Export Word2Vec from dl4j to vector.tsv and metadata.tsv for tensorflow Embedding projector?

Question

I have trained a Word2Vec embedding using deeplearning4j.

How can I export the resulting Word2Vec model to a vectors.tsv and metadata.tsv so I can visualize it using the https://projector.tensorflow.org/ ?

Answer 1

try this out.

# your word2vec model loaded in `w2v_model`
with open('tensors.tsv', 'w') as tensors:
    with open('metadata.tsv', 'w') as metadata:
        for word in w2v_model.wv.index2word:
            metadata.write(word + '\n')
            vector_row = '\t'.join(map(str, w2v_model.wv[word]))
            tensors.write(vector_row + '\n')