6 comments

[ 3.4 ms ] story [ 25.1 ms ] thread
Jupyter notebook with some numbers on performance and accuracy of fastText:

https://github.com/jayantj/gensim/blob/fast_text_notebook/do...

This compares fastText against the word2vec implementation in gensim, on the syntactic/semantic word analogy task.

(raw pull request with comments: https://github.com/RaRe-Technologies/gensim/pull/815)

Note that this is only looking at the embeddings side of FastText (which looks quite impressive).

The classification side is equally interesting, and is better compared to things like VW, eg http://nlpers.blogspot.com.au/2016/08/fast-easy-baseline-tex...

VW is a pretty strong baseline here, so equaling it is a good accomplishment.

That notebook is great! thanks for sharing.
(comment deleted)