Three Approaches to supervised learning for compositional data with pairwise logratios