[slashdave]

Also: you are free to model p(x) without worrying about normalization, something that would be required to maximize likelihood.