Is there way to create labels with patterns without manually accepting each match?

We are trying to create weak labels with patterns. We would like to accept everything that the patterns suggest. Is there any way we can run the match recipe and accept all at once?

Hi Nikolaos,

you can use the PatternMatcher programmatically as well and upload examples via the add_examples method, which might be what you're interested in.

That said, it'd be good to be careful that these weak labels never get confused with actual labels. If you're looking for a weak labelling framework for NER in spaCy, I might recommend skweak. Also, I've had a lot of success by leveraging sense2vec as a weak labelling approach. The video below gives a recent demo:

2 Likes

Thank you for your reply!