Jump to content

Question About Elo


20 replies to this topic

#1 Ens

    Member

  • PipPipPipPipPipPipPipPip
  • Mercenary
  • Mercenary
  • 1,088 posts
  • LocationGermany

Posted 25 October 2014 - 05:41 AM

I´m playing since closed beta and i was supposed to know this ( beat me for this :blink: ) but:

i couldn´t find the posting anymore that explained how the elo is calculated.


was it just by win / loss?
is the "Match Score" in any way taken into account?

Edited by Ens, 25 October 2014 - 07:13 AM.


#2 IceSerpent

    Member

  • PipPipPipPipPipPipPipPipPip
  • 3,044 posts

Posted 25 October 2014 - 07:00 AM

Elo is always based on win/loss only, regardless of the game. Click here for details. There are also posts in Command Chair somewhere regarding specifics of its implementation in MWO.

#3 Warblood

    Member

  • PipPipPipPipPipPipPip
  • The Hunter
  • The Hunter
  • 503 posts
  • LocationMontreal, Qc, Cnd

Posted 25 October 2014 - 07:17 AM

ELO = Win/Loss raito per weight class (start:38min)
http://soundcloud.co...-heat-scale-elo

#4 Ens

    Member

  • PipPipPipPipPipPipPipPip
  • Mercenary
  • Mercenary
  • 1,088 posts
  • LocationGermany

Posted 25 October 2014 - 07:47 AM

exactly what i was looking for. thx!

so in theory the worst of the bad players can have a high ELO when getting pulled by the winning guys....and the match score ( which pretty much tells you how you performed within the match ) is never ever taken into account..

thats just great

Edited by Ens, 25 October 2014 - 07:49 AM.


#5 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 08:09 AM

View PostWarblood, on 25 October 2014 - 07:17 AM, said:

ELO = Win/Loss raito per weight class (start:38min)
http://soundcloud.co...-heat-scale-elo

No. We do have 4 scores - one per weight class, but it's NOT WLR.

Elo is NOT calculated from WLR either. Elo is calculated by predicting the odds of a given team winning a match, and adjusting scores based on the outcome. Elo is not calculated from any stats in fact.

The actual formulas are in the Elo command chair posts here: http://mwomercs.com/...ost__p__1626065

Wins and losses are what determine your Elo - the ONLY thing that determines your Elo - but not the ratio. Also, just because you win doesn't mean your Elo will go up, and just because you lose doesn't mean it will go down.

If you haven't read the command chair posts and don't understand how Elo works(outside of MWO, the wikipedia article linked above is a good read), please educate yourself before spreading more disinformation.

Edited by Wintersdark, 25 October 2014 - 08:10 AM.


#6 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 08:14 AM

View PostEns, on 25 October 2014 - 07:47 AM, said:

exactly what i was looking for. thx!

so in theory the worst of the bad players can have a high ELO when getting pulled by the winning guys....and the match score ( which pretty much tells you how you performed within the match ) is never ever taken into account..

thats just great


Match score is a terrible metric of how you performed in a match.

And the worst of the bad players won't have a high Elo. It changes at most by 50pts per match, and due to the random nature of matches Elo will self-correct over time. Elo does work, within it's limitations. And those limitations apply exactly the same to any other scoring system: If you're one of the top players, you're going to be matched with and against players below your level because there simply aren't enough players at your level looking for a match right in that tiny window of time. The same applies at the other end, too - bottom ranked players will get mixed with higher ranked players because there aren't enough really low-ranked players.

The only way a bad player will get and hold a high Elo score is if he plays primarily in the group queue and always plays with much better teammates. This WILL cause him to get an undeservedly high Elo rating, and will make the Solo queue a terrible place for him as he'll be matched against much better opponents.

Edited by Wintersdark, 25 October 2014 - 08:14 AM.


#7 Abaddonis

    Member

  • PipPipPipPipPip
  • The Scythe
  • The Scythe
  • 129 posts

Posted 25 October 2014 - 08:48 AM

View PostWintersdark, on 25 October 2014 - 08:14 AM, said:


Match score is a terrible metric of how you performed in a match.

And the worst of the bad players won't have a high Elo. It changes at most by 50pts per match, and due to the random nature of matches Elo will self-correct over time. Elo does work, within it's limitations. And those limitations apply exactly the same to any other scoring system: If you're one of the top players, you're going to be matched with and against players below your level because there simply aren't enough players at your level looking for a match right in that tiny window of time. The same applies at the other end, too - bottom ranked players will get mixed with higher ranked players because there aren't enough really low-ranked players.

The only way a bad player will get and hold a high Elo score is if he plays primarily in the group queue and always plays with much better teammates. This WILL cause him to get an undeservedly high Elo rating, and will make the Solo queue a terrible place for him as he'll be matched against much better opponents.



You've completely contradicted yourself in those two paragraphs. Reread. High ELO player gets matched with bad players because ELO is too high. Bad player in group queue carried to high ELO by team now only plays high level players? It's a broken system. The solo queue is a terrible place for everyone.

#8 nehebkau

    Member

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

Posted 25 October 2014 - 08:49 AM

ELO is based on expected wins and expected losses. More unexpected wins = higher ELO and IT sucks, get over it.
At the very least, they could improve it by making ELO scored on weight, not meaning a different ELO per weight class -- that just stupidity --- I mean a multiplier by weight. We all know that weight + pilot skill is the defining factor for success. Two equally skiled pilots, one in an assault and the other in a light, the assault would have the advantage ever single time (please re-read this sentence...i said equally skilled pilots).
That is the nature of the game, all other things being equal, more guns and armor equals a win. Now, ELO simply takes into account your win/loss and, then balances the teams trying to have a fairly even total ELO across both teams. Here is the problem: If you have 3 players with very high LIGHT ELO on one team and 3 players with very high ASSAULT ELO on another team, which team will have the advantage?
The game should have a base ELO, across all chassis – call that the Pilot ELO, then when matchaker is doing its voodo, multiply that ELO by (weight/100). That would give a lower absolute ELO value for a light pilot than the same pilot would have if he were in an assault.
Now before some weenie starts talking about Battle Value, remember there is no way to truly figure out pilot skill for bv to work and, BV would be, frankly, impossibly complex in this customizational game.

So to recap:
ELO not great only accounts for w/l
ELO should take into account drop weight
BattleValue bad for MWO, good for BattleTec.
San Dimas highschool football rules!

Edited by nehebkau, 25 October 2014 - 08:52 AM.


#9 Joseph Mallan

    ForumWarrior

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 35,216 posts
  • Google+: Link
  • Facebook: Link
  • LocationMallanhold, Furillo

Posted 25 October 2014 - 08:52 AM

try tracking where you are placed in your team when you drop and see if you have better games when closer to Alpha Lance? I swear on my family's lives when I am in Alpha I kick ash, In Bravo I see knock down drag out down to the wire matches, Charlie... I Get Rolled Baby Rolled!

#10 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 08:57 AM

View PostAbaddonis, on 25 October 2014 - 08:48 AM, said:



You've completely contradicted yourself in those two paragraphs. Reread. High ELO player gets matched with bad players because ELO is too high. Bad player in group queue carried to high ELO by team now only plays high level players? It's a broken system. The solo queue is a terrible place for everyone.


If you primarily play in the solo queue, your Elo will be correct. If you primarily play in the group queue, your Elo will be correct.

The point of Elo is that it adapts to your most common play. Its not a "per match" system, its an aggregate system. Play in the group queue a lot and win a lot, you'll be matched against similar opponents. Your odd, rare solo queue matches will be incorrectly rated because we don't have separate scores for different queues. If you play in the solo queue a lot though your scores will correct to what they should be there.

My first paragraph was basically "how things are normally" and my second was the exception.

There is no better system, though. Elo isn't perfect - there absolutely are failure cases. But it IS the best system currently. If you had a better system you could make a hell of a lot of money with it.

That the vast majority of players end up with roughly 1:1 WLR is proof of the systems success.


#11 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 09:02 AM

View PostJoseph Mallan, on 25 October 2014 - 08:52 AM, said:

try tracking where you are placed in your team when you drop and see if you have better games when closer to Alpha Lance? I swear on my family's lives when I am in Alpha I kick ash, In Bravo I see knock down drag out down to the wire matches, Charlie... I Get Rolled Baby Rolled!
We know lance assignment is random now. With that on mind, consider that spawn points are not: it could be that you're just not good at Charlie spawn points, or that maps are unbalanced with regards to those spawn points and that puts those mechs at a disadvantage.

However, Elo has no bearing on lance assignment whatsoever. The dedicated servers assign players to lances, not the matchmaker. The dedicated servers assign lances based on tonnage: they are given a randomised list of players from the matchmaker, and then choose lance for each the 4 lightest, 4 heaviest, and remaining 4 players. That's it. (This being solo queue).

In the group queue, players are assigned lances from largest group to smallest.

#12 Abaddonis

    Member

  • PipPipPipPipPip
  • The Scythe
  • The Scythe
  • 129 posts

Posted 25 October 2014 - 09:03 AM

View PostWintersdark, on 25 October 2014 - 08:57 AM, said:

If you primarily play in the solo queue, your Elo will be correct. If you primarily play in the group queue, your Elo will be correct.

The point of Elo is that it adapts to your most common play. Its not a "per match" system, its an aggregate system. Play in the group queue a lot and win a lot, you'll be matched against similar opponents. Your odd, rare solo queue matches will be incorrectly rated because we don't have separate scores for different queues. If you play in the solo queue a lot though your scores will correct to what they should be there.

My first paragraph was basically "how things are normally" and my second was the exception.

There is no better system, though. Elo isn't perfect - there absolutely are failure cases. But it IS the best system currently. If you had a better system you could make a hell of a lot of money with it.

That the vast majority of players end up with roughly 1:1 WLR is proof of the systems success.



The flaw in this thought process is the vast majority of the player are NOT 1:1 skill players. This is not functional design if it is based on a welfare system to carry bad players up to 1:1.

#13 Joseph Mallan

    ForumWarrior

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 35,216 posts
  • Google+: Link
  • Facebook: Link
  • LocationMallanhold, Furillo

Posted 25 October 2014 - 09:04 AM

View PostWintersdark, on 25 October 2014 - 09:02 AM, said:

We know lance assignment is random now. With that on mind, consider that spawn points are not: it could be that you're just not good at Charlie spawn points, or that maps are unbalanced with regards to those spawn points and that puts those mechs at a disadvantage.

However, Elo has no bearing on lance assignment whatsoever. The dedicated servers assign players to lances, not the matchmaker. The dedicated servers assign lances based on tonnage: they are given a randomised list of players from the matchmaker, and then choose lance for each the 4 lightest, 4 heaviest, and remaining 4 players. That's it. (This being solo queue).

In the group queue, players are assigned lances from largest group to smallest.

If you guys say so. but my numbers say different.

#14 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 09:14 AM

View PostJoseph Mallan, on 25 October 2014 - 09:04 AM, said:

If you guys say so. but my numbers say different.
your numbers could be a result of what I said above. Maybe its because of spawn point locations?

Its not just "me saying so". Karl Berg has been very clear aboutbhow the matchmaker and dedicated servers work in assigning players to lances. What's more, its entirely testable. In the solo queue, lance assignment is always by tonnage, with those tonnage groups then randomly assigned to spawns.

Unless you feel that its more likely that Karl is flat out lying(he not only said they weren't but that it was impossible to do so), *and* that even if lances where assigned by Elo that would make any sense.

I think its much more likely that there are balance issues with the survivability of the charlie spawn point which is usually quite distant from the remainder of the team.

In short, your numbers don't say different, you're just drawing an incorrect conclusion from those numbers.

Edited by Wintersdark, 25 October 2014 - 09:14 AM.


#15 EgoSlayer

    Member

  • PipPipPipPipPipPipPipPip
  • Wrath
  • Wrath
  • 1,909 posts
  • Location[REDACTED]

Posted 25 October 2014 - 09:17 AM

View PostAbaddonis, on 25 October 2014 - 09:03 AM, said:



The flaw in this thought process is the vast majority of the player are NOT 1:1 skill players. This is not functional design if it is based on a welfare system to carry bad players up to 1:1.


It's not a welfare system that carries bad players up to 1:1 - it's a system where those bad players are mostly matched against each other so they have an equal chance of winning. The extremes on the top and bottom end, coupled with a match maker timeout to create a match is why either group ends up in the "average" match pool.

Edited by EgoSlayer, 25 October 2014 - 09:17 AM.


#16 Joseph Mallan

    ForumWarrior

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 35,216 posts
  • Google+: Link
  • Facebook: Link
  • LocationMallanhold, Furillo

Posted 25 October 2014 - 09:19 AM

View PostWintersdark, on 25 October 2014 - 09:14 AM, said:

your numbers could be a result of what I said above. Maybe its because of spawn point locations?

Its not just "me saying so". Karl Berg has been very clear aboutbhow the matchmaker and dedicated servers work in assigning players to lances. What's more, its entirely testable. In the solo queue, lance assignment is always by tonnage, with those tonnage groups then randomly assigned to spawns.

Unless you feel that its more likely that Karl is flat out lying(he not only said they weren't but that it was impossible to do so), *and* that even if lances where assigned by Elo that would make any sense.

I think its much more likely that there are balance issues with the survivability of the charlie spawn point which is usually quite distant from the remainder of the team.

In short, your numbers don't say different, you're just drawing an incorrect conclusion from those numbers.

Ok well I will continue to be happy when I see myself in Alpha, Know I am in for a balanced fight in Bravo and tighten my belt for the bumpy ride in Charlie.

It doesn't matter where the placement is, I form up with the main body and proceed whether its from Alpha or Bravo. My inner statistician sees a pattern what ever it may be. :D

Don't forget, I was a part of a team that wrote our machine shop's Statistical Process Control Q1 process for an R&E department at one of the big 3. Now what the trend is... I don't known, but if I had the data, I bet I could pin it down!

Edited by Joseph Mallan, 25 October 2014 - 09:23 AM.


#17 Darth Futuza

    Member

  • PipPipPipPipPipPipPipPip
  • 1,239 posts

Posted 25 October 2014 - 09:21 AM

Nobody understands ELO, it is magical forbidden dark science closely guarded by ComStar. Don't ask too many questions.

#18 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 09:25 AM

View PostJoseph Mallan, on 25 October 2014 - 09:19 AM, said:

Ok well I will continue to be happy when I see myself in Alpha, Know I am in for a balanced fight in Bravo and tighten my belt for the bumpy ride in Charlie.

It doesn't matter where the placement is, I form up with the main body and proceed whether its from Alpha or Bravo. My inner statistician sees a pattern what ever it may be. :D
Your inner statistician may also see confirmation bias.

What I'd ask is: if not due to lance spawn points, why would you get those results? What do you feel is happening?

Let's assume it is Elo. If lances are assigned by Elo, you'd probably have better matches when you where in charlie because you'd know your whole team was better than you. As you're experienced and capable, that would then mean you'd have a full team of experienced and capable players with you.

But... As I said, unless you suggest Karl is lying about how the whole team making system works (and remember, its his code - he wrote the system), its not Elo.

That doesn't mean your observations are wrong, just that either your conclusions are incorrect or your data is suspect.

Keep in mind, and data from before the queue split is also irrelevant, as back then premades filled alpha first, then bravo then charlie, so placement appeared to show players from most skilled to least: players in larger groups play better because of enhanced familiarity, coordination and communication.

Edited by Wintersdark, 25 October 2014 - 09:28 AM.


#19 Joseph Mallan

    ForumWarrior

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 35,216 posts
  • Google+: Link
  • Facebook: Link
  • LocationMallanhold, Furillo

Posted 25 October 2014 - 09:30 AM

View PostWintersdark, on 25 October 2014 - 09:25 AM, said:

Your inner statistician may also see confirmation bias.

What I'd ask is: if not due to lance spawn points, why would you get those results? What do you feel is happening?

Let's assume it is Elo. If lances are assigned by Elo, you'd probably have better matches when you where in charlie because you'd know your whole team was better than you. As you're experienced and capable, that would then mean you'd have a full team of experienced and capable players with you.

But... As I said, unless you suggest Karl is lying about whole the whole team making system works (and remember, its his code - he wrote the system), its not Elo.

That doesn't mean your observations are wrong, just that either your conclusions are incorrect or your data is suspect.

Actually If I was in Alpha that seems to point out that My Elo is higher and thus why I wrack up 3-5 kills and up 6-8 assist with 400-700 damage, and Charlie is 20-300 damage 0-1 Kills and 0-3 assists, cause everyone is that much better than me!

Also Carl does not have to lying... He could just be mistaken. Everyone does make mistakes you know.

Edited by Joseph Mallan, 25 October 2014 - 09:31 AM.


#20 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 25 October 2014 - 09:42 AM

View PostJoseph Mallan, on 25 October 2014 - 09:30 AM, said:

Actually If I was in Alpha that seems to point out that My Elo is higher and thus why I wrack up 3-5 kills and up 6-8 assist with 400-700 damage, and Charlie is 20-300 damage 0-1 Kills and 0-3 assists, cause everyone is that much better than me!

Also Carl does not have to lying... He could just be mistaken. Everyone does make mistakes you know.
Its extremely unlikely he is mistaken in this case. Have you read his posts on the subject? We wrote them during his Matchmaker rewrite, and updated them immediately after. Its his code. There is literally nobody who better understands how that system works.

The key point is that the dedicated servers assign lances, and the dedicated servers do not have access to your Elo scores.

And this specific topic was discussed, too - it was actually central to the discussion at the time.

Also, look at tonnages when you drop solo. You can see immediately how lances are put together - 4 heaviest, 4 lightest, and 4 middle. This always happens. Which group is assigned to which lance varies, so opponents don't know where to expect the assault lance, for example.

But yeah... You've basically got a set of observations:
"I appear to get the best results in Alpha Lance, and the worst in Charlie". I don't dispute that (how could I? I don't have access to your match results).

All I'm saying is that there is an extremely strong possibility that you're incorrect in why that is. Most notably given your unwillingness to consider other options particularly considering how unlikely your original conclusion was.

Edited by Wintersdark, 25 October 2014 - 09:43 AM.






1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users