how to update the pre-trained word2vec model with new train data using genism

Question

Hi I have used the genism to load the Spanish fasttext word2vec model with following code:

Word2Vec = KeyedVectors.load_word2vec_format('../data/raw/wiki.es.vec',
                                             binary=False)

and now I want to train the word2vec with new training sentences.

I know that in the genism can use the following code to train new sentences if the model is loaded:

 model.build_vocab(sentences)  # prepare the model vocabulary
 model.train(sentences, )

But how to update the model if I used the KeyedVectors.load_word2vec to load the model.

Thanks!

Answer 1

You cannot use keyedvectors for resuming training as per Gensim documentation. You will need to save the full model in previous training run to allow resuming training. Keyedvectors remove training related information and other metadata to reduce memory footprint.