def load_pretrained_embeddings()

in ludwig/utils/data_utils.py [0:0]

• 22 lines of code
• 7 McCabe index (conditional complexity)

def load_pretrained_embeddings(embeddings_path, vocab):
    embeddings = load_glove(embeddings_path)

    # find out the size of the embeddings
    embeddings_size = len(next(iter(embeddings.values())))

    # calculate an average embedding, to use for initializing missing words
    avg_embedding = np.zeros(embeddings_size)
    count = 0
    for word in vocab:
        if word in embeddings:
            avg_embedding += embeddings[word]
            count += 1
    if count > 0:
        avg_embedding /= count

    # create the embedding matrix
    embeddings_vectors = []
    for word in vocab:
        if word in embeddings:
            embeddings_vectors.append(embeddings[word])
        else:
            embeddings_vectors.append(
                avg_embedding + np.random.uniform(-0.01, 0.01, embeddings_size)
            )
    embeddings_matrix = np.stack(embeddings_vectors)

    # let's help the garbage collector free some memory
    embeddings = None

    return embeddings_matrix