FACTORBASE : SQL for Multi-Relational Model Learning

We describe FACTORBASE , a new framework that leverages a relational database management system (RDBMS) to support multi-relational graphical model learning. The basic insight behind our approach is that an RDBMS can be leveraged to manage not only big data, but also to manage big models [1, 2]: First, model structure and model parameters can be managed efficiently without having to be stored in main memory. Second, the Structured Query Language (SQL) supports constructing, storing, and transforming structured statistical objects. The FACTORBASE system uses SQL as a high-level scripting language for statistical-relational learning of a graphical model structure. Our implementation shows how the SQL constructs in FACTORBASE facilitate fast, modular, and reliable program development. Empirical evidence from six benchmark databases indicates that leveraging database system capabilities achieves scalable model structure learning.