202 replies to this topic

[Abivard]

Roadkill, on 07 April 2014 - 03:41 PM, said:

It isn't giving false results. The new systems arrive at the same valid ratings more quickly than Elo. And that is why they're being researched. Not because Elo doesn't work, but because the new systems work faster.

I've never claimed that Elo is perfect. It is possible to improve upon Elo.

You, however, have claimed that Elo doesn't work. You are wrong. Elo does work in this environment. It's mathematically proven to work.

I have never claimed Elo does not work for what it was designed for.

None of your links lead to anything more than editorials that espoused your point of view with no collaborating facts. or if they were articles that cited reputable sources they actually supported my arguments, when I pointed that out back then you so conveniently ignored that as you do now.

These OTHER systems do not simply speed up the process, they refine and EXPAND upon, because the system as it was, was to inaccurate even for what it was designed for let alone able to be cut and pasted into anything under the sun.

Pearls before swine.
Keep living on your island of 'Everything working fine, move along now'

[Roland]

Macbrea, on 07 April 2014 - 03:46 PM, said:

You are as bad as the other 23 players in your match. Their teams average elo is close to your teams average elo. And almost always, your elo is close to your teams. I am not sure why people feel they are super special and should get their own bracket.

This isn't actually true. The only stated aspect of the matchmaking according to pgi is that the average elo of the two teams is close to each other.

There is no guarantee that you elo will be close, at all, to all the other people on your team, or the enemy team.

Edited by Roland, 07 April 2014 - 04:00 PM.

[warner2]

Empirically I think Elo does work and that it will only get better as (a) population count rises (let's be positive and hope that happens) and ( PGI introduce the bracketing system so the team has an average Elo but in a bracket which should stop those cases where a really good player is balanced by a really bad player. Empirical reasons for why I think Elo work include:

1. If I group up with a player or players that I know are really good I can see the difference in the players I go against - they are clearly the best players playing. If I group up with players that I know are really bad I can see the difference in the opposite direction - they are clearly very bad players in the matches. This is on average of course, observed over many months.
2. For a long time I played only mediums and heavies, so when I got an assault and later a light (especially the light since it was the last class I got) I noticed straight away that I was at the beginner Elo in those classes as opposed to the classes I usually played in where my Elo had moved.

I think the game is better with Elo than without.

[Abivard]

Roadkill, on 07 April 2014 - 03:34 PM, said:

I think you participated in that "other" thread as well.

It's not an impossibly large number of games. It's no more than 200-300, and may be as few as 50 depending on how confident you want your ratings to be.

But perhaps more to the point, the rating variability introduced by multiplayer Elo is entirely washed out in the noise caused by PGI's matchmaker, which itself is derived from the small player base. When your matchmaker is starting with a window of +/- 175 points and rapidly expands its search to +/- 1000 points, that tells you everything you need to know.

It's not Elo. It's the matchmaker.

Roland, on 07 April 2014 - 03:47 PM, said:

I don't believe you have any mathematical basis for this statement, at all.

If you believe otherwise, please show your math, but I seriously doubt that you are considering the obfuscating factors at hand.

Roland, you have hit the nail on the head when it comes to Roadkill and Elo.

This constant confusion on his part between MM and Elo isn't helping either.

[Targetloc]

Those who rage hardest against Elo and froth at the mere mention of math that can't possibly work because of factors and variables will never consider that no matter how much you think you know about statistics, there's a pretty decent chance that someone else could know more about it than you do.

Amusingly, they're also the same ones that are dead certain they are way better at robots than those other noobs the game keeps forcing on their team to make them lose.

It's an interesting effect. Maybe some cool dudes could do a study on it and get it named after them.

[warner2]

Roland, on 07 April 2014 - 04:00 PM, said:

This isn't actually true. The only stated aspect of the matchmaking according to pgi is that the average elo of the two teams is close to each other.

There is no guarantee that you elo will be close, at all, to all the other people on your team, or the enemy team.

True there is no guarantee but it tries to match within a tolerance first. I blame player population for it widening that tolerance, and perhaps not enough really good players around (or really good players forcing MM to pull in low Elo to balance). I think the 3 Elo brackets they will introduce soon will help here as then at least the team is drawn from a broad pool,

[Roland]

Roadkill, on 07 April 2014 - 03:51 PM, said:

Not getting into another one of these threads.

TL;DR Elo works fine in multiplayer environments. Abivard is convinced otherwise despite multiple research papers that have been presented to him in the past, so there's no point in arguing with him.

As I pointed out, you don't really have any actual math to back up your claims here. You made up numbers, that had no basis in reality.

Not that I blame you. It would be incredibly hard to mathematically demonstrate how much different factors had on the outcome of a game, but you suggestion that you know the number of matches required is only a few hundred is totally baseless. You don't have any proof at all for that statement.

[Targetloc]

Roland, on 07 April 2014 - 04:05 PM, said:

As I pointed out, you don't really have any actual math to back up your claims here. You made up numbers, that had no basis in reality.

Not that I blame you. It would be incredibly hard to mathematically demonstrate how much different factors had on the outcome of a game, but you suggestion that you know the number of matches required is only a few hundred is totally baseless. You don't have any proof at all for that statement.

I'd be interested in trying to figure that out.

I see 2 possible scenarios.

You have a population of N players of known skill levels L, what's the average number of games needed to get them within K rating of their actual rank. (assuming the better team always wins). From there we could start messing with the % chance the better team has to win to see how much longer it takes.


The second scenario assumes everyone else in the population has played a near infinite number of games and all their ratings are exactly where they should be. You know from a massive 1v1 ladder tournament that you are explicitly better than every player below 1800 Elo. Assuming teams are randomly assigned from within your fitness range, the team with the highest Elo always wins, what's the average number of games it takes for you to go from 1400 to 1700? Assuming you count as 1800 when deciding win/loss, but your game rank is used for calculating points gained/lost.

Heh. This is going to require some coffee.

[Roland]

That second assumption you lay out there is one of the major contributors to there being such an immense amount of noise in the signal... Because not only is YOUR rating garbage to begin with, but everyone ELSE'S rating is also garbage. And since all of your actions are obfuscating each others' impact on the overall match results, it dramatically increases the number of matches that would be required to actually have the system stabilize to a point where everyone's rating would be representative of their skill level.

And that doesn't even start to take into account all of the other major contributing factors, such as mech builds, tonnage, mech synergy, etc.

Folks just say vague statements like, "Oh, it'll all just balance out in the end..." and while that's certainly true, "the end" could be after a million games. There's no reason to believe that number is small at all.

[Iskareot]

At first I thought the OP was being funny with the post... then I realized he was serious. My comment back would be "Welcome to the issue of not having choice and or a lobby".

Some solo players want to play other solo players and not even see a 4 man lance to cause imbalance. That's the funny part. You are mad because you cant get 8 other good players ..... Im mad because we are forced to fight 4 or 8 man teams as a solo player

[Roland]

Iskareot, on 07 April 2014 - 06:00 PM, said:

At first I thought the OP was being funny with the post... then I realized he was serious. My comment back would be "Welcome to the issue of not having choice and or a lobby".

Some solo players want to play other solo players and not even see a 4 man lance to cause imbalance. That's the funny part. You are mad because you cant get 8 other good players ..... Im mad because we are forced to fight 4 or 8 man teams as a solo player

It's almost as though the best solution would be to have a queue where only solo players could join, and another queue where teams of any size could join, including solo players who wanted to.

[Ordellus]

EgoSlayer, on 06 April 2014 - 11:16 PM, said:

It's clearly false because what you are proposing implies malice and a programming effort to accomplish. The match maker is looking at your Elo score, not your history. It doesn't care if you just lost your 100th match in a row. Your current Elo score is X. If X is higher than the opposing team average you are expected to win. If X is lower you are expected to lose. It really is that simple and there isn't much more programming to it than that. There is much more in the amount your score changes based on this, but in a nut shell if you are higher you are expected to win and lower you are expected to lose.
http://mwomercs.com/...65#entry1626065

If you really think that anyone, PGI especially, devoted significant logic and programming effort to create a malicious match maker to "make sure you lose enough", then no amount of discussion about programming, elo, and logic is going to convince you otherwise. No elo system ever created has been programed to make sure you lose enough. Find one system that is written that way. You say that is what this does, where is your proof?

"It implies malice thus is false" is the worst attempt at a logical argument ive ever seen.

I can't prove that the matchmaker is looking at my W/L ratio, but I do know that it would be very easy to do and would explain all of the problems and symptoms experienced.

P.s. no one claimed "an elo system" was designed to make us lose.....I said a matchmaking system was designed to make me lose sometimes. ..

Again once you have proof I am wrong I will gladly change my mind. ...

Until then I guess you can go with "obvious is obvious" as your arguement

P.s.s its not "an elo system" its "the elo system".....there are variations, but then its no longer the elo system

[Hellen Wheels]

Roland, on 07 April 2014 - 06:04 PM, said:

It's almost as though the best solution would be to have a queue where only solo players could join, and another queue where teams of any size could join, including solo players who wanted to.

There's best solution....

and then there's PGI solution.

Never the twain shall meet.

[Targetloc]

Right. Which is why I'd start with the basic assumptions and see if that's anywhere near 1,000,000 games. Because if it is, there's no point in going further.

But if it's significantly less, like 100 games, then you could work backwards from the formula to find out what the 'noise level' (% chance the higher ranked team will actually win) would have to be for it to take a million games to sort out. Then you're looking at some of what TrueSkill tries to predict, where there's ranks, and a confidence interval for that rank.

[Bhael Fire]

Charons Little Helper, on 05 April 2014 - 03:28 AM, said:

Yes it does - it just takes a lot longer to get to your proper Elo.

This is a fallacy.

I've been playing since before they put Elo in; there is no "proper" Elo when it's based on a win/loss record that is based on the performance of 11 other random people on your team. Period.

[aniviron]

Deathlike, on 07 April 2014 - 01:24 PM, said:

It could be that PGI forgot to initialize the seeding for randomize().

But, I personally cannot get that out of my system.

The game to me at time runs like this:
1) Play map (say Mordor)
2A) Drop into same map, particularly after a loss
2B ) Drop into same map, after a win (especially if it's Mordor)
3) Massively gripe as I'm probably on the same map again
4) Curse the RNG after dropping into same map after dropping into a different map later

It's a theme.

I don't have a sauce, but I seem to remember a PGI employee saying that each server runs just one map; it could be that it tries to balance server load, and fills capacity on one before shifting to another. Matches take about the same amount of time, and so you get stuck in a cycle of being on the same server/map over and over. But that could be me misremembering, too.

[StaIker]

Bhael Fire, on 07 April 2014 - 06:14 PM, said:

This is a fallacy.

I've been playing since before they put Elo in; there is no "proper" Elo when it's based on a win/loss record that is based on the performance of 11 other random people on your team. Period.

1 Great player cannot move a 50/50 result up to a 90/10. That doesn't mean a great player has no influence at all however. It would not be unreasonable to expect a top 1% player to be winning PUGs at 65% or better, even though he is placed with 11 other random folks. For people in the middle of the skill range, which is most folks who play regularly, they are not going to move the needle that much, perhaps just a few points up or down from 50%. That's because there is a huge gap in capability between the "average" player and a player in the top 1%. That guy is worth 2 or 3 or 4 "average" team mates.

[cleghorn6]

Roland, on 07 April 2014 - 06:04 PM, said:

It's almost as though the best solution would be to have a queue where only solo players could join, and another queue where teams of any size could join, including solo players who wanted to.

Which would achieve what, exactly?

That's a serious question. What would splitting out solo and group players do for match making?

[Joseph Mallan]

I don't have a large sample size at all... but my present W/L is .500 and KR is 1.33 Dropping with just one other player... I really really don't understand the Elo complaining!

cleghorn6, on 08 April 2014 - 05:29 AM, said:

Which would achieve what, exactly?

That's a serious question. What would splitting out solo and group players do for match making?

It would take away one of the poor sports excuses...
...
...
...

What, you asked?

[cleghorn6]

Joseph Mallan, on 08 April 2014 - 05:35 AM, said:

It would take away one of the poor sports excuses...

Yeah, that's the only benefit I can see. The mythical "premade stomp" would be a thing of the past. Like it was when they stopped 8s dropping in the PUG queue.

Oh, wait ...