Density Estimators for Positive-Unlabeled Learning