29 replies to this topic

[ShinVector]

Void Angel, on 30 May 2015 - 09:43 PM, said:

No, it's a total myth. It's a common misconception that Elo (not an acronym) tries to move you to a 1/1 win/loss ratio.

Mazzyplz, on 30 May 2015 - 09:49 PM, said:

that sounds about right.

well it seems my idea was wrong, but the effects are similar some of the time at least

I believe that 1:1 win loss ratio thing was mentioned by PGI themselves... I just can't find the evidence where they said that.

Secondly ELO MM was meant to be used in 1v1 scenario ! If it was 1v1 only for MWO then I believe we would have had a very different experience as in fair match ups.

The problem with ELO MM in MWO is that it is that it is being simplified by using a Team Average ELO score this is probably the ELO hell some people are experiencing.
Lastly the base ELO for new players was set too high I believe. 1300 or 1400 or something like that..
Maybe it should have started from 0.. I dunno...

Thats my thoughts on the matter anyway.

-

Edit:
Russ talks about Elo in the town hall meeting yesterday

Sooo here is some evidence from April town hall.. Russ doesn't want ELO to be just about 'win/loss'.
He even talks about ELO Hell...

Edited by ShinVector, 31 May 2015 - 02:29 AM.

[Ironwithin]

Nothing to add to the discussion but: Elo, name, not an acronym, as Void Angel already pointed out.

[ShinVector]

An example of ELO Hell from today and beating it.. LOL..

Edited by ShinVector, 31 May 2015 - 05:18 AM.

[Void Angel]

As a point of order: this is not really Elo Hell. Elo Hell is a specific phenomenon where players who fall below a certain threshold of Elo ratings find that their personal performance seems inconsequential in determining wins and losses - i.e. victory is determined more by how many rage quitters and "top or I feed" goons show up than by how well the player contributes. What we have here is the matchmaker giving us games that feel lopsided because of the challenge. Remember, the crappy performance of your team may have far less to do with their Elo ratings than it has to do with them all screwing the team by trying to hide behind each other because there's a challenge going on - and they're afraid of dying before they get their points.

ShinVector, on 31 May 2015 - 02:20 AM, said:

Edit:
Russ talks about Elo in the town hall meeting yesterday

Sooo here is some evidence from April town hall.. Russ doesn't want ELO to be just about 'win/loss'.
He even talks about ELO Hell...

What Russ talking about here is tying Elo to an expected match score, rather than an expected victory result. This would make Elo much less susceptible to the luck of the draw when you have a bad team - essentially you're responsible for your team's performance to a much lesser degree, because the game doesn't care about the win, and it's looking for what your match score should be. At a first look, this is a great idea, and a solution to the team Elo problem - to the extent that it exists and/or is solveable. Any ranking system in a team game will always be affected by factors beyond our personal control (i.e. other players,) and given that this is the case, the mathematical foundations of Elo are still sound for MWO. Tying it to match score instead of winning, however, seems like it would significantly reduce confounding variables and make the system more responsive.

At any rate, Elo (including PGI's implementation) still does not care about your kdr. PGI personnel have talked about the "ideal" match that the Elo system is theoretically moving players toward as averaging a 50% win rate over time, but that's not the same thing. Hypothetically the Elo system, given enough time, and a player who never improves or gets rusty, will eventually - given enough players in the queue - generate matches that will, over time, result in a 50% win-loss rate for that player. Whether, given all of those qualifications, this blessed state will be achieved before the heat death of the universe (to say nothing of the sun going nova) is more than I can guess (and I left out different game modes and a shared group/solo Elo rating.)

[ShinVector]

Void Angel, on 31 May 2015 - 12:34 PM, said:

What Russ talking about here is tying Elo to an expected match score, rather than an expected victory result. This would make Elo much less susceptible to the luck of the draw when you have a bad team - essentially you're responsible for your team's performance to a much lesser degree, because the game doesn't care about the win, and it's looking for what your match score should be. At a first look, this is a great idea, and a solution to the team Elo problem - to the extent that it exists and/or is solveable. Any ranking system in a team game will always be affected by factors beyond our personal control (i.e. other players,) and given that this is the case, the mathematical foundations of Elo are still sound for MWO. Tying it to match score instead of winning, however, seems like it would significantly reduce confounding variables and make the system more responsive.

He clearly spoke about the possibility of how a player can get their ELO skewed by playing in a good group all the time and it would possible for that person to reach the max ELO of 2800 just because he plays his role and gets carried all the time.

If he plays solo that's where his ELO 2800 destroys him as ELO MM expects him to carry matches.

This is also known as ELO hell that many have faced before.

Anyway whats Russ says that is actually interesting:
1. They might seperate SOLO ELO and Group ELO.
2. They might adjust the ELO formula.
3. They might reset ELO.

Edited by ShinVector, 31 May 2015 - 07:45 PM.

[Ironwithin]

Again: Elo, not ELO.
Just sayin'.

[Heuvadoches]

Void Angel, on 30 May 2015 - 09:43 PM, said:

No, it's a total myth. It's a common misconception that Elo (not an acronym) tries to move you to a 1/1 win/loss ratio. That's not correct - Elo doesn't care if you win or lose. It does not care what your win/loss ratio is. What Elo cares about is whether or not you are winning as much as it thinks you're supposed to win. So if you win a match against total scrubs, your rating would move very little - if at all. And Elo is fine with that. Elo only adjusts your rating significantly (if at all, depending on the system) if you're winning or losing more than it thinks you should. Otherwise, if you're at the 75% of the player base, and getting matched with a random distribution of players, you would be expected to win 75% of the time, overall - math people will notice that I'm drastically simplifying, but this is accurate as far as it goes. The system works well enough with the middle range of Elo ratings, but as you raise yourself higher and higher in the rankings the pool of people you have to play against that are near your own rating gets smaller... causing longer wait times.

This is where teams come into play. Because the upper ranges of Elo rankings were simply unable to get matches, the Devs had to tinker with release valves in the acceptable range of ratings. So when you get matched up, you can have people who are dramatically more or less skilled than some other players - and they're only matched based on the teams average Elo. Thus, you can have teams where very high-Elo players are matched with total newbies and placed against a team of solidly middle-range enemies. Sometimes, this works out well for the high-range players, and sometimes not.

The trouble is - and here's where the myth comes from - that it's hard to really tell where your teammates actually belong on the Elo scale. First off, the difficulty curve of MWO looks more like a plotting of the Mandelbrot Set - in 3D. There are many skill sets in MWO that players who have favored one style of play may be lacking on other skills: a dedicated gauss sniper may not brawl well; a light pilot may be great at dismembering larger opponents, but have a more difficult time dogfighting other lights. People may be playing new chassis, trying new playstyles, or maybe they just LRM-boated themselves into their current Elo rating and are now trying to adjust to using real weapons.

Add in people's tendency to remember extraordinary events, and you have the makings of the myth that Elo will punish you for "winning too much." Players remember when they do 1000 damage and still lose the match; they don't remember the times when they went head to head with the enemy team and had a close match quite so well - even though both of those results could come from lopsided teams. Then there's cumulative focus fire, player mistakes, etc, etc. When players perform poorly on our team, we remember it, and it's easy to attribute it to low Elo. And when so many players say "Elo is based on KDR" as shorthand for what Elo actually does, the total fabrication that Elo will match you with bad players to bring your KDR down seems plausible - instead of the errant fantasy that it actually is.

These kinds of matches don't happen often in the solo queue. But when they happen, we remember them - and they do tend to happen to higher Elo players more frequently.

Whatever ... you must not be playing then. *ignores*

[Void Angel]

I'm sorry you don't understand the Elo system or the matchmaker, and are uninterested in learning.

Edited by Void Angel, 31 May 2015 - 08:43 PM.

[Void Angel]

ShinVector, on 31 May 2015 - 07:39 PM, said:

He clearly spoke about the possibility of how a player can get their ELO skewed by playing in a good group all the time and it would possible for that person to reach the max ELO of 2800 just because he plays his role and gets carried all the time.

If he plays solo that's where his ELO 2800 destroys him as ELO MM expects him to carry matches.

This is also known as ELO hell that many have faced before.

It's not that the matchmaker expects him to carry a match - it's that the matchmaker couldn't find any teammates near his level and so put him in what it knew was a bad team matchup because it considers a lopsided team to be superior to never generating a match at all.

Isn't really Elo Hell, though people do indeed use that term to refer to it. Elo Hell is a specific thing which predates MWO - it's important to be clear on definitions, because well... that's how things like the matchmaker myth get started.

In any case, I think making the Elo (still not an acronym) system pay attention to match scores is at least worth trying. If it only works as well as the current implementation, we haven't lost much - and if it works better, all the good! Decoupling group and solo Elo sounds like a good plan as well; the queues are dramatically different environments, and Elo in one doesn't necessarily reflect Elo in the other.

Edited by Void Angel, 31 May 2015 - 08:53 PM.

[ShinVector]

Void Angel, on 31 May 2015 - 08:52 PM, said:

It's not that the matchmaker expects him to carry a match - it's that the matchmaker couldn't find any teammates near his level and so put him in what it knew was a bad team matchup because it considers a lopsided team to be superior to never generating a match at all.

Matter of perspective.. People usually want to win.. TROLLOL MM..
Time will tell, I guess on whatever PGI is up too..

Edited by ShinVector, 31 May 2015 - 09:02 PM.