Helper functions to initialise Embedding layer with pre-trained GloVe embeddings.

tf_glove_embeddings.py

import zipfile
from io import TextIOWrapper
import numpy as np
import tensorflow as tf
import os

GLOVE_EMBEDDINGS = {
    'glove.6B.50d'   : ('http://nlp.stanford.edu/data/glove.6B.zip'        , 50),
    'glove.6B.100d'  : ('http://nlp.stanford.edu/data/glove.6B.zip'        , 100),
    'glove.6B.200d'  : ('http://nlp.stanford.edu/data/glove.6B.zip'        , 200),
    'glove.6B.300d'  : ('http://nlp.stanford.edu/data/glove.6B.zip'        , 300),
    'glove.42B.300d' : ('http://nlp.stanford.edu/data/glove.42B.300d.zip'  , 300),
    'glove.840B.300d': ('http://nlp.stanford.edu/data/glove.840B.300d.zip' , 300),
}

def download_glove_embeddings_index(pretrained_name):
    try:
        url, dimension = GLOVE_EMBEDDINGS[pretrained_name]
    except KeyError:
        raise ValueError(f"Unknown pretrained embedding name '{pretrained_name}'")
    
    archive_fname = os.path.basename(url)
    archive_fname = tf.keras.utils.get_file(archive_fname, url)

    embeddings_index = {}
    with zipfile.ZipFile(archive_fname) as z:
        with z.open(f'{pretrained_name}.txt') as f:
            f = TextIOWrapper(f, encoding='utf-8')
            for line in f:
                values = line.split()
                word = values[0]
                coefs = np.asarray(values[1:], dtype='float32')
                embeddings_index[word] = coefs
    return embeddings_index, dimension

# get glove coeff matrix
def get_glove_embeddings(tok, pretrained_name='glove.6B.300d'):
    embeddings_index, embeddings_dim = download_glove_embeddings_index(pretrained_name)
    if tok.num_words:
        max_num_words = tok.num_words
    else:
        max_num_words = len(tok.word_index) + 1
    word_index = tok.word_index
    num_words = min(max_num_words, len(word_index)+1)
    embeddings_matrix = np.random.uniform(low=-0.25, high=0.25, size=(num_words, embeddings_dim))
    for word, i in word_index.items():
        if i >= max_num_words:
            continue
        embedding_vector = embeddings_index.get(word)
        if not embedding_vector is None:
            embeddings_matrix[i] = embedding_vector
    # return: (input_dim, output_dim, embeddings_matrix)
    return (num_words, embeddings_dim, embeddings_matrix)