Optimal Allocation Of Information Granularity To The Inputs Of Granular Neural Networks