Fairness in the Eyes of the Data: Certifying Machine-Learning Models