word2vec about variational-xray-report-gen HOT 7 CLOSED

You can download the file from here https://github.com/ncbi-nlp/BioSentVec ([BioWordVec vector] https://ftp.ncbi.nlm.nih.gov/pub/lu/Suppl/BioSentVec/BioWordVec_PubMed_MIMICIII_d200.vec.bin).

from variational-xray-report-gen.

ok thank you very much! I downloaded some file from gensim related to word2vec, but then it weighed only 1GB. and when I use it for training I am getting stuck at the gradient backpropagation.

Another issue I faced was when calculating priors I was getting negative values of the variance from logit outputs. But then I added sigmoid at the output of priors. But I guess all issues probably started because of me using the wrong word2vec.

Also, the one you use in the code is .pkl file. Is it just pickled version of the file you shared?

from variational-xray-report-gen.

model.vectors.shape
Out[20]: (16545452, 200)

That's what I am getting when I am reading the file using gensim's keyedvectors. Does that look right? So I just pass this vector as embedding initialization?

from variational-xray-report-gen.

I tested it with these embeddings, and it still falls because of negative values of prior's variance

from variational-xray-report-gen.

Hello! Thank you for sharing your work! Running the training requires bio_word2vec_weights.pkl file. Can you please tell where I can get this file?

Hi,did you find the bio_word2vec_weights.pkl file?

from variational-xray-report-gen.

Hello! Thank you for sharing your work! Running the training requires bio_word2vec_weights.pkl file. Can you please tell where I can get this file?

Hi,did you find the bio_word2vec_weights.pkl file?

Hi, have you solved the problem yet?

from variational-xray-report-gen.

Hi. I assume the format of the file for these pre-trained weights provided here https://github.com/ncbi-nlp/BioSentVec, was changed meanwhile. Maybe you can try using the currently provided format and load them like that, possibly using fasttext library.

from variational-xray-report-gen.