When Machine Learning Gets Personal: Understanding Fairness of Personalized Models