word2vec

to_embeddings.py

# Initialize word2vec. Context is taken as the 2 previous and 2 next words
model = Word2Vec(sentences, window=5, size=100, workers=4)
model.train(sentences, total_examples=len(sentences), epochs=1000)

# get ordered vocabulary list
voc = model.wv.index2word

# get vector size
dim = model.vector_size

# get most similar words
sim = model.wv.most_similar('holmes')

# Convert to numpy 2d array (n_vocab x vector_size)
def to_embeddings_Matrix(model):  
    embedding_matrix = np.zeros((len(model.wv.vocab), model.vector_size))
    word2idx = {}
    for i in range(len(model.wv.vocab)):
        embedding_matrix[i] = model.wv[model.wv.index2word[i]]
        word2idx[model.wv.index2word[i]] = i
    return embedding_matrix, model.wv.index2word, word2idx