Is it possible to use gensim word2vec model in deeplearning4j.word2vec?

Question

I'm new to deeplearning4j, i want to make sentence classifier using words vector as input for the classifier. I was using python before, where the vector model was generated using gensim, and i want to use that model for this new classifier. Is it possible to use gensim's word2vec model in deeplearning4j.word2vec and how i can do that?

Answer 1

Yes, it's possible since Word2Vec implementation defines a standard to structure its model.

To do this:

1. Using gensim, save the model compatible with Word2Vec implementation:

   w2v_model.wv.save_word2vec_format("path/to/w2v_model.bin", binary=True)
   

2. From DL4J, load the same pre-trained model:

   Word2Vec w2vModel = WordVectorSerializer.readWord2VecModel("path/to/w2v_model.bin");
   

In fact, you could test the model in both codes and you should see the same results, for instance:

With gensim:

print(w2v_model.most_similar("love"))
print(w2v_model.n_similarity(["man"], ["king"]))

And with DL4J:

System.out.println(w2vModel.wordsNearest("love", 10));
System.out.println(w2vModel.similarity("man", "king"));