9

I have a text file with my precomputed word vectors in the following format (example):

word -0.0762464299711 0.0128308048976 ... 0.0712385589283\n’

on each line for every word (with 297 extra floats in place of the ...). I am trying to load these with Gensim as KeyedVectors, because I ultimately would like to compute the cosine similarity, find most similar words, etc. Unfortunately I have not worked with Gensim before and from the documentation it's not quite clear to me how to do this. I have tried the following which I found here:

word_vectors = KeyedVectors.load_word2vec_format('/embeddings/word.vectors', binary=False)

However this gives the following error:

ValueError: invalid literal for int() with base 10: 'the'

'the' is the first word in the text file, so I suspect that the loading function is expecting something to be there that is not. But I can't find any information on what should be there. I would highly appreciate a pointer to such information or any other solution to my problem. Thanks!

iloveseals
  • 93
  • 1
  • 4

1 Answers1

8

You can see here an example of Word2Vec format. The first line is supposed to contain the number of words you have in your file followed by the dimension of your vectors. This is probably why your script is returning you an error.

In your example :

1 300
word -0.0762464299711 0.0128308048976 ... 0.0712385589283
EtienneG
  • 300
  • 5
  • 14
  • 7
    Thanks! That's exactly the information I needed! Also I found out that my text file was actually in the GloVe format, so I ended up using these lines: ```from gensim.scripts.glove2word2vec import glove2word2vec glove2word2vec(glove_file, tmp_file)``` After that, I could import my vectors with ```KeyedVectors.load_word2vec_format(tmp_file)```. – iloveseals Apr 10 '18 at 16:23
  • Thanks! The comment was my solution as well! :-) – Sandro_V Sep 24 '19 at 08:46