Adding custom wordvectors to my pipeline

[Gitclop]

I've build custom wordvectors and created a blank model with
spacy init vectors
Now if i load the model will i loose all the components from the stock models, like de_core_news_sm?
can i use something like
nlp.add_pipe('vectors', source=pathToVectors)
?
Or how would i just load the custom vectors onto my existing pipeline?

[Gitclop]

I got it working by iterating the gensim model and adding each vector manually to a new spacy vocab.

from gensim.models import Word2Vec
from spacy.vocab import Vocab

gensim_model = Word2Vec.load(my_w2vmodel.model)
vocab = Vocab()

for word in gensim_model.wv.index_to_key:
      vector = gensim_model.wv.get_vector(word)
      vocab.set_vector(word, vector)

nlp.vocab.vectors = vocab.vectors

I would like to know if this could cause problems with other pipeline components and what would be the prefered way to load custom wordvectors in spacy.

[svlandeg]

Hi! You can indeed either use spacy init vectors or vocab.set_vector. Whether or not this will influence existing trained components depends on their settings. See the docs here - if include_static_vectors is set to True and you're changing the vectors after training, then that might negatively influence the model's performance.