Data transformations for relational learning

In response to the growing need for analysis of relational and graph data, it is necessary to ensure interoperability between different machine learning methods that process this type of data. Among the most important is inductive logic programming (ILP), which is based on the representation of relational data in a form derived from the Prolog language and learning rules from the W3C RDF standard. This thesis focuses on the ILP system Popper and the RDFRules system, which implements the AMIE approach for the analysis of RDF data. At the time of writing this thesis, there were no publicly available tools that would allow conversion between these formats; that is, if the input data was in RDF format, it was not possible to easily import it into the Popper system and vice versa. This thesis, therefore, describes two new libraries programmed in Python that allow users to transform Prolog atoms to RDF triples and vice versa. The tools are designed to support n-ary predicates and both positive and negative examples in ILP representation. The converters were evaluated on four datasets, including Michalski’s trains dataset. Both libraries are publicly available on GitHub.