Quote:
Originally Posted by udaykamath
I understand the proofs and the theoretical argument. The point that you underline is "give me the breakpoint" and i will give the error bound in terms of examples N etc. Now practically for an algorithm how do we find a breakpoint? If someone comes up with an algorithm f(x), how is breakpoint or VC dimension actually computed? Are there any formal steps etc ?

There is no general, systematic way for finding the break point of a learning model, but for many of the popular models, the break point has already been estimated (approximately if not exactly). In one of the homeworks in the online course, this problem is addressed in a geometric case.