Forget it, ws is wrong. If you can figure out a function of a vector of weights w that will ensure </ ? (i,j){w is exactly as likely as %/(i,j){w then you can replace (+ +/) with that function. But I suppose that begs the question.
-Dan