Give this component a copy of its own tok2vec

[koayst]

Hello,

This question is related to: projects/tutorials/ner_double/. Part of the code as shown below.

import spacy

nlp = spacy.load("en_core_web_md") # load the base pipeline
drug_nlp = spacy.load("en_ner_drugs") # load the drug pipeline

give this component a copy of its own tok2vec
drug_nlp.replace_listeners("tok2vec", "ner", ["model.tok2vec"])

The last comment said "give this component a copy of its own tok2vec". Isn't a tok2vec object of ["model.tok2vec"] is instantiated to replace the tok2vec layer in "ner" component ?

I did a drug_nlp.pipe_names and it returned -> ['tok2vec', 'ner'].

[koayst]

I think I know the answer.

Looking at the config.cfg file again. My understanding now is "ner_drug" uses another tok2vec component. There are 2 independent tok2vec components loaded into the pipeline. One tok2vec for "nlp" and other tok2vec for drug_nlp.

Please comment if my understanding is correct.

[polm]

That's correct!

Statistical components depend on the tok2vec they were trained with, so if you move them to another pipeline you need to bring the tok2vec along.