On Using Decision Tree Coverage Criteria forTesting Machine Learning Models