Jump to content

Elo Is Coming: What To Expect


277 replies to this topic

#261 Evex

    Member

  • PipPipPipPipPip
  • 154 posts

Posted 19 February 2013 - 09:43 AM

My only concern for this is that ELO was installed last patch, so how accurate is it ? What I mean is how accurate is it when every one had all ready a certain number of x wins and y loses. For instance I have a win loss score of 90 wins and 160 loses. This number was most likely lower from last patch when elo was implemented. For instance last patch a player was given 1300 elo with say a 140/200 win loss but another player with a 50/200 was given the same 1300 elo. You could say by skill the score would of equalized to what it should be. The problem is since the 1300 elo wasn't implemented when every one had a 0/0 win/loss stat then the elo numbers are slightly off to begin with. The only way to fix that would to have set every ones win/loss to 0 during the last patch with no one knowing it. For instance my win rate could of really be 10/80 on PGI servers due to the elo patch, but I'm being shown my win/loss is 90/160.

#262 Mechteric

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Overlord
  • Overlord
  • 7,308 posts
  • LocationRTP, NC

Posted 19 February 2013 - 09:57 AM

View PostEvex, on 19 February 2013 - 09:43 AM, said:

My only concern for this is that ELO was installed last patch, so how accurate is it ? What I mean is how accurate is it when every one had all ready a certain number of x wins and y loses. For instance I have a win loss score of 90 wins and 160 loses. This number was most likely lower from last patch when elo was implemented. For instance last patch a player was given 1300 elo with say a 140/200 win loss but another player with a 50/200 was given the same 1300 elo. You could say by skill the score would of equalized to what it should be. The problem is since the 1300 elo wasn't implemented when every one had a 0/0 win/loss stat then the elo numbers are slightly off to begin with. The only way to fix that would to have set every ones win/loss to 0 during the last patch with no one knowing it. For instance my win rate could of really be 10/80 on PGI servers due to the elo patch, but I'm being shown my win/loss is 90/160.



don't worry so much, its supposed to naturally evolve over time. If you are good, you will probably get some wins in for a while until it increases enough to match you against better pilots. Or the opposite if you were not as good.

#263 Mackman

    Member

  • PipPipPipPipPipPipPip
  • 746 posts
  • LocationCalifornia

Posted 19 February 2013 - 10:01 AM

View PostEvex, on 19 February 2013 - 09:43 AM, said:

My only concern for this is that ELO was installed last patch, so how accurate is it ? What I mean is how accurate is it when every one had all ready a certain number of x wins and y loses. For instance I have a win loss score of 90 wins and 160 loses. This number was most likely lower from last patch when elo was implemented. For instance last patch a player was given 1300 elo with say a 140/200 win loss but another player with a 50/200 was given the same 1300 elo. You could say by skill the score would of equalized to what it should be. The problem is since the 1300 elo wasn't implemented when every one had a 0/0 win/loss stat then the elo numbers are slightly off to begin with. The only way to fix that would to have set every ones win/loss to 0 during the last patch with no one knowing it. For instance my win rate could of really be 10/80 on PGI servers due to the elo patch, but I'm being shown my win/loss is 90/160.


I don't even understand what you're worried about. W/L in a game with a learning curve this high is extremely inaccurate as a way of measuring skill. By only measuring wins and losses from the most current patch, they insure that the W/L they'll be working with is the most recent, and most accurate, depiction of your performance in-game and your potential performance in future games.

#264 Arete

    Member

  • PipPipPipPipPipPip
  • Bridesmaid
  • Bridesmaid
  • 390 posts

Posted 19 February 2013 - 10:26 AM

I'd like the ELO rating to be visible personally, at least for myself. I'm all OK with it begin hidden for others (just as the current stats). I've played quite a bit of HoN, and even though I've never been more than an average skilled player (hanging at about 1600) it's fun to know that I've improved, and it's good to know that I have to focus when I get on the higher end of my skill level. Every time I reach 1650+ I know that I really need to play my best to win, which is fun.

Would also be fun to be able to compare a bit with the rest of the community, If we'd start at 1500, and the top players are at 1800+, then knowing if I'm at 1300 or 1700 is quite interesting. Also, will make recruiting for making larger groups better (on large teamspeak servers for pickup groups for example). Just put up a channel with "4man 1500-1700" and a group of similarly skilled players can be up and running in a heartbeat.

#265 Redoxin

    Member

  • PipPipPipPipPipPip
  • Ace Of Spades
  • 263 posts

Posted 19 February 2013 - 11:39 AM

The opening post was a very good post about the coming Elo system. Finally someone that understood it. Worded it much better than I ever could. I got to say it was somewhat frustrating that many didnt understand what a fundamental change this new matchmaking phase would be.

I also made a post on the MWO forums about this, and how I dont like that Elo wont be displayed:
http://mwomercs.com/...system-phase-3/

Edited by Redoxin, 19 February 2013 - 11:45 AM.


#266 Exilyth

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bridesmaid
  • 2,100 posts
  • LocationTerra

Posted 19 February 2013 - 12:00 PM

Nice to see that match making is being improved upon.

But... wasn't ELO invented for static games like chess, whoose rules don't change over time (e.g. no new chess pieces)?

#267 Chromanin

    Member

  • PipPip
  • Overlord
  • Overlord
  • 30 posts

Posted 19 February 2013 - 01:09 PM

Well this seems very unfair to me.

What can I expect to happen now?

This is what my stats look like:
302 / 234Kills / Death
191 / 243Wins / Losses

1.29Kill / Death Ratio


Edited by Chromanin, 19 February 2013 - 02:48 PM.


#268 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 February 2013 - 01:11 PM

View Post80Bit, on 19 February 2013 - 08:05 AM, said:



Correct me if I am wrong, but the last time population numbers were visible were at the end of the closed beta period, was it not? And I think you are overstating the population requirement to have Elo work. 90% of the players in an Elo system occupy a fairly narrow range of Elo ratings. 2000 being a minimum size for polling has nothing to do with Elo match making. For the majority of players the Elo system should have little problem finding fair matches. Outlying players do face a new problem, of either high queue times, or frequently being places in matches out of their range. You are correct that population size will affect the system, but mainly for the fringe, not the bulk of players.


I disagree that a BV system would work better than an ELO system. In a BV system, you are trying to make predictions based on the equipment's performance instead of the players performance. But it is clear that a K2 in the hands of an unskilled player is vastly different than one in the hands of a skilled player. It is folly to try and predict results by equipment when you have a much better indicator, past performance. Even better, they appear to be narrowing it down to past performance by mech or at least by class, which will match make based on your performance in a mech, rather than by the mech itself.

Population of the que has everything to do with how well an Elo matchmaking system will work. First off Elo systems are mathematically based around the idea that skill is distributed in a Gaussian manner across the population. Population size effects the variance and therefor the standard deviation of a distribution. Population size relates directly to mean and therefor margin of error. The smaller the population the greater the margin off error. Which results in not only less accurate rankings but forces you to allow for a greater spread in matchmaking, in order to take into account margin of error.

Couple margin of error with a distorted standard deviation of player skill and it's not 10% of the population that are outliers, It's the 1 % of the 1%. Meaning your matchmaker isn't doing enough segregation. For all intensive purposes, everyone is still piled together. So what's the point of the system? Even if you set the system up to only allow matches of 1 standard deviation of a players score, Elo would do a poor job of predicting outcomes, due to the inherent inaccuracy of player scores and standard deviation of the group. I'm not saying the system won't find you any matches, I'm saying you will be getting pretty much the same matches as before, minus 1% of the population. With a 40% win rate, that won't help you at all.

BV isn't a predictive system or ranking system, it's a handicapping system. Not all mech's are created equal, and therefore when two random mechs of the same class a paired, both do not have an equal chance at success. The chance for equal success is tantamount to the validity of Elo scores. Elo is based on the assumption that both players are starting with the same number of chess pieces.

What BV and handicapping system do is, adjust the game state to give players of different setups equivalent chances at victory. Give them same amount of chess pieces. We all ready have a handicapping system, and one that isn't very good either. It states all assaults are = to all other assaults, ect, ect. The choice isn't between Elo and BV. It's between Elo and BV or complete random matchmaking and BV, or Elo with current handicapping and random matches with current handicapping. If matchmaking largely turns out to be the same post patch, than the time would have been better spent developing a better handicapping system.

Edited by Grits N Gravy, 19 February 2013 - 01:30 PM.


#269 Chromanin

    Member

  • PipPip
  • Overlord
  • Overlord
  • 30 posts

Posted 19 February 2013 - 02:00 PM

lol, just tested this, I haven't lost a single match yet. The entire game went to super easy all of a sudden with 4-5 kills per match.

EDIT, just after a couple of games my stats raised enormously.
318 / 234Kills / Death
200 / 243Wins / Losses
1.36Kill / Death Ratio

Edited by Chromanin, 19 February 2013 - 02:47 PM.


#270 Calimaw

    Member

  • PipPipPipPipPip
  • 149 posts
  • LocationIowa

Posted 19 February 2013 - 03:13 PM

Posted Image

#271 TheRulesLawyer

    Member

  • PipPipPipPipPipPipPipPip
  • 1,415 posts
  • LocationChicagoland

Posted 19 February 2013 - 03:18 PM

One of the big issues I've seen with Elo for mid-range players is the new player injection point. Where ever they stick this, there is going to be a much wider spread of skill levels compared to rank. There will be smurfs who want to pub-stomp, there will be terri-bad new players that need to sink. If your skill is right near that injection point you get a lot of really random games. Above and below are fine.

I really hope they have something in mind to stop the smurfs. Once your win rate is fixed at 1:1 you get a lot of the "I'm uber l33t, I should win 90% of my games!" player wanting to pub stomp rather than playing people their own level.

#272 Calimaw

    Member

  • PipPipPipPipPip
  • 149 posts
  • LocationIowa

Posted 19 February 2013 - 03:31 PM

20 minutes later I'm in a game, woo hoo!

#273 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 February 2013 - 06:08 PM

View PostMackman, on 18 February 2013 - 08:26 PM, said:


How the hell could it make things worse than a purely random distribution? How? Even if it literally just divides the player base in half, it would give us better matches than we have right now.


So that happened.


http://mwomercs.com/...41#entry1922841

Solo dropers sat in long ques, failed to find matches. Many of the higher end pug players were forced to fill out the other half of crappy 4 man teams fighting against 8 mans. Mid to low end solo droppers couldn't find matches. The quality and mech parity of the matches was overall reported to be lower.

All of these are symptoms of a blown standard deviation, small population pool and very narrow grouping protocols in an Elo matchmaker. PGI actually tuned the MM to the worst possible settings at release. Which was to look for very narrow matching criteria. Thus we had the dual fiasco of bad matches due to inaccurate Elo scores and long que times due to a small pool and a narrow match function.The solution to quicker matches will be to widen the range of Elo scores a solo dropper can face. Thus making the system function almost as identical to a random drop system. Now do you see why I said Elo could make things worse? If you had BV and random drops you'd arguably be in a much better place right now.

#274 p4r4g0n

    Member

  • PipPipPipPipPipPipPipPip
  • Knight Errant
  • 1,511 posts
  • LocationMalaysia

Posted 19 February 2013 - 06:37 PM

Some kind of bug maybe? -> refer to Matthew Craig's post here

#275 Targetloc

    Member

  • PipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 963 posts

Posted 19 February 2013 - 06:49 PM

Steering wheel underhive? Seriously? Are you elitist ******** judging people based on their input devices now?

Wheel and pedals is a perfectly matched control scheme for a mech. Maybe you just need to learn how to drive?!

Check out me and my A1 getting 8-0 on multiples on River City. You might learn a few things.



#276 Reported for Inappropriate Name

    Member

  • PipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 1,767 posts
  • LocationAmericlap

Posted 02 April 2013 - 06:16 PM

Only match wins and losses effect your ELO.


lmao get punished for the mistakes of others, nice system pgi, your player ranking system doesn't even account for skill.

#277 FupDup

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 26,888 posts
  • LocationThe Keeper of Memes

Posted 02 April 2013 - 06:25 PM

View PostBattlecruiser, on 02 April 2013 - 06:16 PM, said:


Only match wins and losses effect your ELO.


lmao get punished for the mistakes of others, nice system pgi, your player ranking system doesn't even account for skill.

Holy necro, Batman!

#278 Reported for Inappropriate Name

    Member

  • PipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 1,767 posts
  • LocationAmericlap

Posted 02 April 2013 - 06:46 PM

View PostFupDup, on 02 April 2013 - 06:25 PM, said:

Holy necro, Batman!

it's what i do





1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users