Learning programs by combining programs

The goal of inductive logic programming is to induce a set of rules (a logic program) that generalises examples. Inducing programs with many rules and literals is a major challenge. To tackle this challenge, we decompose programs into \emph{non-separable} fragments, learn fragments separately, and then combine them. We implement our approach in a generate, test, combine, and constrain loop. Our anytime approach can learn optimal, recursive, and large programs and supports predicate invention. Our experiments on multiple domains (including program synthesis and inductive general game playing) show that our approach can increase predictive accuracies and reduce learning times compared to existing approaches.Comment: early draft of manuscript (more of a technical report) to accompany code release. arXiv admin note: text overlap with arXiv:2109.0781