A good exemple of ranking problem.

[Kamikaze28]

(09-10-2012 01:19 AM)Joggies Wrote:  Clearly an ideal system would give more weight to recent games to account for the fact that people get better as they play more.

Such a system might discount the relevance of a any result by 1/x^2 where x is how many matches you've played since. (You could obviously modulate the strength of the formula for the desired 'stickiness' of course)

I think the 'ideal system' you are referring to is a theoretical construct which is capable of matching people as perfectly as possible with all the knowledge it gathered so far. This of course is nigh impossible to implement.

Nevertheless, I agree with you. And before we descend into a sheer endless rabbit hole of possible mathematical descriptions for various aspects of the matchmaking system, I would just like to stop this topic of discussion in its tracks by saying that there are lots of different ways to modify, tweak, tune, fine-tune and re-tune systems of this sort. We don't need to attempt to enumerate them. Alex has probably done his research before his implementation and sees the same issues we see and discuss here.