LDAModel DepracationWarning under Python 3

[mattilyra]

I noticed that when training an LDA model under Python 3 with NumPy 1.10 or 1.9 I get a long long list of DepracationWarnings from NumPy, so much so that it kills the browser running a notebook. This doesn't happen under Py2 with NumPy 1.10.

I think it's because the ids is a list not a ndarray. The warning refers to two lines in ldamodel.py

• https://github.com/piskvorky/gensim/blob/develop/gensim/models/ldamodel.py#L375
• https://github.com/piskvorky/gensim/blob/develop/gensim/models/ldamodel.py#L401

Both use the ids list to index a ndarray. The ids are just feature ids from the current document (?) so I tried a quick fix of just changing the ids to ndarray which makes the warnings go away, but I haven't tested if it has some other consquences, I don't see why it would.

The DepracationWarnings (just two of them, there are many more actually produced)

/Volumes/LocalDataHD/conda/envs/py34/lib/python3.4/site-packages/gensim/models/ldamodel.py:375: DeprecationWarning: non integer (and non boolean) array-likes will not be accepted as indices in the future
  expElogbetad = self.expElogbeta[:, ids]
/Volumes/LocalDataHD/conda/envs/py34/lib/python3.4/site-packages/gensim/models/ldamodel.py:401: DeprecationWarning: non integer (and non boolean) array-likes will not be accepted as indices in the future
  sstats[:, ids] += numpy.outer(expElogthetad.T, cts / phinorm)

[piskvorky]

I think this may be related to this discussion: #448 (comment)

Yes, we definitely want to get rid of these warnings!

[mattilyra]

So the offender is actually 561 which passes the numpyified doc(s) into .inference()

for chunk_no, chunk in enumerate(utils.grouper(corpus, chunksize, as_numpy=True)):                                                                                                                                                           
    reallen += len(chunk)  # keep track of how many documents we've processed so far
    if eval_every and ((reallen == lencorpus) or ((chunk_no + 1) % (eval_every * self.numworkers) == 0)):
        self.log_perplexity(chunk, total_docs=lencorpus)

Changing as_numpy=False seems to work, my suggested change above actually created some other weirdness.

[tmylk]

These DeprecationWarnings are not present in the latest build. Closing.