Prodigy with scispacy models

, , ,
1

I am totally new to Prodigy

My environment:
Python: 3.7.0
Spacy: 2.3.5
Scispacy: 0.2.5
Prodigy: 1.10.6

I used the spacy model "en_core_web_sm" to annotate a given text with no issue:
prodigy ner.teach ner_news en_core_web_sm ./news_headlines.jsonl --label PERSON,EVENT

However, I have a different type of text "biomedical" and I would like to use one of the scispacy model for annotation e.g., "en-ner-bc5cdr-md" for disease and chemical. I though scispacy models will be treated as spacy models in Prodigy?

I installed the "en-ner-bc5cdr-md-0.2.0" and I used:
prodigy ner.teach ner_disease en-ner-bc5cdr-md ./dis_source_text.jsonl --label DISEASE,CHEMICAL

I got this error: "Can't find model 'en-ner-bc5cdr-md'. It doesn't seem to be a shortcut link, a Python package or a valid path to a data directory". However, the models works fine when I load it in the same environment in a python code.
nlp = spacy.load("en_ner_bc5cdr_md")

with pip list, the model is there as it is the case with "en_core_web_sm"

any ideas in how to solve the issue?

2

Hi!

I think you need to specify the model with underscores instead - i.e. en_ner_bc5cdr_md - could you give that a try?

3

Hi
Thanks a lot ! this was the mistake.

4

Sure, no worries! Happy to hear it works now.