-1

I am running an experiment and I want to know if it is possible to retrain the word2vec Google news vectors using only hyperparameter that i want. And also assign the values that I want to these hyperparameters. Thanks

Jay
  • 11
  • 3
  • 1
    Possible duplicate of [Is it possible to re-train a word2vec model (e.g. GoogleNews-vectors-negative300.bin) from a corpus of sentences in python?](https://stackoverflow.com/questions/35117491/is-it-possible-to-re-train-a-word2vec-model-e-g-googlenews-vectors-negative300) – petezurich Aug 28 '18 at 04:33

1 Answers1

0

Not really, because they haven't shared the exact corpus they used.

And if you like your own corpus better – because it includes more words, or the right word-senses for your application, or even just because it's the only thing you have – you don't really need to "re-train" other vectors, you can just make your own from scratch.

There may be ways to inject those vectors into your model for a "fast start", but that adds complexity for unclear benefits. If your corpus is any good, perhaps even better for your purposes than the one used for GoogleNews, mixing-in the influence from there may be complexity for no net benefit.

gojomo
  • 52,260
  • 14
  • 86
  • 115