Quote:
Originally Posted by z_beeblebrox
Can't you, from OPR, which TBA calculates anyway, simply predict the scores of the remaining matches and use those to predict rankings?
|
Prediction is tricky. It's easy to do a deterministic prediction (win/lose), but that's almost useless. How close is the win? How likely is it that that outcome will occur? What's far more valuable in most situations is a probabilistic prediction.
For a simple OPR-based probability, you could do something like this:
Code:
%RedWin = (Red1OPR+Red2OPR+Red3OPR)/(Red1OPR+Red2OPR+Red3OPR+Blue1OPR+Blue2OPR+Blue3OPR)
Basically, this says that Red has a 75% chance to win if Red has 75% of the OPR.
However, you also have to consider here that OPR -- the basis for that probability we generated -- is not very good for some games. This year was one, and 2012 comes to mind readily. Even in games where OPR is actually pretty good (like 2013), OPR should never be regarded as the 'god metric' that most people think it is. See
my whitepaper for more of a numerical analysis.
A better method of prediction in games like 2013 or 2012 where scoring was more linear and separable (and where scores were approximately normally distributed) would be to calculate the average points teams were putting up and the standard deviation in those points. You could then generate a normal model for total red points, with:
Code:
mu = Red1Average + Red2Average + Red3Average - Blue1Average - Blue2Average -Blue3Average
sigma = sqrt(Red1StandardDeviation^2+Red2StandardDeviation^2+Red3StandardDeviation^2+Blue1StandardDeviation^2+Blue2StandardDeviation^2+Blue3StandardDeviation^2)
%RedWin = normalcdf(0, 10^99, mu, sigma)
But we digress here.
TL;DR: Yes, could be implemented, but OPR-based predictions are overrated.