Notes on fasttext

Some gist about fasttext:

  • Basically 3 packages, wordvector, text classification and compression.
  • Text classifications is really comparable with other deep methods.  Another Web's wisdom is here.
  • Running the tasks are trivial for proficient unix users.  So I don't want to repeat them here.  The examples also run end-to-end and they are fast.
  • Unlike what I thought though, fasttext doesn't quite setup a deep-learning-based word-classification, but as I said, that's not the point.
  • Compression was known to be so good such that it can fit to be embedded devices.
  • Users also got granted patents to use the source code freely.  So good stuffs.

Some other nice resources one can follow:

http://debajyotidatta.github.io/nlp/deep/learning/word-embeddings/2016/09/28/fast-text-and-skip-gram/

http://sebastianruder.com/word-embeddings-1/index.html#continuousbagofwordscbow

http://papers.nips.cc/paper/5021-distributed-representations-of-words-and-phrases-and-their-compositionality.pdf

http://textminingonline.com/fasttext-for-fast-sentiment-analysis

 

Leave a Reply

Your email address will not be published. Required fields are marked *