Say I have a set of continuously arriving outcomes that have 'goodness' ratings attached. I don't know the distribution of said ratings, but I have some idea of the range, and I have a 1-level deep history of past ratings for some of them. I want to know how many outcomes I should wait for before grabbing one and calling it good. In this case, the payoff for picking sooner is fairly high.
I know there is some sort of actual method for this. The problem is sort of akin to Amazon's old Gold Box Model: given a price, you pick it now or you let it go and wait, giving up a known gain for a potential better future deal.
Can anyone tell me what I'm trying to think of here? This is terribly hard to Google search for :-p
-- Avani (proof positive that you can do years of ML without learning basic stats...)