Bug Fix 1779

[saroufimc1]

• Not sure about the header.
• Keeping the whole embedding matrix for subwords, and only creating a dictionary for the subwords using the hashing function. In the previous version, the index was incorrectly assigned in the subword dictionary.

[menshikh-iv]

@manneshiva @jayantj can you look into this PR?

[manneshiva]

@saroufimc1 You might be mixing up two issues here -- #1261 and #1779. It is best to have a different PR for each of these issues. As #1261 still needs more discussion to agree upon a solution, I suggest you only address issue #1779 here -- i.e. correct the assignment of model.wv.syn0_ngrams after trimming of unused ngrams (model.wv.syn0_ngrams.shape[0] <= self.bucket). Also, don't forget to add a test case in gensim/test/test_afsttext_wrapper.py to catch this error in future.

[saroufimc1]

@manneshiva Sure, will do that.
Thanks,
Carl

[piskvorky]

Needs a better PR title and description.

Issue link belongs to description (Github's hyperlink), not the title.

[menshikh-iv]

ping @saroufimc1, what's the status of this PR?

[menshikh-iv]

ping @saroufimc1, what's the status here?

[menshikh-iv]

I close this PR because this looks abandoned.
@saroufimc1 feel free to re-open when you continue this.

[saroufimc1]

Sorry was on vacations. Uploading my fix today. Thanks.