override _log_progress() in word2vec to track training loss

[usul83]

I came across this suggestion from @villmow in this thread #1172 and thought I'd implement since @menshikh-iv thought it was a good idea and doesn't seem to have been done by anyone else.

First time I've contributed so appologies if not followed correct process etc.

[menshikh-iv]

Hi @usul83,

as I see, this is definition from https://github.com/RaRe-Technologies/gensim/blob/408a714e836484e914a7e3622b4f0f99a6d27ddd/gensim/models/base_any2vec.py#L1148 why this copy-paste needed (this method already inherited automatically)?

[usul83]

It should be a little different from the any2vec version, it adds: self.get_latest_training_loss() / example_count

[menshikh-iv]

@usul83 but in this case, compute_loss must be True, otherwise - this will not work, need to check this case.

[usul83]

It won't work, but it will give a 0 rather than raise an error (since self.running_training_loss = 0 at line 556 whether compute_loss true or not). Maybe could add a check and a warning for that situation?

[menshikh-iv]

No warning needed, simply check, if compute_loss is False - use parent method, otherwise - your.
And also, please train an model with compute_loss=True and submit an log here (I want to see how it will look).

[menshikh-iv]

this PR fully overlapped by #2135, for this reason, I close current.
@usul83 thanks for your work!