Handling high-dimensional data in statistical compilation: a relational approach