Traditional strategies to build machine learning based classification systems employ discrete labels as targets. This limits the usefulness of such systems in two ways. First, the generalizability of these systems is limited ...