Computing Subsets of Equivalence Classes for Large FSMs