Quote:
Originally Posted by sricharan92
Pr(f(x) != g(x)) should be calculated considering all x points and not just the training data. Am I right?

You are right. Considering all
points can be done analytically or approximated by Monte Carlo using a new, large set of points generated at random.