Pruning the Unlabeled Data to Improve Semi-Supervised Learning