pyfastText

• description

  • test code for fastText.py(python wrapper of fastText)
  • installation

  pip install fasttext

• word vector

  • train

  python train_wordvec.py --train data.txt --model model --type=skipgram
  
  # you can adjust hyper-parameters by editing train_wordvec.py
  # for example,
  #   dim=100, epoch=5, ... 
  # data.txt is just a big text file like 'text9'

  • lookup

  python lookup_wordvec.py --model=model.bin
  facebook
  [-0.21015416085720062, 0.3572459816932678, ..., 0.40391290187835693, 0.2624731957912445]

• classification

  • train

  python train_classifier.py --train=amazon_review_polarity.train --model=polarity
  
  # training files are downloadable via https://github.com/facebookresearch/fastText/blob/master/classification-results.sh 

  • evaluate

  python evaluate_classifier.py --model=polarity.bin --test=amazon_review_polarity.test
  P@1: 0.9110425
  Number of examples: 400000

  • predict

  python predict.py --model=polarity.bin
  bad design , the rubber/plastic material ripped within moments at two points where the volume and phone jacks meet .
  [u'__label__1']