Jump to content

Elo Worthless


298 replies to this topic

#81 Sarsaparilla Kid

    Member

  • PipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 664 posts
  • LocationGold Country

Posted 13 November 2013 - 01:39 PM

View PostMudhutwarrior, on 13 November 2013 - 01:06 PM, said:



ELO works best for me? I am sure not above average and I pug only so I don't know where your obsevation comes from. I get stomped consitently and by very good premades. My kd is .74 so explain that one please.


Seems like Grits N Gravy exactly described your situation as being in Elo Hell...you're running up against very good premades as a skilled solo dropper, and it feels like you're losing a lot more than winning, but you're stuck in the same Elo range because your expected loss to a better team doesn't affect you as much as an unexpected win against a better team.

#82 Mudhutwarrior

    Member

  • PipPipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 4,183 posts
  • LocationThe perimieter, out here there are no stars.

Posted 13 November 2013 - 01:47 PM

View PostNick Makiaveli, on 13 November 2013 - 01:17 PM, said:


Please explain what you think KDR has to do with MWO's version of ELO? Winning and losing matter, not your personal KDR.

Again with the evil premades. You really have issues with losing don't you? You say you are "not above average" yet it's their fault you lose?


What is it with you clowns using "evil premades"? Did I say that? I said I get placed agianst good premades. I said nothing derogatory. Really what is this the Karl Rove battalion? More spin and a tilt a whirl here.

#83 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 13 November 2013 - 01:52 PM

View Postnehebkau, on 13 November 2013 - 01:32 PM, said:

We don't need to go into all the maths <fecal matter> in regards to this issue. I'ts simple, if you are joining as a group your total ELO should be higher than your individual ELO. The bigger the group the higher that is. This would also account for the MUCH MUCH MUCH higher use of in-game VOIP by pre-made groups over pugs.

The question being, how much better are 4 people who are on voip, have compatible mechs and have an understanding of each others play-style than 4 strangers, using chat with incompatible mechs? I'd say 50% better by thumbnailing it. How about 8 players? I'd say still, 50% better. I'd even go so far as to say 2 are still 50% better.

Shouldn't be too hard manage in code.



That's still not the problem though. Here's the issue -

You drop in a 4man. You guys are the bomb-diggity and win a LOT. Regardless of how the Elo bumps your rating when it adds you to a team on how it expects you to perform you play competitive and use voip with your buddies and win a lot of games.

Elo helps the matchmaker by making a prediction on who will win based on everyones score. Your score is, lets say, 2,000 because when you drop with your buddies you're a dervish and beat other highly skilled teams. It essentially says that 'against an enemy team whos Elo score is 2000, you'll win half the time. Higher will beat you more often, lower less often'. Even if the matchmaker treats your team like it's worth 2500 it's essentially doing the same thing to all the other premades. It's like if you have 10 people who each have $500 and want to know who's richest. You give them all an extra $500, they're all richer! Just... now they're all still holding the same amount of money, it's just more.

Make sense?

Then you pug.

The matchmaker sees you as worth an Elo of 2,000 still even though without your team you're probably closer to 1600. So it drops you in a match with two teams whos AVERAGE is 1800. Your presence in the team is offset by someone with an Elo score of 1600 as well, as the matchmaker thinks you'll pull a lot of weight to make up for it.

The result is that instead of your team getting 1 ringer and 1 slightly below average guys your team gets 2 below average guys (for that matches predicted value anyway). When you lose the matchmaker reduces your score way more than normal because it thinks you, with a score of 2,000, should absolutely have helped win a match against a team of 1800 value.

So you pug and you get dropped in matches against people way better than you are able to beat when pugging. Your Elo drops....

Then you go back in 4mans. Your Elo has dropped to 1800 instead of 2,000. So has 2 of your teammates. Now your whole teams Elo is lower so.... the 4 of you get dropped against people you are better than and you stomp all over them until you get back to your 2,000 score.

The result? You're throwing off predictions for matches when you pug which makes your score drop more quickly than it normally would. Then you go back to your premade and you're sandbagging the team so your whole team is now roflstomping nubs (by comparison). The cycle will rinse and repeat.

I'm inclined to agree with a gaussian curve and lower k-value, I think we've got enough differentiation in the player base to support that it's been a lot of months and who knows how many thousands and thousands of matches.

I'm just really, really of the opinion that scores for premade and pug need split.

#84 Mudhutwarrior

    Member

  • PipPipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 4,183 posts
  • LocationThe perimieter, out here there are no stars.

Posted 13 November 2013 - 01:52 PM

View PostSarsaparilla Kid, on 13 November 2013 - 01:39 PM, said:


Seems like Grits N Gravy exactly described your situation as being in Elo Hell...you're running up against very good premades as a skilled solo dropper, and it feels like you're losing a lot more than winning, but you're stuck in the same Elo range because your expected loss to a better team doesn't affect you as much as an unexpected win against a better team.


Well I tend to think its not ELO or MM. My beleif in that is that a month or two ago I didnt fare well but I did much better than now. Now there seems no relief whatsover. I just cant begin to imagine anyone staying under those conditions. Im a hard case and tenacious is my way but that not the same for the majority. I just decided to stay till it gets better or the game dies and the servers go quiet. I just dont see new players anymore and cant think of the last time I saw a noob question in chat. I did several times a day a month or two back. Now nothing. Is that elo or lack of blood in the game?

#85 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 13 November 2013 - 02:16 PM

View PostMudhutwarrior, on 13 November 2013 - 01:52 PM, said:


Well I tend to think its not ELO or MM. My beleif in that is that a month or two ago I didnt fare well but I did much better than now. Now there seems no relief whatsover. I just cant begin to imagine anyone staying under those conditions. Im a hard case and tenacious is my way but that not the same for the majority. I just decided to stay till it gets better or the game dies and the servers go quiet. I just dont see new players anymore and cant think of the last time I saw a noob question in chat. I did several times a day a month or two back. Now nothing. Is that elo or lack of blood in the game?


So remember Elo isn't always a direct representation of your straight wins. If you win a bit more than you lose you'll slowly but steadily gain Elo. What it sounds like (not a lot of new players in your teams, etc) is that you've moved far enough up to be out of new player Elo bands with any frequency.

Change mechs recently? Did you used to play something that brought home the wins more often? Also remember that it's by weight class and not mech.

That sounds far more like Elo, you're just playing in a more experienced environment. It's a bloody conversion, you're also going to get pulled up to flesh out matches at higher Elo.

For example your Elo right now might be 1600. You'll get pulled into matches between teams with an 1800 average Elo, you being paired with some 2,000 Elo sharpshooter. The result? Most of the people on the other team might be better than you. Even more painful - you get points added or removed from your Elo relative to the difficulty of the team you're playing against. If you're getting pulled up as filler for matches against better teams you might only lose a handful of points each time you lose but gain a pile when you win.

Result? In that instance you might lose 60% or 70% of your matches and still remain Elo-neutral, possibly even continuing to gain points. Each win is worth way more, each loss costing you way less.

On the plus side if you keep plugging away you'll push through into the Golden Paradise. I'm teasing a bit but generally put in a more competitive environment people tend to improve faster. If you can continue to gain Elo you'll end up being on the other end of that spectrum. You'll be the higher-tiered player that lower-tiered people are brought in to feed upon. There's a certain point where it suddenly flips - now you're winning all the time but your wins are worth very little. You lose occasionally but the losses are steep. You'll probably run a 1.2 or 1.3 win/loss rate just to stay neutral. Statistically at that point MOST players are lower scored than you and as such you'll play the role of wolf among sheep.

Patience gets rewarded in an Elo system. Just keep trying to do better and the math will reward your skill and perseverance.

#86 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 13 November 2013 - 02:33 PM

View PostMudhutwarrior, on 13 November 2013 - 01:06 PM, said:



ELO works best for me? I am sure not above average and I pug only so I don't know where your obsevation comes from. I get stomped consitently and by very good premades. My kd is .74 so explain that one please.

Let's look at a worse case scenario your score is -2 standard deviations away from the mean. Meaning you are well below average skill. The match maker allows for a difference of 2 standard deviations from the mean. As there are not enough people at your skill level or below for the majority of your matches, you tend to get matched against higher ranked players.

You see a lot of premades for the inverse reason. The pool of players with high Elo is limited, if there is no match they are mostly likely to get a match in the Standard deviation pool of matches, where you're likely to have the majority of your matches too.

This is also Elo hell. You lose no Elo points from a loss against these high ranked teams and have little opportunity to gain enough Elo points to get out of this predicament even if your skill improves. As you see few games where you can actually make enough Elo points to move closer to the mean score, which would give you better matchmaking.

View PostMischiefSC, on 13 November 2013 - 01:52 PM, said:

I'm just really, really of the opinion that scores for premade and pug need split.


The thing about a Gaussian and lower K factor Elo, is it allows matchmaking to self segregate better. So you don't need to split pools or have different Elo scores for premades. 4 man's well tend to matched against 4 man's more frequently, as the Elo score reflective of this play will be of a lower range but more highly populated.

Worst case scenario, If highly ranked players found queue times to long, you simply adjust just their match making criteria. Allowing skilled players to have a higher matchmaking ceiling. So some with a score at +1.5 Standard deviations could get a match against someone with 3 and vice verse. Though some above 1.5 Standard deviations could not get a match against someone with the mean score. Ceilings and floors in worst case scenarios are better solutions than splitting the queue or attempting to modify scores.

#87 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 13 November 2013 - 02:56 PM

View PostMischiefSC, on 13 November 2013 - 01:28 PM, said:

I also absolutely don't get how this wouldn't extend weight times.

With out changing the population, you basically double the population at the mean score. If you double the population you can half the matchmaking criteria and not effect queue times. Once you get above or below 1.5 standard deviations, the population thins out more. This might require tweaks to matchmaking criteria for players ie floors and ceilings as I discussed in my previous post. If you don't want to try and hack a solution you could get away with leaving it as this group represent 13% of the population.

#88 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 13 November 2013 - 02:56 PM

View PostGrits N Gravy, on 13 November 2013 - 02:33 PM, said:

Let's look at a worse case scenario your score is -2 standard deviations away from the mean. Meaning you are well below average skill. The match maker allows for a difference of 2 standard deviations from the mean. As there are not enough people at your skill level or below for the majority of your matches, you tend to get matched against higher ranked players.

You see a lot of premades for the inverse reason. The pool of players with high Elo is limited, if there is no match they are mostly likely to get a match in the Standard deviation pool of matches, where you're likely to have the majority of your matches too.

This is also Elo hell. You lose no Elo points from a loss against these high ranked teams and have little opportunity to gain enough Elo points to get out of this predicament even if your skill improves. As you see few games where you can actually make enough Elo points to move closer to the mean score, which would give you better matchmaking.


The thing about a Gaussian and lower K factor Elo, is it allows matchmaking to self segregate better. So you don't need to split pools or have different Elo scores for premades. 4 man's well tend to matched against 4 man's more frequently, as the Elo score reflective of this play will be of a lower range but more highly populated.

Worst case scenario, If highly ranked players found queue times to long, you simply adjust just their match making criteria. Allowing skilled players to have a higher matchmaking ceiling. So some with a score at +1.5 Standard deviations could get a match against someone with 3 and vice verse. Though some above 1.5 Standard deviations could not get a match against someone with the mean score. Ceilings and floors in worst case scenarios are better solutions than splitting the queue or attempting to modify scores.


I get your point and I agree - just that when a premade pugs he's skewing results.

Someone drops premade for 400 matches and pugging for 50 matches. The variance between his skill in a premade and his skill in a pug isn't unreasonable to say is 100-200 points, possibly more depending on how good his friends are. That's enough to skew Elo predictions for the whole team. Most people drop premade sometimes, many drop premade 50% or more. That introduces the potential of a lot of skewed results given that it will magnify the more people with a mismatched premade value vs their actual pug value. It self-reinforces because when they premade again the losses to Elo they'll have taken will drop the overall value of their premade team so it then rolls a bunch of easy wins to get everyones Elo back up. They take that skewed result into pugs, skew results there until they seat. Then their results are skewed when they bring them back to premade games.

Make sense? That's a lot of variance and match prediction disruption that could be removed by just splitting the pools and would give everyone more consistent results and less variance across everyones matches.

As to fixing queue times for higher Elo players I'd say widen tonnage instead of Elo variance. It's always a big deal but with Spiders, Jenners, Shadow Hawks all being powerful and effective low tonnage mechs (especially in organized 4man groups) you can account for more tonnage mismatch with higher skill. This would also help reduce what I think we all agree is the issue we want to solve - wide Elo ranges dropping in the same match. If everyone on my team is competent I'm comfortable with more tonnage mismatch. That'll go my way as often as it goes against me. Elo level mismatch I can't see and account for. Tonnage I can. if I see we're all in lights but I know everyone is about my skill we can make a plan. If I see a balanced tonnage range on my team I have no idea how many of those people are keen-eyed killers and how many are playing with a steering wheel and pedals.

#89 Tombstoner

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bridesmaid
  • Bridesmaid
  • 2,193 posts

Posted 13 November 2013 - 03:04 PM

View PostMalino, on 11 November 2013 - 02:47 PM, said:

Hi,

Been playing a while and now we have more variety in mechs I'm seeing more and more games worthless because tonnage balance is so far out.

Whats the point of ELO when you're got a team of half lights with the remainder mediums and heavies facing off against half a team of Assaults, then heavies and mids.

Regularly I'm seeing matches with 2-300 tonnes difference between the teams. This leads to the inevitable steamrollering going on latley.

ELO, nice idea for one on one matches. Sucks for 12 -v- 12.

Mal

well ELO works great with a large population. with a small pop it sucks. you wait for a long time and you get matches with big weight differences. i have notices a substantial increase in delays and tonnage as well. population must be low.

The diversity you see is from the influx of new players from the official release and the favorable PC gamer article.
they dont know any better and run what they think is cool and not effective.

i think PGI suffered a big drop in player numbers with the ghost heat and gauss nerf fiascos.
partially countered again with the influx of new players from an angle article.

#90 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 13 November 2013 - 03:05 PM

View PostGrits N Gravy, on 13 November 2013 - 02:56 PM, said:

With out changing the population, you basically double the population at the mean score. If you double the population you can half the matchmaking criteria and not effect queue times. Once you get above or below 1.5 standard deviations, the population thins out more. This might require tweaks to matchmaking criteria for players ie floors and ceilings as I discussed in my previous post. If you don't want to try and hack a solution you could get away with leaving it as this group represent 13% of the population.



OOOOOOHHHH!!!!

Holy ****, fair enough. With a gaussian distribution you've got all but the furthest extremes of scores in a thicker population to pull from.

Honestly? People in the furthest edges are going to go one of two directions anyway. Highest Elo range is probably pure premade who's going to spend way more of their time in 12mans - no real impact on pugging games. When they do play they're going to drop largely with less skilled people but differences in performance between the top 5% are pretty narrow, I suspect most of that difference is going to be expressed in good communication and coordination between teams more so than ubermensch solo players.

The lowest range is probably trying to avoid choking on their own tongue more than they're worried about score results and, well, the bottom 7% of performers are probably going to either change or churn out of the game after 300 matches anyway. When they do play they'll serve as padding for people in the lower but not incompetent range and help 'boost' the band above them. I don't mean to sound critical but people don't end up in the bottom of the heap without a lot of poor decisions.

Okay, I like your idea, it's mathematically sound and perfectly logical, pending some telemetry from PGI showing why it wouldn't work lets go to Canada and shout at their windows until they do it.

#91 nehebkau

    Member

  • PipPipPipPipPipPipPipPipPip
  • 3,386 posts
  • LocationIn a water-rights dispute with a Beaver

Posted 13 November 2013 - 03:16 PM

View PostMischiefSC, on 13 November 2013 - 02:16 PM, said:

Result? In that instance you might lose 60% or 70% of your matches and still remain Elo-neutral, possibly even continuing to gain points. Each win is worth way more, each loss costing you way less.


And there lies the problem. Somone loosing 70% of their matches through no real fault of their own is going to quit the game in frustration. I don't really care about the person who'se ELO is inflated because they usually play in a group and then solo. I care about the numerous people who just pug and want to have fun and end-up being used as fodder to an 8-man because MM has a stroke.

Edited by nehebkau, 13 November 2013 - 03:22 PM.


#92 AC

    Member

  • PipPipPipPipPipPipPipPip
  • Elite Founder
  • Elite Founder
  • 1,161 posts

Posted 13 November 2013 - 03:25 PM

View Postnehebkau, on 13 November 2013 - 03:16 PM, said:


And there lies the problem. Somone loosing 70% of their matches through no real fault of their own is going to quit the game in frustration. I don't really care about the person who'se ELO is inflated because they usually play in a group and then solo. I care about the numerous people who just pug and want to have fun and end-up being used as fodder to an 8-man because MM has a stroke.



BOOM! ^ THIS

Plus, I think some of you misunderstand how ELO balancing works. You are giving it much more intelligence that I think PGI programmed into it.

#93 Dirkdaring

    Member

  • PipPipPipPipPipPipPip
  • Survivor
  • 685 posts
  • LocationTwycross

Posted 13 November 2013 - 03:32 PM

^ what he said.

4 man drops in queue. No other premades available.
You have high elo, you get put on other side against them.

Rest fills in accordingly by ELO/Weight.

Thats my guess anyhow.

#94 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 13 November 2013 - 03:41 PM

View PostGrits N Gravy, on 13 November 2013 - 02:33 PM, said:

The thing about a Gaussian and lower K factor Elo, is it allows matchmaking to self segregate better. So you don't need to split pools or have different Elo scores for premades. 4 man's well tend to matched against 4 man's more frequently, as the Elo score reflective of this play will be of a lower range but more highly populated.

Worst case scenario, If highly ranked players found queue times to long, you simply adjust just their match making criteria. Allowing skilled players to have a higher matchmaking ceiling. So some with a score at +1.5 Standard deviations could get a match against someone with 3 and vice verse. Though some above 1.5 Standard deviations could not get a match against someone with the mean score. Ceilings and floors in worst case scenarios are better solutions than splitting the queue or attempting to modify scores.


For that to work though you'd have to wipe and restart Elo - which is highly unlikely to happen. You're still going to get the same sort of situation with premades as well - it's not about premades getting to drop against less skilled players to inflate their score it's that a player in a premade team plays better and has a higher score. So while you may accurately play at a 2000 Elo while in your 4man team your skill is more like 1600 when pugging.

Make sense? It's not about accuracy while in a premade. It's that while you're in the premade you play at one skill level and while you pug you play at another. If you primarily premade it will skew your value when you pug significantly.

#95 Sarsaparilla Kid

    Member

  • PipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 664 posts
  • LocationGold Country

Posted 13 November 2013 - 03:53 PM

View PostMischiefSC, on 13 November 2013 - 03:41 PM, said:


For that to work though you'd have to wipe and restart Elo - which is highly unlikely to happen.


An Elo restart would make things exciting for a bit, I'm sure... :D

#96 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 13 November 2013 - 04:10 PM

View PostAC, on 13 November 2013 - 03:25 PM, said:



BOOM! ^ THIS

Plus, I think some of you misunderstand how ELO balancing works. You are giving it much more intelligence that I think PGI programmed into it.

It's the logistic implementation used by most chess federations. Compare the formula from mwo http://mwomercs.com/...ost__p__1626065 to the one in Wikipedia http://en.wikipedia....matical_details

That implementation will always derive scores into a logistic distribution. The graph of MWO player's Elo scores, published by them, is solid proof of that. Dark blue is players with over 50 gamesPosted Image

This type of implementation can only work quickly with wide matchmaking criteria. Elo hell is a function of it's operation under the conditions it runs in.

#97 Novakaine

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Veteran Founder
  • Veteran Founder
  • 5,728 posts
  • LocationThe Republic of Texas

Posted 13 November 2013 - 04:13 PM

It's because smurfs designed it.

#98 Tekadept

    Member

  • PipPipPipPipPipPipPipPip
  • Knight Errant
  • 1,290 posts
  • LocationPerth, Australia

Posted 13 November 2013 - 04:23 PM

View PostGrits N Gravy, on 13 November 2013 - 04:10 PM, said:

It's the logistic implementation used by most chess federations.

Imagine if Chess was a 16v16 Free for all... :D

Edited by Tekadept, 13 November 2013 - 04:36 PM.


#99 Roadbeer

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Elite Founder
  • 8,160 posts
  • LocationWazan, Zion Cluster

Posted 13 November 2013 - 04:26 PM

View PostTekadept, on 13 November 2013 - 04:23 PM, said:

Imagine if Chess was a 16v16 Free for all... :P


Fixed :D

#100 cSand

    Member

  • PipPipPipPipPipPipPipPipPip
  • 4,589 posts
  • LocationCanada, eh

Posted 13 November 2013 - 04:38 PM

View PostRandalf Yorgen, on 11 November 2013 - 04:28 PM, said:

with 6 lights if you can be patient you can pick off the enemy team one a time or if they are pushing for your base, your mediums/hevies fight a delaying action while your lights storm their base. They would never get beck in time and if one or two mediums did you send out two lgiths to slw them down.


I had a match the other night... 5 or 6 lights on our side, and a mixture of med/heavy/assault for the other half of the team.

We cleaned up, if I recall correctly. I think II have a screenshot which I'll post if I can find it when I get home





3 user(s) are reading this topic

0 members, 3 guests, 0 anonymous users