rodekruis/anonymization-app

spaCy ensemble

Opened this issue · 0 comments

import spacy

nlp_fast = spacy.load("en_core_web_sm")  # load the fast pipeline
nlp_medium = spacy.load("en_core_web_md")  # load the medium pipeline
nlp_slow = spacy.load("en_core_web_lg")  # load the slow pipeline

assert nlp_slow.vocab["cat"].vector == nlp_medium.vocab["cat"].vector

nlp_slow.add_pipe(
    "ner",
    name="ner_medium",
    source=nlp_medium,
    after="ner",
)
nlp_slow.add_pipe(
    "ner",
    name="ner_fast",
    source=nlp_fast,
    after="ner",
)

nlp_fast_and_slow = nlp_slow

doc = nlp_fast_and_slow(
    "My David Berentein and I love NLP.")