Occam factors and model-independent Bayesian learning of continuous distributions