Central word and context

[刘士渤]

Excuse me，if I want the window to see the whole document and set it large enough，will the gensim choose every word to be the central word and other words to be context？

[Gordon Mohr]

See the documentation for the parameters `window` & `shrink_window` for ways to arrange for a much-larger (including full-length-of-text) window where all words have equal weight.

In particular, if `window` is larger than your largest text (and the `Word2Vec`/`Doc2Vec`/`FastText` models only support 10,000-token texts), every word could be considered for every context.

And, if `shrink_windows` is turned off (set to `0` or `False`), then the default technique of always using some random smaller-than-`window` window each context, as an efficient way to weight nearer words more, will be turned off – meaning every word within `window` token-positions will be considered an equal part of the context. 

Note that such large & never-shrunk windows will result in relatively-longer runtimes. 

- Gordon

[刘士渤]

Thank you very much，by the way what optimizer does gensim.models.Word2Vec use, is it SGD?

[Gordon Mohr]

Yes, the training loop of Gensim's Word2Vec (closely following the original word2vec.c released by the Google researchers with the word2vec paper) is essentially SGD.

- Gordon