How to calculate VC dimension for matrix factorization (Netflixlike) tasks
Let's say we have some users (u) and some movies (m), and some ratings. The variable we control is d, a number of factors for each movie/user.
We would like to know how many data points we need according to VC theory. This number probably depends on d, so in other words, what d we can afford given the data points we have.
So, how do we go about calculating VC dimension here? Is it (u + m) * d?
