Author Topic: ELO Scores for Ladder and not only  (Read 10837 times)

Offline Igognito

  • Axe Thrower
  • ****
  • Posts: 406
    • View Profile
ELO Scores for Ladder and not only
« on: November 10, 2016, 11:59:15 AM »
Hi all,

Yesterday I implemented the ELO formulas for ladder.
I'm starting this forum topic to discuss some of the small details and issues.

======== Topic 1 Initialization =========
First lets discuss initialization of values:
We have a few options:
1) Do nothing in ~50 games the values should start be representative
2) Initialize some values by taking the wins - loses of old ladder games and assume you play with 1300 ELO
3) Initialize by re-introducing some games from the history
4) Initialize manual some players

all options have problems as you might think: (3) is the most hard to implement, (4) is biased.
(1) is the most representative but slow method
(2) is fast, requires not so much implementation but might not be representative.

====== Topic 2: Inflation/deflation =====
ELO has in-build mechanism to protect you from inflation/deflation.
There is a parameter K which affects how much the score increases/decreases per game.
By keeping that parameter High we get to have fast increases/decrease of our scores
By keeping that parameter Low we maintain the scores

the usual strategy is to keep the parameter High for the first games (right now set to 30 games)
then use a mid value for non Master players with more than some games
and a low value for Master players

All these are functions that we can tune. What is the best values always depends on the specific application of the formula.
We can have arbitrary K's and over time we will get it tuned to serve our purposes.

I want to point out that in the end it doesn't matter that much! ELO is a relative score. All that ELO says is that if you have a difference of 200 points it means that you would win/lose ~74% of the time.

The highest your ELO goes the harder it becomes to increase it and the lower it goes the harder to lower it.
So assuming that people will start playing in a short time (I estimate ~50 games per player) the ELO should be representative! Meaning that if you see players that are 200 elo higher than you, your chances to win are low.
Also a player playing vs a player with 200 less ELO will win few points of ELO.

That is another trick of the ELO... You want to play always against marginally better players than you! In that way you increase your score!
Actually, that is the only way to inflate your ELO. If you play easy picks then your ELO will slowly increase (in some cases you might even earn 0 ELO).
And when you will lose 1 game over a low ELO (just by chance) then you will lose loads of points (K)! As a result you want to play with players that are of equal power or slightly better than you.

====== Topic 3: Inactivity ====
This is the only real problem with ELO. If a player stops playing the ELO remains constant. Which means a player that reaches a high ELO might choose to become inactive in order to maintain their score.
This is not urgent but we need to find a counter for that!
One simple idea is after a time of inactivity we modify the K of that player. This is what most chess associations do, but only for players with no title.

======= Topic 4: Map specific ELO =====
Well if you see the ELO scores right now you will notice that the first player to get a push was Unu!
Obviously, all his games are nwtr! That is something we cant easily control.

Players like Smeagol or Unu or some only chop players will be able to gather high ELO by playing only 1 map and being unbeatable on that map.
I think PB and Chop maps are not taken in account currently.

I was thinking whether we should implement multiple ladders: GoW ladder, Chop ladder, PB ladder, NWTR ladder; Mixed ladder (includes any map).
Implementation wise it is not so hard as it sounds. But it will be a space burden for the DB.

======== Topic 5: Team play ELO; ffa ELO =====
It is doable but not so simple. Also it is not clear what it means.

There are more topics too speak about but lets start with these for now!

I started this topic to get some brain storming.
Cheers

Offline I hate naggers

  • Ogre Mage
  • ********
  • Posts: 2345
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #1 on: November 10, 2016, 12:08:22 PM »
Elo. Not ELO

Offline I hate naggers

  • Ogre Mage
  • ********
  • Posts: 2345
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #2 on: November 10, 2016, 12:13:58 PM »
======== Topic 1 Initialization =========

1) Do nothing in ~50 games the values should start be representative
because we have this new player mechanism - the usual strategy is to keep the parameter High for the first games (right now set to 30 games). I would call them "placement matches".

However, i believe this should be the only use of the k parameter.
Instead of inactivity-prevention mechanism, i would simply introduce some sort of point inflation that would promote activity (but very minor). It would encourage being active.

>map specific
I dont think chops, customs or pballs should have their ladders. I would like to see a "map of the week" system from some map pool (known bnes for starters) which would give 30% more elo points for a week for victories or something. The map of the week would rotate

team elo is irrevelant. ignore it

but you have your ideas, i have mine, lets see how your ladder turns out
« Last Edit: November 10, 2016, 12:28:10 PM by CLAW the racist »

Offline Certified MENSA Genius Brain (smart)

  • "The Architect"
  • Global Moderator
  • Dragon
  • *****
  • Posts: 5384
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #3 on: November 10, 2016, 12:15:09 PM »
I'm assuming people get a +/- Elo bump at the time a ladder game is reported?  If so, perhaps we could show that on the player.php page.

Example:
00Logan (#4) defeated SPB- (#12) on "Garden of War" for 35 Elo points at 03:20:45 on 11/10/2016.
    

Offline I hate naggers

  • Ogre Mage
  • ********
  • Posts: 2345
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #4 on: November 10, 2016, 12:18:38 PM »
right now i believe the system is in beta phase and only a side stat on the page. maybe we can get commited or add sorting by Elo if it proves to be accurate and/or people like it

Offline Igognito

  • Axe Thrower
  • ****
  • Posts: 406
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #5 on: November 10, 2016, 01:32:44 PM »
It is in beta mode :-P

Keeping players active is one of our targets, the question is how to do so. I do not think by giving extra points you motivate them.

About map of the week, that might be complicated but interesting as by changing the map we will get people to play other maps too.

Also I was thinking that all tournament matches should count more!
This will be made possible by the tournament organizer providing the results of all matches.

By the way the ELO (Elo) system will not replace the ladder system.
ELO is only a system for comparing among players. Its target is to give you an indication of the chance you have to win.

Also a future target is to get Elo to be reported inside the game with a command like /stats (This is not something I can promise yet but I will look at it)

Also ELO will be a tool for creating the tiers! One good thing with ELO is that you cant cheat it easily after it normalizes.

Cheers

Offline I hate naggers

  • Ogre Mage
  • ********
  • Posts: 2345
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #6 on: November 10, 2016, 01:38:38 PM »
Keeping players active is one of our targets, the question is how to do so. I do not think by giving extra points you motivate them.
?????? if youre trying to motivate them by ANY kind of ladders, RANKS (and in Elo ladder, POINTS affect RANKS) is the only thing u got

Offline Igognito

  • Axe Thrower
  • ****
  • Posts: 406
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #7 on: November 10, 2016, 02:39:16 PM »
To be honest, I think it would motivate them more if we gave prizes for reaching an ELO rating instead of giving more points...

say that we give you a special banner for reaching 2000 ELO...
in general ELO will be useful for several tasks :-)

Offline the_magician

  • Grunt
  • ***
  • Posts: 93
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #8 on: November 10, 2016, 04:14:08 PM »
cool idea mang, would love to see this implemented

cheers!

Offline Winchester

  • Ogre Mage
  • ********
  • Posts: 1737
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #9 on: November 10, 2016, 08:55:05 PM »
when you will lose 1 game over a low ELO (just by chance) then you will lose loads of points
With all the smurfs, I don't know how this would work, like if you vsing someone with alot higher elo, and you win, then he logs on his smurf his with low elo cause his worried he'll lose again and wins consistently and makes you lose it all + more ? I can name plenty of people that would do that just out of spite for losing a game or two to players that they're better against.


Offline Certified MENSA Genius Brain (smart)

  • "The Architect"
  • Global Moderator
  • Dragon
  • *****
  • Posts: 5384
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #10 on: November 10, 2016, 11:29:49 PM »
That, and the risk of people newb bashing and dodging pros in general, is why we use a rung-operated case's system.  The elo stuff is cool but I like it as just an added skill rating there for extra info more than as a ranking system, which would be horrible - just see the built-in ladder history for evidence.
    

Offline Paper_Boy

  • Grunt
  • ***
  • Posts: 131
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #11 on: November 11, 2016, 01:00:04 AM »
Can't we have 2 ladders that try both methods on 2 pages? We shouldn't encourage dodging and inactivity ever. It was pretty cool seeing the team ladder rating and your winning % with certain players etc... That was a cool metric to have that  always interested me more than singles

Offline mousEtopher

  • Administrator
  • Ogre Mage
  • *****
  • Posts: 2065
    • View Profile
    • War2.me
Re: ELO Scores for Ladder and not only
« Reply #12 on: November 11, 2016, 09:19:13 AM »
Yep, definitely can. Would like to fix double's ladder too even if we don't make an Elo version of that
squeak!

Offline Certified MENSA Genius Brain (smart)

  • "The Architect"
  • Global Moderator
  • Dragon
  • *****
  • Posts: 5384
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #13 on: November 11, 2016, 09:34:04 AM »
The built-in ladder is already an elo system.  We've had it count all tvb games before.  The result was always some 35-0 nobody sitting at the top after carefully bashing newbies on a gimmick map.

Here's my thought, put the elo ranking as a column on the ladder page, like it is in this test, and let people click on the column header to sort by elo ranking if they are interested
    

Offline Igognito

  • Axe Thrower
  • ****
  • Posts: 406
    • View Profile
Re: ELO Scores for Ladder and not only
« Reply #14 on: November 11, 2016, 10:11:07 AM »
Hi all,

===== Topic 1: ELO vs Ladder/ELO presentation ======
ELO and ladder are two different things. I thought that this was pretty clear.
The only relation the two have is that ELO ratings are computed by the same games that count for ladder.
Also that we currently present the ELO at the ladder page.

Well, I believe that the ELO should be presented in many different locations like:
Status page.
and ingame /stats  command.

How doable is that is a second question.

======= Topic 2: ELO & Smurfs ==========

About smurfs and ELO and the scenario described.

Yes the system is not bullet proof and malicious usage can abuse it a bit; but the target is for people to get an idea of the players strength compared with their own strength instead of actually making a ranking.

ELO gives you statistical information. If player A has a score of 1700 and player B has a score of 1500 that means that player A has ~75% chance to win player B.
When a smurf starts winning the ELO will increase rapidly so; the smurf wont be so smurfish anymore ;-)

Of course I should add 2 more information in the site: how much games a player has played and their tier!
If you see a player with ELO 1300 and 0 games and a player with ELO 1300 and 100 games you realize that you have a high error on your estimation for the first player.

After the system stabilizes I would consider to use the opponent Game # information in order to protect from smurfs.

Example: Player A: ELO 1700; Games 300; Player S: ELO 1300; Games 0. Player S wins!

S earns their ELO points normally.
A would normally lose quite some ELO in this case but instead as Player S has 0 Games I reduce the amount of lost ELO. I haven't thought the formula and the math yet. But that is an easy Smurf protection we can do.

Thus ELO can become an anti Smurf protection tool after a while!

=========== Topic 3: Team ladder and Team ELO ==========
That should go to the TODO list...

Method 1
If teams are balanced (equal number of players) we could avg the ELO ratings.

Method 2
Take the ELO of the strongest team player as the basis.
Reduce all other ELO's by 1000 and sum them. (Yes players with less than 1000 ELO penalize the team)
This works also for non balanced teams

Method 3
Try to match player vs player
Best of Team A vs Best of Team B etc..
Handle it as multiple 1v1 games if balanced; otherwise devise something weird. lol

Finally we can use the math from: http://sradack.blogspot.pt/2008/06/elo-rating-system-multiple-players.html
This is similar with my summing approach.

Conclusion: It is possible but maybe not desirable.
Still this is no team score. This is just a way to award ELO points to each player for games with teams.

About team ladder, I haven't been involved on that but perhaps in the future I will...

Cheers