Jump to content

Proof That Elo/mm Is Working?


66 replies to this topic

#21 cSand

    Member

  • PipPipPipPipPipPipPipPipPip
  • 4,589 posts
  • LocationCanada, eh

Posted 29 January 2015 - 12:33 PM

View PostDuke Nedo, on 28 January 2015 - 02:11 PM, said:

Most common win result: 12-4, most common loss result: 5-12. Very few 12-0, 0-12 actually!





In a lot of matches, you see guys complaining about their team / the MM / LRMS :P... when the final result is 4-6 to 12.

Which is ridiculous because that is actually a pretty good match.

We don't have respawns, so many people are used to that COD style where there is no "loss" per se... you just continually get second chances until the matches just end the score tallies up. You get one chance/match here which can be frustrating/hard for people used that "endless lives" style.

#22 F4T 4L

    Member

  • PipPipPipPipPipPipPip
  • 767 posts

Posted 29 January 2015 - 12:39 PM

View PostDuke Nedo, on 28 January 2015 - 02:11 PM, said:

Ok sorry, just needed a catchy title. :ph34r:

Just a friendly suggestion for everyone complaining about Elo and matchmaking in the solo queue. After having a few win-streaks and loss-streaks too many I decided to track all end results and plot them for fun. Plot below is cumulative wins and losses, -1 for a loss, +1 for a win.

Posted Image

It was a bit less cyclic and periodic so far than it subjectively felt like so I would recommend doing this to get a more objective picture of how it really looks. There was a spot where I felt pulled down towards 0 by the MM in this sample, but over all the trend actually looks much better than perceived and fits well with my global 1.133 w/l ratio after reset and 1.10 for archive.

Sample still small, growing slowly every day... Too early to say if anything about anything, except that the perceived pain is a bit worse than it actually is.

Edit: Added a graph of distribution of end results, plotted the opponents score on wins and own teams score on losses. Most common win result: 12-4, most common loss result: 5-12. Very few 12-0, 0-12 actually!

Posted Image


Great data. However.. you are a single sample. What we need is:

Easy individual access to our own cumulative data, plus regularly published stats for all of this stuff, in bulk.

I'd especially love to see what happens to those first 20 game new players, personally.

IDK why PGI dont publish this **** alrdy. Oh wait, I'm not an idiot, I do know that.

#23 cSand

    Member

  • PipPipPipPipPipPipPipPipPip
  • 4,589 posts
  • LocationCanada, eh

Posted 29 January 2015 - 12:41 PM

View PostF4T 4L, on 29 January 2015 - 12:39 PM, said:


I'd especially love to see what happens to those first 20 game new players, personally.


Oh I can tell you

Like any new player to any other game.... pwnage to the max. Some guys will have some decent skills to start out but will still likely get pwned.

The goal is to have a reasonable grasp of things after those 25 matches are up

#24 LordSkippy

    Member

  • PipPipPipPipPipPip
  • Overlord
  • Overlord
  • 451 posts
  • LocationAustin, TX

Posted 29 January 2015 - 12:46 PM

View PostcSand, on 29 January 2015 - 12:33 PM, said:



In a lot of matches, you see guys complaining about their team / the MM / LRMS :P... when the final result is 4-6 to 12.


I've seen people rage quit with "TEAM FAIL!!!1!" while our team was was up two or more kills. For some, if they died, it's always the team's fault and failure.

#25 Fate 6

    Member

  • PipPipPipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 2,466 posts

Posted 29 January 2015 - 12:58 PM

Graphs that don't clearly display the data (no labels on axes, etc) are really hard to read btw.

It seems like matchmaking doesn't really work if you win a lot more than you lose. It should pretty much even out at some point. I have the same experience though, where I seem to win more than I lose (at least when I'm in a mech that I'm good with). Right now I'm sitting at 55% winrate after 1000 games or whatever it is I've played since the reset, and that is a good deal higher than 50%. In a game like League of Legends when a champion hits a winrate around 55% they become the target of the next balance change.

#26 Duke Nedo

    Member

  • PipPipPipPipPipPipPipPipPip
  • CS 2023 Top 12 Qualifier
  • CS 2023 Top 12 Qualifier
  • 2,184 posts

Posted 29 January 2015 - 01:03 PM

View PostFrDrake, on 29 January 2015 - 12:29 PM, said:

One more followup question, are you "tryharding" in these matches, using optimal mechs/loadouts or is this data from you leveling up new mechs?


Something in between perhaps, I have been skilling up mechs a lot during these games but I always run the best build I can come up with, all upgrades and modules. So, 50% will be "meta"-mechs and the first 50% of each mech will be un-elited "meta"-mechs... Did some matches in Gargoyles too though, that pulls it down a bit...

#27 Duke Nedo

    Member

  • PipPipPipPipPipPipPipPipPip
  • CS 2023 Top 12 Qualifier
  • CS 2023 Top 12 Qualifier
  • 2,184 posts

Posted 04 March 2015 - 11:40 AM

Edited the OP with another 250 games. Interesting deadlock at 1.0 w/l ratio since 26. January...

#28 Holdfast

    Member

  • PipPipPip
  • Liquid Metal
  • Liquid Metal
  • 67 posts
  • LocationNoVa

Posted 04 March 2015 - 12:32 PM

I did a much smaller analysis of my win / loss record (http://mwomercs.com/...44#entry4190544) where it leveled out at after something like 350 matches:

My first 30 matches, which includes mastering the mech (not really relevant): Win ratio: 1.73
The next 341 matches: Win ratio: 1.38
The latest 187 matches: Win ratio: .99
I'm no where near being elite, so the pool of people I end up pugging with is still fairly large. I rarely recognize the names, one way or another. But yah, I agree with the conclusion that ELO works, for whatever that's worth.

#29 Duke Nedo

    Member

  • PipPipPipPipPipPipPipPipPip
  • CS 2023 Top 12 Qualifier
  • CS 2023 Top 12 Qualifier
  • 2,184 posts

Posted 04 March 2015 - 11:23 PM

View PostHoldfast, on 04 March 2015 - 12:32 PM, said:

I did a much smaller analysis of my win / loss record (http://mwomercs.com/...44#entry4190544) where it leveled out at after something like 350 matches:

My first 30 matches, which includes mastering the mech (not really relevant): Win ratio: 1.73
The next 341 matches: Win ratio: 1.38
The latest 187 matches: Win ratio: .99
I'm no where near being elite, so the pool of people I end up pugging with is still fairly large. I rarely recognize the names, one way or another. But yah, I agree with the conclusion that ELO works, for whatever that's worth.


Nice, always interesting to read data like this! I couldn't read out from your post how many games you had in mediums before you started recording so in your case it could definitely be an Elo climb. In my case I now have 5500 recorded drops on the stats page, so I would have had 5000+ drops before my first match in this dataset. With such a long baseline before measuring, the stats I recorded manually should not be related to my Elo going up I think, unless I made a skill quantum leap lol.

What I wanted to catch was the streakyness of solo games, and that started out rather well with one or two win/loss streaks, then I just started winning, and then suddenly on the 26. or 27. January (27. was a tuesday, was there a patch?) the graph became (unnaturally) flat. My archived stats say w/l ratio 1.121 over 5500 games, but after that date it's deadlocked on 1.000 for 300 games. Not the biggest sample, but the impression is that there is less I can do to affect the outcome of each match now. Perhaps I am just a late arriver to the infamous Elo hell. :)

#30 Ralgas

    Member

  • PipPipPipPipPipPipPipPip
  • Overlord
  • Overlord
  • 1,628 posts
  • LocationThe Wonderful world of OZ

Posted 04 March 2015 - 11:41 PM

View PostDuke Nedo, on 04 March 2015 - 11:23 PM, said:


Nice, always interesting to read data like this! I couldn't read out from your post how many games you had in mediums before you started recording so in your case it could definitely be an Elo climb. In my case I now have 5500 recorded drops on the stats page, so I would have had 5000+ drops before my first match in this dataset. With such a long baseline before measuring, the stats I recorded manually should not be related to my Elo going up I think, unless I made a skill quantum leap lol.

What I wanted to catch was the streakyness of solo games, and that started out rather well with one or two win/loss streaks, then I just started winning, and then suddenly on the 26. or 27. January (27. was a tuesday, was there a patch?) the graph became (unnaturally) flat. My archived stats say w/l ratio 1.121 over 5500 games, but after that date it's deadlocked on 1.000 for 300 games. Not the biggest sample, but the impression is that there is less I can do to affect the outcome of each match now. Perhaps I am just a late arriver to the infamous Elo hell. :)


It may not be as bad as it's made out, but there are so many variables to consider...

People having a bad day, current bracket players available for that particular drop, undergeared or unmastered mechs in play in the drop (if i remember rightly elo don't care about chassis or the state it's in it just uses your score for that particular weight class) Unbalanced team loadouts, mismatched weight drops and i'm sure there are plenty of others.

I can easily see where anomalies crop up with minimal fault to the current system (not in any way saying it cant be improved though) and in that light seeing it get so close is quite a surprise

#31 Brother MEX

    Member

  • PipPipPipPipPipPip
  • The Patron Saint
  • The Patron Saint
  • 407 posts
  • LocationRANDIS IV

Posted 05 March 2015 - 01:29 AM

View PostLordKnightFandragon, on 28 January 2015 - 02:13 PM, said:

Luckyyyyyy, you get on the side that wins more often then not, whats the secret?
That he is favoured by the UNFAIR MatchMaker !

I, on the other hand, have a bad standing with the matchmaker which reachently has deteriorated to less than 1 win every 4 games ( worse than 75% loss ratio, below 25% win ratio ... whatever you call it ! )

#32 Holdfast

    Member

  • PipPipPip
  • Liquid Metal
  • Liquid Metal
  • 67 posts
  • LocationNoVa

Posted 06 March 2015 - 11:29 AM

(quoting seems bonked for me at the moment)

Duke Nedo Says: "Nice, always interesting to read data like this! I couldn't read out from your post how many games you had in mediums before you started recording so in your case it could definitely be an Elo climb. In my case I now have 5500 recorded drops on the stats page, so I would have had 5000+ drops before my first match in this dataset. With such a long baseline before measuring, the stats I recorded manually should not be related to my Elo going up I think, unless I made a skill quantum leap lol."

I had probably around a thousand-ish matches in mediums, and then I discovered lrm skirmishing in the cent and then the 4j (which got so blessed by the quirks fairy). The 4j was pretty much the only medium mech I was playing for this time frame, and I was significantly better in it than in anything I had played before.

Also, I have a vague memory (all my memories are vague, actually) that PGI did something to improve their queueing algorithm around then. Maybe they fixed the ELO? That would also explain why you keep seeing the same people, as their ELO would have bee affected the same way.

I really, rally wish they'd have more transparency into their stats collection.

#33 Tarogato

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 6,558 posts
  • LocationUSA

Posted 06 March 2015 - 01:06 PM

Elo is divided by weight class, so perhaps you should track your W/L for each weight class individually.

I'll start doing the same and tracking my own data, because I love stuff like this. I'll also track my matchscore rating (1-12). See you in about 500 matches, this will take a while...

#34 ThrashInc

    Member

  • PipPipPipPipPipPip
  • Caladbolg
  • Caladbolg
  • 248 posts

Posted 06 March 2015 - 01:41 PM

I think you would have to also math out matches that occur during events or mech release weeks.

For example, regardless of your "best" weight class, the MM was throwing matches with 8 lights vs 6 assaults during Resistance release, so you might have been the only assault, or one of 8 panthers.

My assault W/L is considerably higher than any other bracket, with the overall W/L being 105 games up.

#35 J0anna

    Member

  • PipPipPipPipPipPipPip
  • The Merciless
  • The Merciless
  • 939 posts

Posted 06 March 2015 - 01:56 PM

While I agree that ELO appears to work, I'm not so sure the MM algorithm is optimal. Wins and losses is only one part of elo, it also includes whether the MM thinks you should win or lose. Win when you should and you gain little to no elo, same as losing, but win when you shouldn't (or vice versa) and big changes to elo. I think ELO needs quite a few drops per weight class to level out, like in the few 100's range. However, I still believe that elo works. I propose a simple test: Figure out your average round in a mech (I like about 200 to 300 drops). Then track your w/l in that mech when you perform above average and below. I have noticed a definite trend that I lose a much greater % of the time when I don't perform up to standards, then when I do.

#36 ThrashInc

    Member

  • PipPipPipPipPipPip
  • Caladbolg
  • Caladbolg
  • 248 posts

Posted 06 March 2015 - 01:59 PM

I lose a lot of 800-1200 damage games, I wouldn't be too set on that.

Imagine you're doing 1000 damage, a few kills, but everyone else is in trial mechs, subpar variants to XP, or crap builds, no ECM, all LRMs, etc.

It's very easy to have the MM tell you to hoist and not have a big enough backpack.

#37 Yokaiko

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Bad Company
  • Bad Company
  • 6,775 posts

Posted 06 March 2015 - 02:22 PM

View PostGundamStompyRobotChurchy, on 06 March 2015 - 01:59 PM, said:

I lose a lot of 800-1200 damage games, I wouldn't be too set on that.

Imagine you're doing 1000 damage, a few kills, but everyone else is in trial mechs, subpar variants to XP, or crap builds, no ECM, all LRMs, etc.

It's very easy to have the MM tell you to hoist and not have a big enough backpack.


Start an account get in a light and see how many big name unit tags you see.

I assure you it won't be a non-trivial percentage

#38 Jman5

    Member

  • PipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 4,914 posts

Posted 06 March 2015 - 02:47 PM

thanks for the updated data. It was good to see your w/l even out eventually. I wonder what that big climb in wins meant. Perhaps you leveled up in skill and it took a little bit for your score to readjust. Maybe it had to do with the mech choice or you were playing at a low population time.

#39 J0anna

    Member

  • PipPipPipPipPipPipPip
  • The Merciless
  • The Merciless
  • 939 posts

Posted 07 March 2015 - 12:37 AM

View PostThrashInc, on 06 March 2015 - 01:59 PM, said:

I lose a lot of 800-1200 damage games, I wouldn't be too set on that.

Imagine you're doing 1000 damage, a few kills, but everyone else is in trial mechs, subpar variants to XP, or crap builds, no ECM, all LRMs, etc.

It's very easy to have the MM tell you to hoist and not have a big enough backpack.


But as I said, that's a MM issue, not an elo one. What we never know is whether the MM thinks we should win or not. WoT does this so much better, and I wish PGI would share this info with us. Ideally, the MM should limit the distribution of ELO ranges on each team. That is, everyone should be +/- 10 elo from team average - that would ensure we were all playing against equal opponents. Just setting the team ELO's equal to each other fails when one team varies greatly from the average.

#40 -Vompo-

    Member

  • PipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 532 posts
  • LocationFinland

Posted 07 March 2015 - 02:11 AM

Even without any matchmaker the good and bad teams equal out eventually with enough games. The only thing that is common with all the games is the player himself or herself. The better players end up with positive W/L. The average player has 1 win for every game lost and the players with negative W/L can only blame themselves.

Group queue is somewhat different though since you can have the same team or somewhat same team all the time.





6 user(s) are reading this topic

0 members, 6 guests, 0 anonymous users