I was planning to use Prodigy to improve NER from Spacy. However, it would be great to have class-wise metrics for Spacy’s models to understand performance better as well as identify what classes should be focussed on during labelling. Do you guys have those numbers handy?
Sorry for the delay getting back to you on this. I snoozed the issue because I wanted to get the numbers for you, but in the absence of that:
The answer unfortunately is no, we don’t have the metrics computed. I’d like to update our scoring script to output a json file with more detailed reporting. As a rule of thumb, the rarest entity types are always going to have the worst performance per type.