I think what you're looking for is a Bayesian average.
http://en.wikipedia.org/wiki/Bayesian_average
Basically you give everybody N fake games at 50% for the purposes of ordering the players. Lets say 100 games.
1-0 = 50.5% instead of 100%
10-0 = 54.5% instead of 100%
200-100 = 62.5% instead of 66.7%
So the more games you play, the closer to the actual average your score becomes.
Or the better idea -- ignore win% completely since it's basically meaningless.
We have some great improvements to the player lists coming in the next week or so - stay tuned!
thats actually decent idea but tbh it doesn't reflect shit
cuz playing solo winrate can be much different than playing with 4 other skilled players
Thanks for the feedback guys!
@mattieshoes I'll look into that, thanks mate.
@Jason Looking forward to it. :)
Por favor, se conecte para postar comentários.
IMHO, win rates of the Top 50 should actively take into account the number of matches played, and not just be based on a lower boundary (200 games at the moment?).
I would like to propose the use of fuzzy weights to take into account the number of games played, thus resulting in an fuzzified win rate (FWR).
Sample:
FWR = W1*AWR + W2*log_k(GP)
where AWR is the actual win rate, GP is the number of games played, and W1 and W2 are fuzzy weights. k here is a multiple of 10, and is used to adjust the fuzzified win rate to correspond to a fixed number of games (say 1000), thus penalizing/rewarding the win rates those with less than/more than k games played.