On Computing Probabilistic Explanations for Decision Trees