out of them at random.

But that would be a bad assumption. Even in this magical land of giant Elo gaps, why would you assume that? Are you trying to say for example, matchmaking will take a two 300-Elo and three 400-Elo players pair them against a five 700-Elo players? Because that assumption is nonsensical.

The only reasonable alternative would be that matchmaking tried to make even teams with uneven players...like it always had. Wait...if the teams are somewhat even...what would that mean? Hmmm...maybe your assumption is baseless?

Also:

Quote:
Combine with diminishing elo changes over time and you have brownian motion.
Elo changes do not diminish based on time. Elo changes pretty much drops down to around 11-12 points per game. Except for rare cases of large Elo gaps. It is like the system was designed even for the land of large Elo gaps that most players do not experience.

More importantly, you have not actually proven that someone who should be at a higher Elo will be stuck at a lower Elo. You've merely stated that at some extreme Elo, the Elo gaps might become bigger. Without actually saying why the teams will be so badly constructed that a superior player cannot work his or her way out.

You've merely stated at some Elo that everything is completely random. And because you said that at some Elo, everything is completely random, those everything is completely random. I would hope as someone who is supposedly trained in the sciences would recognize that your line of reasoning is largely fallacious.

I would also hope that you'll recognize that disproving power of a counter-example.

http://na.leagueoflegends.com/board/....php?t=1498738

Guy goes from -18 Elo to 1520. Clearly, even the land of extreme Elo gaps is not purely random enough to keep him down. Because it wasn't purely random to start with.


Comment below rating threshold, click here to show it.

Bezant Gebal

Senior Member

01-25-2012

Elo hell is when your Yi is asked pregame where he's going, and he asks "what did I call?" and no one really cares if he said it or not, pregame can be spammy. He's given top anyhow like he wanted, but because no one answered what he called, he announces he will afk. He then offers pithy chat throughout the game saying things like he sure is glad he didn't waste his time with this game (despite the clear irony).


Comment below rating threshold, click here to show it.

syn3rgyz

Member

01-27-2012

Quote:
Originally Posted by Panzerfaust View Post
It's been explained at least piece-wise, but in simple terms, assuming you belong above your current Elo:

Since you are better than the average player at your Elo, an essentially random team selection of those who are deemed to be, at least currently, at the same Elo as you, will result in two teams that if you were judging on Elo alone, would be quite even. Since you are actually stronger than your Elo, you give your team an edge to win; being the better player doesn't always mean you will win. Players have good days and bad days, players get flat out unlucky when an essentially random choice leads to their downfall, and players overestimate their ability more often than not.

From wiki, the law of large numbers "...is a theorem that describes the result of performing the same experiment a large number of times. According to the law, the average of the results obtained from a large number of trials should be close to the expected value, and will tend to become closer as more trials are performed."

This means that when you flip a coin once, you're not going to expect a 50/50 result. Twice, while you could get a 50/50 result, it's not necessarily likely. Flip a coin 100 times, and you would expect close to 50 heads. Flip a coin 100000 times, and you would expect the percentage to get even closer to 50%. It's impossible to accurately guess a single coin flip, but when you're dealing with exceptionally large numbers, you can be reasonably sure that you can predict, within a minor margin of error, the ratio. Similarly, when dealing with Elo, you will never find your 'proper' Elo after only a game or two, but as you play more games your Elo gets more accurate. Of course Elo calculations are more complicated than flipping a coin; even ignoring the other players, your skill level (~chance of flipping heads) is constantly fluctuating.

In short, it's pretty much the best solution to an impossible problem. See the NP-Complete problem set if you want to see some even more annoying problems.
doesn't account for the amount of jungle shen, AP ashe and "I'm last pick but i'm solo top and if i'm not i'll just feed" that I encounter


Comment below rating threshold, click here to show it.

TwistWrist

Senior Member

01-28-2012

Sure it does. It assumes you'll play many more games and that particular game is insignificant. Because the system is based on large numbers. Not one game.

So if your proof of Elo hell starts with a story of a few games, stop. Because it shows ignorance of how Elo works.


Comment below rating threshold, click here to show it.

syn3rgyz

Member

01-28-2012

what if i have played over 150 games and 99% of those i own my lane and my team just feeds or trolls?


Comment below rating threshold, click here to show it.

TwistWrist

Senior Member

01-28-2012

Quote:
Originally Posted by syn3rgyz View Post
what if i have played over 150 games and 99% of those i own my lane and my team just feeds or trolls?
Then, you need to realize LoL is a team game and leave your lane. I seriously doubt you are "own"ing your lane. If you win your lane hard enough, then you should have plenty of time to help your team win the other lanes as well.

Hard for your teammates to lose their lane if you win it for them.


Comment below rating threshold, click here to show it.

Panzerfaust

Emissary of the League

01-28-2012

Quote:
Originally Posted by syn3rgyz View Post
what if i have played over 150 games and 99% of those i own my lane and my team just feeds or trolls?
Two things here. First, you're not going to remember the 'other' 99% of the games, the ones where the other team had feeders/trolls/etc as regularly; it's just part of human nature.

Second, realize that winning your lane doesn't truly mean anything about your skill. It's entirely possible that while you may be a great early game player, you lack the map awareness or late game skills required to make use of the advantage you gained early. There are also plenty of match-ups that are in your favor; getting more minions/creeps doesn't mean you're winning such a lane, you want to completely dominate that lane.