We unify submodular/supermodular ratio problems and show general algorithms like SuperGreedy++ and min-norm point methods are surprisingly efficient and scalable, outperforming specialized methods in theory and practice