Generic predictor that not depends on data itself, this assumes the original data is in L2(R) and there's a sampling theorem.
