#vc-dimension

Question 1: Single choice

The VC dimension of a class is the largest number of points that its functions can label in every possible way with $\pm 1$ . Thinking about the magnitude of the functions, why can it be a poor measure of how complex a class is?

It only sees the sign of each function, not how large its values or weights are1
It always equals the number of parameters, so it adds nothing beyond counting weights2
It changes with the data distribution, so it differs from one dataset to another3
It is defined only for linear classifiers, so other model classes have no VC dimension4

Timed test