Machine Learning with Multitype Protected Attributes: Intersectional Fairness through Regularisation