Get the most similar word in a spaCy model vocabulary based on word vector distance

spacy_get_most_similar_word.py

import numpy as np
import spacy
from spacy.language import Language
from typing import Callable

def get_most_similar_word(nlp: spacy.language.Language, word: str, metric: Callable[[np.array, np.array], float]) -> str:
  # Get the vector of representation of the query word
  vector = nlp.vocab.get_vector(word)

  # Get the most similar vector by row index in the word vector matrix
  index = argmax([metric(vector, v) for v in nlp.vocab.vectors.data])

  # Mapping of row index -> string hashes
  m = {v: k for k, v in nlp.vocab.vectors.key2row.items()}
  
  # Now get the string representation of this hash
  return nlp.vocab.strings.as_string(m[index])