ELO Rating system

[Overstrand]

I worked on this for a bit with Myth and this is what we have:

This is mainly Myth's work (obviously) Thanks Myth

----------------------
Squads start off with 1000 rating points.
When a TWD game has been played, the new rating of the squad will be determined by the following formula:

New Rating = Old Rating + K(W-P)

P = Probablitity of winning
K = Maximum rating change
W = 1 if the squad has won, W = 0 if the squad has lost


P = 1 / (1 + 10^(-difference_in_ratings / 400))
K = 50





Example:
wolvencreed (rating 1000) vs Pallies (rating 1400)

first we'll calculate the probability of winning of wolvencreed over Pallies:
difference_in_ratings = 1000-1400 = -400

P = 1 / (1+10^(-(-400) / 400))
P = 1 / (1+10^1)
P = 1 / 11
P = 0.09 = 9% chance of wolvencreed winning over Pallies.
then ofcourse, the chance of Pallies winning over wolvencreed is 1-0.09 = 0.91 = 91%

- Scenario: wolvencreed wins!
so, for wolvencreed: W = 1
new rating = old rating + K * (W-P)
new rating = 1000 + 50 * (1-0.09)
new rating = 1000 + 45.5 = 1046

for Pallies: W = 0
new rating = 1400 + 50 * (0-0.91) = 1400 + (-45.5) = 1354.5 = 1355



-

K (maximum rating change) is set to 50 in this example. We're considering to lower K when a squad plays more games.

-



This system has been blatantly ripped from the starcraft ladder rating system. ( http://www.battle.net/ladder/ladderfaq.shtml )

[Webbb Llama v1.5]

Thoroughly confusing but it looks good! Damn I want to play some TWD =)

[Dillusion]

what the fuck did you just say, jesus christ

[Wix]

WT !#$%!*%*@%*!&#^$%^!#$)

seeing that wolvencreed only had a 9% chance of winning, and their rating only went up a snitch seems very insignifcant

[X Machine]

Actually, K=50 is rather high in ELO. Taken together with that it takes 30 games to reach a significant rating I think wolvencreed should be happy with that large an increase.

[aphix]

geez, this is taxing on the old brain..

Ok, after over, myth then bleen explained it, i think i have it.

Correct me if im wrong ok. .. (and im taking a mid season perspective)

======================

Onix (1280) vs. Cripples (1631)

Probability's

1/(1+10^(-(-351)/400))
1/(1+10^0.88)
1/9.8
______________
Onix Wins = 11%
Crips Wins = 89%

Scenario1 - Cripples Win

CnR = 1631+50*(1-0.89)
CnR = 1631+5.5 = 1637

OnR = 1280+50*(0-0.11)
OnR = 1280+(-5.5) = 1275

Onix's new rating = 1275 (move lower)
Crip's new rating = 1637(move higher)

Scenario1 - Onix Win

OnR = 1280+50*(1-0.11)
OnR = 1280+45.5 = 1325

CnR = 1631+50*(0-0.89)
CnR = 1631+(-44.5) = 1587

Onix's new rating = 1325 (move higher)
Crips new rating = 1587 (move lower)

======================

Thats probably all wrong

-aphix

[aphix]

What i have seen in the football version, and other ELO systems on other models is where they take the number of games played into account.

-aphix

[Mythrandir]

yes; K might depend on the games played

[Fiction]

Originally posted by Webbb Llama v1.5
Thoroughly confusing but it looks good! Damn I want to play some TWD =)

lmao

same here, i'm not sure of what's happening there with all those numbers... and x's and parenthesis... but i'm sure you guys put a lot of work into and it'll work! And i'm sure i'll get it sooner or later. Lez get TWD goin

[RabbitRapist]

how about starting with 10 points..

win = +1
lose = -1

why confuse people?

[Mythrandir]

how about no?

[Webbb Llama v1.5]

Originally posted by Mythrandir
how about no?

Myth 4 sysop

[pv=nrt]

starting out at 1 is a bad idea, because then you have to go into decimal points to get detailed scores. not that decimals are that hard but it looks cleaner without having 30 squads scores with a period (isn't it a comma in the rest of the world?) after the first number. you can't just do +/- 1 after each win becuase then old squads wanting to exploit the system could face new squads with barely any basers and win easy points. starting at 1000 is fine. though i personally would like to see K=100. just so that there are more points to be won or lost per game.i n a loser sort of way makes it more exciting.

[Mythrandir]

makes sense (the K=100 part)

[Sleepy Weasel]

The math looks fine to me, it's not really complicated people.


I suppose there is a problem with say a new "vet" squad forming. They'd start at 1000 points, could beat someone with like 1600 points and knock them down close to 50. With the max possible you could lose being 50 it's not really that huge of a problem, it's still a slight problem though.

It would be nice if it could be done so losing to a newly formed squad wouldn't knock you down quite as much, maybe only half as much, but would still improve them as much as normal if they won. Thus they could get to their "proper" rating quickly without destroying the ratings of sqauds above them on their way up.

It would also help out the problem of a group of "vets" creating a fake aliased squad pretending to be new and just take down a highly ranked squad a bunch of points.

So yeah, something like until a squad has played 5-10 games or so they only drop a squad half as much as usual is my idea. I'm sure you can think of a better mathmatical way to do this, but I'm sure you follow my reasoning.