Reading Glove Embeddings

read_glove_embeddings.py

import numpy as np

class Embedding(object):
  def __init__(self, unk_token=None):
    self.unk_token = unk_token
    self.word2id = {unk_token: 0}
    self.id2word = [unk_token]
    self.vectors = []
    
  def __len__(self):
    return len(self.word2id)
  
  def __setitem__(self, word, vector):
    if word not in self.word2id:
      self.word2id[word] = len(self)
      self.id2word.append(word)
      self.vectors.append(vector)      
  
  def __getitem__(self, word):
    word_id = self.word2id.get(word, 0)
    return self.vectors[word_id]
    
  def __contains__(self, word):
    return word in self.word2id
  
  def words_to_ids(self, words):
    ids = [self.word2id.get(w, 0) for w in words]
    return ids
  
  def words_to_vectors(self, words):
    ids = self.words_to_ids(words)
    vectors = self.vectors[ids]
    return vectors
    
  def vectors_to_numpy(self):
    self.vectors = np.zeros_like(self.vectors[:1]).tolist() + self.vectors
    self.vectors = np.asarray(self.vectors)
    assert len(self.vectors.shape) == 2, "Some error occured in vectorization"    


def read_embeddings(path, all_vocab):
  embedding = Embedding()
  with open(path) as fp:
    for line in fp:
      line = line.split("\t")
      assert len(line) > 1, f"Incorrect line: {line}"
      word = line[0]
      vector = [float(x) for x in line[1:]]
      if word in all_vocab:
        embedding[word] = vector
      if len(embedding) > len(all_vocab):
        break
  embedding.vectors_to_numpy()
  return embedding