total_vec don't exists

[set92]

I wanted to save the most important vectors or similar words, but I tried to cut the length of the embeddings using model.save_word2vec_format('try.c.w2v', total_vec=10000) but it says total_vec doesn't exists, but it's in the documentation.

[menshikh-iv]

Hi @set92, please upgrade your gensim version pip install -U gensim, this feature comes from 2.1.0.
FYI - documentation always for latest version (right now - for 3.1.0)

[set92]

Im in 3.1.0, and before I was in 3.0.1 but I get the same error. I load a model using model = gensim.models.Word2Vec.load('../../models/model_w2v_medic_v2.bin') and then I try to save it with model.save_word2vec_format('try.c.w2v', total_vec=10000) and it gives the error TypeError: save_word2vec_format() got an unexpected keyword argument 'total_vec'

[menshikh-iv]

@set92 you called this one (method of Word2Vec class), but add link to method of KeyedVectors class.
Please try model.wv.save_word2vec_format('try.c.w2v', total_vec=10000)

[set92]

Yes, you are right, I forgot about wv in model.wv.save_word2vec_format, thanks a lot.

[piskvorky]

@menshikh-iv what's the difference between model.save_word2vec_format and model.wv.save_word2vec_format? That's some really strange API, we should fix that.

[menshikh-iv]

@piskvorky we should remove this deprecated methods for avoiding confusion, don't fix.

[piskvorky]

Sure, but how can this even happen? How did they get out of sync?

I was under the impression the deprecated version is just an alias of the new one (until we remove it for good).

[menshikh-iv]

@piskvorky

I was under the impression the deprecated version is just an alias of the new one (until we remove it for good).

From the beginning, that's exactly the way it was (i.e. full sync of interfaces), but later - #1256 this PR added total_vec param to KeyedVector.save_word2vec_format https://github.com/RaRe-Technologies/gensim/pull/1256/files#diff-741f69634ff2e1b2c4ce544e0d82a1e6R121 and doesn't update "deprecated" Word2Vec.save_word2vec_format

[piskvorky]

Yeah, time to remove it.

Where do we keep the list of deprecated things + their schedule for removal (version)? Or do you just grep all source code before every release to find them?

[menshikh-iv]

@piskvorky I have no idea, I assume that there is no such list (all these "deprecations" make before I came).

The first time, docstrings "forwarding" was added in #1251 (April 11), explicit deprecation added in #1274 (May 3), that looks like a fix for the previous PR.

[menshikh-iv]

My plans are release minor (feature version) on next week and major (with change package structure, removing outdated stuff, etc) in next year (probably in January).

[piskvorky]

Let's append a clear deprecation plan with each deprecation: in which version will the code be actually removed? (3.2.0 etc)

And then before each release, grep the code base for deprecations and actually remove the scheduled ones.