Fix deprecations in SoftCosineSimilarity

[piskvorky]

When running our CI test suite, I see an array of deprecation warnings:

https://dev.azure.com/rare-technologies/gensim-ci/_build/results?buildId=287&view=logs&jobId=f9575ddc-dec8-54e6-9d26-abb8bdd9bed7&j=f9575ddc-dec8-54e6-9d26-abb8bdd9bed7&t=180156a9-2bf9-537d-c84a-ef9e808c0367

Some are from gensim, some from scipy:

@Witiko could you please have a look? Is it something your existing PR already addresses?

If not, can you please fix those? Thanks.

[piskvorky]

…and we're also still using the deprecated smart_open.smart_open apparently, CC @mpenkov :

gensim/test/test_api.py::TestApi::test_load_dataset
gensim/test/test_api.py::TestApi::test_multipart_load
  d:\a\1\s\.tox\py37-win\lib\site-packages\smart_open\smart_open_lib.py:254: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function
    'See the migration notes for details: %s' % _MIGRATION_NOTES_URL

[piskvorky]

And more:

gensim/test/test_datatype.py: 2 warnings
gensim/test/test_doc2vec.py: 14862 warnings
gensim/test/test_fasttext.py: 96 warnings
gensim/test/test_keyedvectors.py: 20 warnings
gensim/test/test_word2vec.py: 84 warnings
  D:\a\1\s\gensim\models\keyedvectors.py:1478: DeprecationWarning: tostring() is deprecated. Use tobytes() instead.
    fout.write(utils.to_utf8(prefix + str(key)) + b" " + row.tostring())

The whole CI log could do with a careful inspection.

[Witiko]

@piskvorky I can remove the deprecated parameters together with their tests and open a PR. Please note that this will break backwards compatibility, i.e. the PR should only be merged if we are going for the next major release of Gensim.

[piskvorky]

Thanks! Yes, we are going for the next major release. Current develop no longer supports py2.x etc, it's all 4.0.0+.

[Witiko]

Most of the deprecation warnings are fixed by #2940 except ...

…and we're also still using the deprecated smart_open.smart_open apparently, CC @mpenkov :

gensim/test/test_api.py::TestApi::test_load_dataset
gensim/test/test_api.py::TestApi::test_multipart_load
  d:\a\1\s\.tox\py37-win\lib\site-packages\smart_open\smart_open_lib.py:254: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function
    'See the migration notes for details: %s' % _MIGRATION_NOTES_URL

... this one, which originates from api.load("__testing_matrix-synopsis", return_path=True) where api is gensim.downloader. This likely needs to be fixed in gensim-data.

[piskvorky]

Awesome, thanks!

except … this one, which originates from api.load("__testing_matrix-synopsis", return_path=True) where api is gensim.downloader.

It's probably the test line a bit lower down, without the return_path=True. That line will load the __testing_matrix-synopsis dataset, which does indeed call smart_open.smart_open():
https://github.com/RaRe-Technologies/gensim-data/releases/tag/__testing_matrix-synopsis

Gensim-data releases are immutable, so we should make a new release of the __testing_matrix-synopsis dataset, with smart_open.smart_open replaced by smart_open.open. Not urgent, and definitely not blocking for this PR.

@mpenkov feel free to merge #2940 when ready.