Often the structure that relates objects or variables in machine learning tasks is assumed to be constant, for example, the data comes in samples of fixed size as in all of the models presented in Section 3.1. Sometimes the samples are structured, like molecules, or related to each other in an individual manner, like web pages. First-order logic, or equivalently, relational modelling is needed to represent such structured data. Inductive logic programming aims at learning logic programmes from data by combining machine learning and first-order logic, but let us first discuss logic programming in general.