Created by: GoogleCodeExporter

Just released a Ruby module that builds an index of a binary word2vec vector 
file, so your code can seek directly to the right position in the file for a 
given word or term. For example, the word "/en/italy" in the English 
"freebase-vectors-skipgram1000-en.bin" file is at byte position 116414 position.

The module also computes a locally-sensitive hash for each vector in a binary 
word2vec file, so you can do a nearest neighbor search (i.e. cosine distance) 
much faster. I get a couple orders of magnitude better performance on my 
machine, with a 10 bit random projection LSH.

https://github.com/someben/treebank/blob/master/src/build_word2vec_index.rb

Thanks for the project, Tomas.

Best,
Ben

Original issue reported on code.google.com by bgimp...@googlemail.com on 23 Sep 2013 at 3:48