Can you be a bit more specific about using insurance_qa_embeddings.py?

[wailoktam]

Hi, thanks for your great work and support. I have got insurance_qa_evaluation.py running with Kera's own embedding layer. How do I bridge insurance_qa_embeddings.py to insurance_qa_evaluation.py?
I want to substitute parts of the input sentences with their synonyms or antonyms and see how training and prediction goes? Any suggestion on how it can be done?

[codekansas]

Hmm that seems pretty interesting, let me know how it goes. Embeddings can be trained in different ways, or you can use the ones I provided. You can make your own embeddings and specify them here. insurance_qa_embeddings.py is a stand-alone file for generating embeddings using Word2Vec, and it saves the resulting embeddings in a particular location which you can load elsewhere. So you could mess with the embeddings, save the file, then load it and run the program.

[wailoktam]

Hi, thanks for your prompt reply.

So the following line is necessary for using the embedding layer in Keras?

'initial_embed_weights': np.load('word2vec_100_dim.embeddings')

Is my understanding correction: Keras won't train its own word embeddings. We need to supply the weights generated from word2vec/glovec by the above line ?

However, is the file "models/word2vec_100_dim.h5" in insurance_qa_embeddings.py renamed to
"'word2vec_100_dim.embeddings'"in insurance_qa_evaluation.py or are they different files?

If I want to use a different dataset, I need to generate a different "'word2vec_100_dim.embeddings', right? How should I do it? I know how to save a model in Genism but this "word2vec_100_dim.embeddings" is not the model file yield from the save command in Genism, right?

Many thanks in advance.

[codekansas]

Keras will train its own word embeddings, it just works better if you start with word2vec embeddings (you can choose whether or not they are trained as well). I think they're the same file, probably just a naming difference. And yes, you should generate a new embeddings file. The file word2vec_100_dim.embeddings is, specifically, the embedding part of the Gensim model (consult the code to see what this means).