Assessing the performance of macromolecular sequence classifiers