THE reference used to make this thread turned out to be BS.
This is not how FAF trueskill works.
Player A and player B
Au - player A skill
Ao - player A width of Gaussian around Au
Bu - player B skill
Bo - player B width of Gaussian around Bu
Coming into a game you have these 4 numbers, leaving the game you
will get those 4 numbers updated based on the outcome of the game.
if Au wins then
new Au = Au + Ao*(some arbitrary positive constant)*(some arbitrary positive function(not of Ao but of some other bs)*(Au-Bu)
if Au loses then change + to - above
new Ao = Ao*(1-some positive number)
So what we have here is a parameter Ao that always decreases with every game and makes the effect of future games on Au progressively smaller with the number of games.
Such a set up makes sense when you are studying entities that possess an underlying CONSTANT true Au that you are trying to figure out, essentially using the law of large numbers. The more you play the closer you will get to your true Au, assuming your true Au is constant.
Please tell me how this is not absurd.
ref. http://research.microsoft.com/en-us/pro ... tails.aspx
a very badly written BS piece