Approximate learning of high dimensional Bayesian network structures via pruning of Candidate Parent Sets