Asked to label pairs that are already marked

[mmcneill]

Hi all, I have a question about the intended behavior of mark_pairs and read_training more generally. I'm loading a small training file and then calling console_label to add additional labels. I'm being asked to label pairs that were already labeled before and in fact exist if I look at deduper.training_pairs. I would have thought that a pair could not simultaneously be in training_pairs and active_learner.candidates. 

Is this the intended behavior? If so, is there an existing method to remove pairs from candidates once they are marked?

Thank you!