Learning from Positive and Unlabeled Data under the Selected At Random Assumption