get_document_topics returns and empty list

[senjed]

I trained the lda model and want to get topic distribution for a new document. However, for some documents the get_document_topics function returns and empty list and I get this warning:

/home/anaconda2/lib/python2.7/site-packages/gensim/models/ldamodel.py:676: RuntimeWarning: overflow encountered in divide
gammad = self.alpha + expElogthetad * np.dot(cts / phinorm, expElogbetad.T)

This is the way I call the function:

topic_vector = [ x[1] for x in self.ldamodel.get_document_topics(new_doc_bow , minimum_probability= 0.0, per_word_topics=False)]

[horpto]

Hi @senjed

What does self.ldamodel.inference([new_doc_bow]) return ?
If this expression returns empty list, so what is the value of self.ldamodel.num_topics ?

[senjed]

Hi,
It returns (array([[nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan]], dtype=float32), None)
and num_topics is 20.

[horpto]

@senjed
Please, try to increase epsilon in gensim/models/ldamodel.py likeit is said in #2115 (comment)

[senjed]

it worked thanks! I am closing the issue.