32 replies to this topic

[Davers]

YMMV, but I feel Elo is working for me. Sure I have been in matches where I have stomped the enemy team, and matches I have received the same. But I have also had many matches that were very close. I have also had matches that the scoreboard didn't reflect how close the match was. I have close to 50/50 (387-385) win/loss.

[DeadlyNerd]

Davers, on 14 May 2014 - 09:32 AM, said:

YMMV, but I feel Elo is working for me. Sure I have been in matches where I have stomped the enemy team, and matches I have received the same. But I have also had many matches that were very close. I have also had matches that the scoreboard didn't reflect how close the match was. I have close to 50/50 (387-385) win/loss.

Win/loss ratio is barely relevant. If you lose 12 matches by a stomp and win 12 matches by a stomp, you're still 50/50.
Anyway, back to the topic, because my initial argument is that currently, by using Elo rating, the matchmaker is taking too long. What I want is a junk punch free free for all, where you know you're with randoms against randoms, and where you can hop into a game within seconds.
Waiting 2-3 minutes for a very possible stomp on either side is not exactly fun for vets, and specially not fun for newbies.

Flaming oblivion, on 14 May 2014 - 09:22 AM, said:

I laughed when I read this , Dude there tightening up elo not loosening it , Why ? Because fights aren't balanced and people are leaving because of it, They got to learn ? Lmao , no they don't if they get shafted in to many games there just going to quit or play ALOT less , like most are now because the matchmaker in MWO is nothing more then a troll , this will never happen I thought my previous post would of been end of topic, Its such a bad idea.

I wish I could just ignore this blunder, but you're shooting yourself in the foot. First, they're loosening the criteria, not tightening. If you're so informed, you'd know this.(3/3/3/3 is not a criteria for Elo rating)
Second, they're loosening the matchmaker criteria, not Elo. Elo rating is calculated as it's calculated, there's no such thing as loosening it. Wikipedia is your friend, read about Elo.

Also, if the matchmaker is a troll, why are you supporting it with your post?
And if the matchmaker is such a troll, how's playing without it worse?

Screech, on 14 May 2014 - 09:10 AM, said:

OK it doesn't work great but it is better then only tonnage matching. How many assaults do I need to core in the back without them ever turning around do I need do to prove it? I will take the man over the machine.

What is that going to prove that will help you in this argument? I'm talking about matchmaker taking too long to find a game, not about ego stroking vs newbies. Roids clouding your judgment?

Edited by DeadlyNerd, 14 May 2014 - 12:07 PM.

[Flaming oblivion]

DeadlyNerd, on 14 May 2014 - 12:02 PM, said:

Win/loss ratio is barely relevant. If you lose 12 matches by a stomp and win 12 matches by a stomp, you're still 50/50.
Anyway, back to the topic, because my initial argument is that currently, by using Elo rating, the matchmaker is taking too long. What I want is a junk punch free free for all, where you know you're with randoms against randoms, and where you can hop into a game within seconds.
Waiting 2-3 minutes for a very possible stomp on either side is not exactly fun for vets, and specially not fun for newbies.


I wish I could just ignore this blunder, but you're shooting yourself in the foot. First, they're loosening the criteria, not tightening. If you're so informed, you'd know this.(3/3/3/3 is not a criteria for Elo rating)
Second, they're loosening the matchmaker criteria, not Elo. Elo rating is calculated as it's calculated, there's no such thing as loosening it. Wikipedia is your friend, read about Elo.

Also, if the matchmaker is a troll, why are you supporting it with your post?
And if the matchmaker is such a troll, how's playing without it worse?


What is that going to prove that will help you in this argument? I'm talking about matchmaker taking too long to find a game, not about ego stroking vs newbies. Roids clouding your judgment?

So your not aware of the changes to ELO buckets that are supposed to be happening , tightening the gap . They're not going to edge more towards anyone being able to face anyone that would be the end of MWO in months. I don't support the current matchmaker it needs to be much much stricter , however what your suggesting if I understand correctly would be so so much worse then what we already have , removing pilot skill from the equation when match making , I promise you would be absolutely disastrous for this games long term future.

Edited by Flaming oblivion, 14 May 2014 - 12:51 PM.

[RG Notch]

Flaming oblivion, on 14 May 2014 - 12:48 PM, said:

So your not aware of the changes to ELO buckets that are supposed to be happening , tightening the gap . They're not going to edge more towards anyone being able to face anyone that would be the end of MWO in months. I don't support the current matchmaker it needs to be much much stricter , however what your suggesting if I understand correctly would be so so much worse then what we already have , removing pilot skill from the equation when match making , I promise you would be absolutely disastrous for this games long term future.

You understand that this "tightening" will last until people start complaining that they are waiting too long for matches or you'll discover that the "bucket"s are so large you still end up with vets v. noobs. That's not an Elo problem, it's a low player count problem. All the talk about Elo is silly since it might work well if there were enough players to have a good distribution.

[MischiefSC]

I'd love a pug only queue.

That the OP doesn't understand Elo is alright, it's not uncommon. Math is hard. Which is fortunate, it means I get paid reasonably well for understanding it.

Elo works just fine, especially given the players population. If we had a large enough population you'd want to identify where the average mean was and then use a persons Elo vs the average mean to identify the degree of uncertainty between their current estimated Elo and average, thus allowing you to more swiftly swing the k-factor in adjusting their Elo to more quickly seat them at an accurate score. You could also include variables to their Elo based on their success with specific chassis and loadouts - for example if they're good with AC20s then you modify their Elo for that match by their statistical average improvement in win/loss for matches with AC20s.

Currently though you just need a rough estimate of their skill and if they fall over 1,000 or over 1500 and fit them into their relative bucket.

Elo isn't likely to be causing an issue with wait times. Currently it's matching you in one of three buckets - 0 to 1,000, 1,001 to 1,500, 1501 to 2800. Unless that got backed out of the MM as well?

Regardless though you contribute about 8% to your teams success. That variable can be solved for over a sufficient sample size. Saying that math doesn't work because of the size of teams is flat out silly. It's alright if you don't understand why, it's alright if you're frustrated by the nature of PvP in a game like MW:O with the population that MW:O has and the disparity between variables in player tools (mechs, loadouts, skill) and player population (enough players playing with enough different mechs, loadouts and skill level with them to populate every match with a pair of balanced teams) is why matches can be painfully uneven.

Hence 3/3/3/3 will be awesome - it'll help, within the limited parameters allowed, to balance the mech/loadout factor, which makes fine tuning the skill factor easier.

Elo though is simply a mathematical equation for extracting your influence on the probability of you winning or losing, regardless of your team, in any environment. Mixed or solo is irrelevant save in how it alters required sample size to find results within a viable margin of error.

That's the other part. Margin of error. Elo in MW:O currently has a pretty good sized one - which doesn't matter, because to match players across both skill and weight with the population level it has a rough estimate is just fine. It wouldn't work to rank players from 1 to 100,000 as 'who's the best', but it doesn't need to. Just say 'are you well below average? Around average? Well above average? OKAY! Here you go, have fun, shoot shoot stomp stomp!'

[Screech]

DeadlyNerd, on 14 May 2014 - 12:02 PM, said:

What is that going to prove that will help you in this argument? I'm talking about matchmaker taking too long to find a game, not about ego stroking vs newbies. Roids clouding your judgment?

How is not wanting to get matched up against new players in a PUG only queue ego stroking? Sorry I don't want some solo queue to farm out new players in trial mechs. If there was a solo queue I would want to face people that had significant drops under their belt regardless of the weight of their mech. Sorry I willing to wait longer to have this happen.

And roids? Bro, do you even lift?

[Lykaon]

There are just to many factors that have nothing to do with Elo ratings or comparative skill of players.

We have ECM as a potent game changer yet not accounted for in match balance

We have superior top tier weapons using a superior damage dealing mechanic yet the number of meta humping mech builds per side is not taken into account.

We have weapons sytems that vary widely in effectivness based upon the map.LRMs on Alpine or Tormalaine vs Forest colony or river city?

So even if we had perfect skill and tonnage matching we would still have mech builds meta use and ECM altering the variables.

Solo only queues may even be more vulnerable to these factors.

with 4 player premades at least 4 per team is highly likely to be using competative builds in a coperative manner rather than potluck for 12.

So,equipment and weapon balance is a potent factor as well as skill level.

[DeadlyNerd]

Flaming oblivion, on 14 May 2014 - 12:48 PM, said:

So your not aware of the changes to ELO buckets that are supposed to be happening , tightening the gap . They're not going to edge more towards anyone being able to face anyone that would be the end of MWO in months. I don't support the current matchmaker it needs to be much much stricter , however what your suggesting if I understand correctly would be so so much worse then what we already have , removing pilot skill from the equation when match making , I promise you would be absolutely disastrous for this games long term future.

There's no tightening the gap, the game, and devs explicitly explained this, loosens Elo criteria if the player doesn't get a match for a longer time. You're ending up with newbs anyway.
Also, you're, which is really no surprise, missing the point of this thread, a separate pool for those that don't want to wait.
Talk about long term, you're failing short term(or even history) right here.

PS. There's nothing like "player skill" when a player can get carried from low Elo to high Elo, just by being lucky. Generally individual player skill doesn't matter when considering organized teams. Why do you think a football team comprised of worlds best players would suck? Because there would be absolutely no team play(which is crucial if you want to use Elo for rating).

MischiefSC, on 14 May 2014 - 01:13 PM, said:

I'd love a pug only queue.

That the OP doesn't understand Elo is alright, it's not uncommon. Math is hard. Which is fortunate, it means I get paid reasonably well for understanding it.

Elo works just fine, especially given the players population. If we had a large enough population you'd want to identify where the average mean was and then use a persons Elo vs the average mean to identify the degree of uncertainty between their current estimated Elo and average, thus allowing you to more swiftly swing the k-factor in adjusting their Elo to more quickly seat them at an accurate score. You could also include variables to their Elo based on their success with specific chassis and loadouts - for example if they're good with AC20s then you modify their Elo for that match by their statistical average improvement in win/loss for matches with AC20s.

Currently though you just need a rough estimate of their skill and if they fall over 1,000 or over 1500 and fit them into their relative bucket.

Elo isn't likely to be causing an issue with wait times. Currently it's matching you in one of three buckets - 0 to 1,000, 1,001 to 1,500, 1501 to 2800. Unless that got backed out of the MM as well?

Missing the point completely AND a wall of text? Oh happy day. Aside the fact that Elo doesn't, never did, and never will work for groups of randomly selected players, I'm not talking about Elo rating causing stomps. I'm talking about an Elo free pool, where wait times are barely noticeable.
Derailing is against forum CoC, but I'll humor you this time, because you're quite funny.

Screech, on 14 May 2014 - 01:38 PM, said:

How is not wanting to get matched up against new players in a PUG only queue ego stroking? Sorry I don't want some solo queue to farm out new players in trial mechs. If there was a solo queue I would want to face people that had significant drops under their belt regardless of the weight of their mech. Sorry I willing to wait longer to have this happen.

And roids? Bro, do you even lift?

Because the same way you can stomp newbs, you can get stomped by equally good players that happen to be on the enemy team(geee, who'd thought of that possibility).
Again, nobody would be forcing you to play in that queue. You could stay in your "balanced" queue, waiting 2-3mins to find a "balanced" match for all I care.

You obviously don't, as I can see roids are still in effect.

Lykaon, on 14 May 2014 - 02:01 PM, said:

There are just to many factors that have nothing to do with Elo ratings or comparative skill of players.

I'm not talking about Elo being bad for matchmaker because stomps, I'm talking about Elo being bad for matchmaker because wait times.

Edited by DeadlyNerd, 14 May 2014 - 09:54 PM.

[MischiefSC]

DeadlyNerd, on 14 May 2014 - 09:50 PM, said:

Missing the point completely AND a wall of text? Oh happy day. Aside the fact that Elo doesn't, never did, and never will work for groups of randomly selected players, I'm not talking about Elo rating causing stomps. I'm talking about an Elo free pool, where wait times are barely noticeable.
Derailing is against forum CoC, but I'll humor you this time, because you're quite funny.

So you don't get what Elo is or how math works. That's okay, a lot of people don't.

So why don't you show me in the matchmaker where the Elo aspect of it is adding 2 or 3 minutes to wait times. Go ahead. In fact, why don't you show me exactly how the matchmaker works in terms of criteria.

Then I want you to point out in the Elo equation and how it's applied in MW:O where it won't function. Seriously, would love to see it. Show me how it's not possible to solve for a persistent ~8% factor with adequate sample size.

So this is where you make assumptions again and state them like facts again, while avoiding doing any of the above - because, well, you can't. Not the last half. If you actually educated yourself on the matchmaker you absolutely COULD show how it uses Elo and how the matchmaker fills matches and with what criteria, at least in a general sense. You can not however predict the impact of the matchmakers use of Elo on match fill times, because you don't have that data. You're just making an assumption, a completely uninformed and absolutely baseless assumption and trying to pass it off as a fact.

Which it isn't.

So that takes us back to the start -

You don't understand Elo or the matchmaker but you had to wait 3 minutes for a match and you saw where someone said 'Elo' once, so you're an expert now.

Flat out, without question, you're wrong. On all accounts. It is hypothetically possible that your particular Elo there isn't 23 other people even approximately close. At last report from CIG Elo was simply used to break people into tiers of 0-1000, 1001-1500, 1501-2800 to fill matches though it started with closest to your score first though it's possible they rolled it back without telling anyone. If they did however, again, Elo would be a minimal impact on your match fill time unless you were way, way on one end of the scale or the other. If you're not dropping every match in a 4man poptart tryhard then you're probably not at the 2800 line and unless you're piloting your trial mech with a steering wheel you're probably not at the 0 mark.

Which points to populations at the time you're dropping and tonnage. Potentially combined with an increased number of premades (too many premades make puzzle-filling teams with close enough tonnage more difficult). Elo parameters have been widened so far at this point that it was intended to still fill matches with a 3/3/3/3 format. With 3/3/3/3 removed the 3 tier approach to Elo makes it far easier.

Also Elo works just fine for teams. If that's not the case you need to show, mathematically, why not. Because the fact that it does work has been proven, repeatedly. Not by me but by a variety of statisticians, not the least of which is the stunningly expensive investment in the TrueSkill system Microsoft put together. Really rather brilliant; it can not only seat your Elo but take your performance in, say, CoD and use it to pre-seed you in something like Checkers. It's based on the Elo equation but uses a gaussian distribution and standard deviation to predict a belief for the relative skill compared to the mean. This lets it very quickly settle your Elo in a wide range of games based on performance in other games. It does so with such efficiency that it actually can use a Bayesian inference to rank players numerically - not just a general score but place you pretty specifically in a leaderboard style ranking.

Based on Elo though. Why? Because Elo is just an equation. It's an equation for solving for your impact on a win/loss environment and deriving a score from that which is then modified based on its prediction of your odds of having won or loss, based on your history of winning or losing in comparable matches in the past.

TrueSkill is NOT Elo - it's a complex equation based off of it. The difference is that TrueSkill is designed not simply to roughly estimate your odds of winning against different opponents but to rank you in a leaderboard format and pre-seed your Elo score and create a useful belief prediction of your relative skill so that it can move you very quickly towards an accurate ranking.

The matchmaker in MW:O is NOT Elo either - it uses a different k-factor and doesn't use a standard curve for the rankings (nor should it). It also simply uses that data to roughly place players for skill along with relative weight. Ideally it'll do so via 3/3/3/3 but for now it's just rough skill, rough weight matching. It is not, has never been and should never be designed to try and fit people into a leaderboard style ranking or attempt to predict with fine accuracy exactly what your skill is. It'd be a waste of time - there isn't the population to fill that precisely. It does exactly what it needs to do and does so correctly.

You're 8% of your teams success. Depending on your skill it can take ~30 matches to roughly place you, ~90 matches to place you with reasonable accuracy within each weight class. Less if you always run the same chassis and loadout, a bit more if you vary it a lot. It's got a wide margin of error - which doesn't matter as the real factor is and has always been population.

Here is the link to the thread showing the mathematical process by which Elo score and k-factor for use by the MW:O matchmaker is derived. Please explain, with support equation, how that doesn't work and exactly why it's impossible to solve for an 8% consistent factor in a win/loss environment over ~100 samples. Seriously, love to see it, if it's legit and provable it would make you famous the world over and literally change the way we look at mathematics.

We all know you're not able to, so go ahead. Gimmie some baseless, incorrect assumptions put as fact with no supporting evidence. It's been working well for you so far. It's like you're wandering out in the open in an XL Atlas while I'm in a poptart Highlander. You're not even making me work for it. While you're at it why don't you show me the telemetry showing what the weighting of relative Elo score of players is for the wait time for your matches. Would love to see it.

[Karl Streiger]

I'm with you OP.
The not so funny part of the current MM - > if you hit play and found almost instantly a game - with working MM and enough players that should happen, the game is exactly as you want it.

But with each second the MM is searching the chance is rising that the teams will be a horrend mix of different ELOs.

Considering the ammount of time you are waiting for a battle + the team composition - you have 2 negatives.
So the good old CB - MM would be the better option.

[Violent Nick]

DeadlyNerd, all things realistically considered, i like your idea. I'm not miffed about wait times but more concerned about how the elo affects gameplay given the varied pilot competence. Having a seperate non elo queue could help. To make it work properly would probably require a tier system where players under x amount gxp are in a seperate pool before qualifying to have their elo used for matching. I think this would allow them to ease into the game gradually and learn how to play on their own terms. I don't mean to brag here but my highest dmg was a respectable 1052 scored yesterday, i specialise in playing tactically using harder weapons like gauss etc.. (still get my butt handed to me regularly and make my fair share of mistakes) but some matches are terrible simply because my team and the enemy are at totally different levels which no one can see. I've had so many instances of trying to forge some teamplay only to find no one, or only half the team gets it. Every now and then 12 lonewolves manage to work together when led by the few random experienced players but this is rare. For the sake of the pugs, the game and our sanity, please take this on board pgi..

Edited by Nick86, 16 May 2014 - 06:23 AM.

[EgoSlayer]

DeadlyNerd, on 14 May 2014 - 09:50 PM, said:

<snip>


I'm not talking about Elo being bad for matchmaker because stomps, I'm talking about Elo being bad for matchmaker because wait times.

Apparently you don't realize that it was class matching (3/3/3/3) that broke the matchmaker, not Elo. Because while there is a nice bell curve distribution of people for Elo ratings and far more likely to find a match for your rating; there is not a bell curve distribution of mechs by weight *or* class. That coupled with 3 different and exclude-able game modes is why the match maker takes time to find matches.

An Elo only queue would be much faster than a weight/class only queue because there isn't an even distribution of mechs by class or weight. Elo Plus anything is going to take longer, just like Class plus anything (E.g. class or elo plus game mode, Class + elo + game mode, etc) because every criteria reduces the pool of available players. If anything look at WoW, which has literally millions of concurrent players - the wait queues for some classes are very long because way more people play those classes. We have the same problem here with an uneven distribution for each mech class, only without the millions of concurrent players.

What would help is a stop light chart of wait times by class - that would change behaviors for some who don't want to wait. Instant queuing in a medium? OK. 5 Minutes for an Assault? Guess I'll play a Medium.

Edited by EgoSlayer, 16 May 2014 - 07:56 AM.

[Mister Blastman]

EgoSlayer, on 16 May 2014 - 07:52 AM, said:

What would help is a stop light chart of wait times by class - that would change behaviors for some who don't want to wait. Instant queuing in a medium? OK. 5 Minutes for an Assault? Guess I'll play a Medium.

This is actually a pretty good idea.