Ignore or reject in text with many entities

Hi! Iam new using prodigy and I have some doubts about choosing mainly ignore or reject:

  • When we run ner.teach and we find a text that has some entities with correct predictions and some entities with wrong predictions, what should we choose: accept, reject or ignore?

  • When we run ner.make-gold and we have a text where there are no entities should we chose reject (as if we did understand the text but there is none of the entities we need) or ignore (as if we didn´t understand the text at all).

When you do ner.teach, it should just highlight one entity at a time, rather than all the entities in the piece of text. Is it highlighting multiple spans for you?

When there are (correctly) no entities I would answer accept. See this thread for a discussion of when to accept or reject. tl;dr: if you’d be happy with your model returning those spans, accept, and if you’d not be happy, reject.

Hope that helps! (Just chiming in while @ines is off being awesome)

1 Like

Thanks a lot for your answer Andy!. We will chose accept and see what happens with our model.

1 Like