LearningGlobalTransparentModelsConsistentwith LocalContrastiveExplanations