load_vectors does not take keyword arguments?

[jwohlwend]

Looking at the function definition for load_vectors, it doesn't seem to take optional arguments, so I'm wondering how line 206 in Vocab.py can work:

self.load_vectors(vectors, unk_init=unk_init, expand_vocab=expand_vocab)

It would be nice if load vectors took all arguments to Vectors, so as to specify cache, or unk_init as above. Please let me know if I'm missing something..

[nelson-liu]

Yes, that looks like a bug --- I'm assuming @jekbradbury wrote that when there was a different API for load_vectors.

It would be nice if load vectors took all arguments to Vectors, so as to specify cache, or unk_init as above. Please let me know if I'm missing something..

I'm a bit confused -- can you explain your rationale for doing this? Couldn't you just pass cache or unk_init when creating a Vector object, then pass the Vector to load_vectors? Or did you want to do something like have a different value than default for one of the pretrained vector aliases (e.g. if you wanted a different unk_init for glove)?

I agree that it'd be nice to have for the latter case, but a temporary workaround would be to just manually instantiate the Vector class in question and pass these kwargs in (see this dict for what each alias refers to).

[jwohlwend]

Yes, I had the latter in mind. The reason I ran into this was that I was using the pretrained embeddings in two different projects and realized that I was downloading the word vectors twice instead of using a single cache folder.

I was loading the string alias from a config file, which is why I thought it'd be nice to have that option in the load_vectors function, but you're right there's no particular reason that I can't make my own aliases to instantiate the Vector class with the appropriate kwargs. Thanks for your response!

[allenanie]

Encountering a similar problem but with unk_init argument. It seems like the unknown keyword defaults to 0 for now, which is fine, but in certain settings it might be better to randomly initialize unknown keywords.