PHIL is a suite of algorithms for learning both the parameters and the structure of Hierarchical Probabilistic Logic Programs (HPLP) from data. The parameters are learned applying gradient descent (DPHIL) or Expectation Maximization (EMPHIL). To perform structure learning, PRIL initially generates a large HPLP and applies a regularized parameter learning on it. Then clauses with small values of probabilities are dropped.
See the GitHub repository.