Gradient-based boosting for statistical relational learning: the Markov logic network and missing data cases