ValueError on Vectors.most_similar

[ZeeD]

Hi.
I'm trying to understand if it's possible to use spacy to find synonyms. I found that I could use Vectors.most_similar, but I'm having troubles with even a basic code.

from spacy import load
import numpy

nlp = load('en_core_web_sm')
queries = numpy.asarray([numpy.random.uniform(-1, 1, (300,))])
most_similar = nlp.vocab.vectors.most_similar(queries, n=10)

This snippet is basically the example I found on the most_similar documentation, but if I run it, it ends with this error:

  File "c:\Program Files\Python37\lib\runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "c:\Program Files\Python37\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "C:\Users\vito.detullio\Desktop\workspace-srm-cognitive\_prove\spacynltk.py", line 6, in <module>
    most_similar = nlp.vocab.vectors.most_similar(queries, n=10)
  File "spacy\vectors.pyx", line 479, in spacy.vectors.Vectors.most_similar
ValueError: [E198] Unable to return 10 most similar vectors for the current vectors table, which contains 0 vectors.

can you help me understanding what I'm doing "wrong"?

[ZeeD]

Ok, found the "culprit" - I'm using en_core_web_sm that has

0 keys, 0 unique vectors (0 dimensions)

switching to a different model, such as en_core_web_md the error is gone

[polm]

Yes, the small models don't have word vectors.

[ZeeD]

Coming back to my "original" problem, I have written a simple script, to find word "synonims"

from numpy import asarray
from spacy import load  # @UnresolvedImport

nlp = load('en_core_web_md')

words = ['cinema']
keyss, _, scoress = nlp.vocab.vectors.most_similar(asarray([nlp(word).vector
                                                            for word in words]),
                                                   n=10)
for word, keys, scores in zip(words, keyss, scoress):
    strings = [nlp.vocab.strings[key] for key in keys]
    print(f'word={word!r}')
    for string, score in zip(strings, scores):
        print(f'\tstring={string!r}, score={score!r}')

this snippet "works" - as in extract something.
my theoretical basis unfortunately is little, so I would like to know... what have I extracted? are word with "similar vectors" actually comparable to synonims?
Are the results (both "teoretically" and "pratically") comparable to the ones I could get, for example, from wordnet synsets?

[polm]

Are the results (both "teoretically" and "pratically") comparable to the ones I could get, for example, from wordnet synsets?

No, they are not going to be synonyms in the traditional sense. The best you can hope for is that they are words that are used in similar contexts (= surrounding words). A simple example of how this differs from real synonyms is that antonyms in the same category, like "black" and "white", will usually be close in embedding space.

There's a lot of research on this topic, you might search for "word vectors" and "synonyms" or "antonyms", or check the Jurafsky and Martin book.

https://web.stanford.edu/~jurafsky/slp3/