Skip to main content

A Python interface for Facebook fastText library

Project description

fasttext is a Python interface for Facebook fastText.

Requirements

fasttext support Python 2.6 or newer. It requires Cython in order to build the C++ extension.

Installation

pip install fasttext

Example usage

This package has two main use cases: word representation learning and text classification.

These were described in the two papers 1 and 2.

Word representation learning

In order to learn word vectors, as described in 1, we can use fasttext.skipgram and fasttext.cbow function like the following:

import fasttext

# Skipgram model
model = fasttext.skipgram('data.txt', 'model')
print model.words # list of words in dictionary

# CBOW model
model = fasttext.cbow('data.txt', 'model')
print model.words # list of words in dictionary

where data.txt is a training file containing utf-8 encoded text. By default the word vectors will take into account character n-grams from 3 to 6 characters.

At the end of optimization the program will save two files: model.bin and model.vec.

model.vec is a text file containing the word vectors, one per line. model.bin is a binary file containing the parameters of the model along with the dictionary and all hyper parameters.

The binary file can be used later to compute word vectors or to restart the optimization.

The following fasttext(1) command is equivalent

# Skipgram model
./fasttext skipgram -input data.txt -output model

# CBOW model
./fasttext cbow -input data.txt -output model

Obtaining word vectors for out-of-vocabulary words

The previously trained model can be used to compute word vectors for out-of-vocabulary words.

print model.get_vector('king')
# or just use a nice syntax
print model['king'] # get the vector of the word 'king'

the following fasttext(1) command is equivalent:

echo "king" | ./fasttext print-vectors model.bin

This will output the vector of word king to the standard output.

Load pre-trained model

We can use fasttext.load_model to load pre-trained model:

model = fasttext.load_model('model.bin')
print model.words # list of words in dictionary
print model['king'] # get the vector of the word 'king'

Text classification

This package can also be used to train supervised text classifiers and load pre-trained classifier from fastText.

In order to train a text classifier using the method described in 2, we can use the following function:

classifier = fasttext.supervised('data.train.txt', 'model')

equivalent as fasttext(1) command:

./fasttext supervised -input data.train.txt -output model

where data.train.txt is a text file containing a training sentence per line along with the labels. By default, we assume that labels are words that are prefixed by the string __label__.

We can specify the label prefix with the label_prefix param:

classifier = fasttext.supervised('data.train.txt', 'model', label_prefix='__label__')

equivalent as fasttext(1) command:

./fasttext supervised -input data.train.txt -output model -label '__label__'

This will output two files: model.bin and model.vec.

Once the model was trained, we can evaluate it by computing the precision at 1 (P@1) on a test set using classifier.test function:

precision_at_one, nexamples = classifier.test('test.txt')
print 'P@1:', precision_at_one
print 'Number of examples:', nexamples

This will print the same output to stdout as:

./fasttext test model.bin test.txt

In order to obtain the most likely label for a list of text, we can use classifer.predict method:

texts = ['example very long text 1', 'example very longtext 2']
labels = classifier.predict(texts)
print labels

This interface is equivalent as fasttext(1) predict command. The same model with the same input set will have the same prediction.

API documentation

Skipgram model

Train & load skipgram model

model = fasttext.skipgram(params)

CBOW model

Train & load CBOW model

model = fasttext.cbow(params)

Load pre-trained model

File .bin that previously trained or generated by fastText can be loaded using this function

model = fasttext.load_model('model.bin')

Attributes and methods for the model

Skipgram and CBOW model have the following atributes & methods

model.model_name       # Model name
model.words            # List of words in the dictionary
model.dim              # Size of word vector
model.ws               # Size of context window
model.epoch            # Number of epochs
model.min_count        # Minimal number of word occurences
model.neg              # Number of negative sampled
model.word_ngrams      # Max length of word ngram
model.loss_name        # Loss function name
model.bucket           # Number of buckets
model.minn             # Min length of char ngram
model.maxn             # Max length of char ngram
model.lr_update_rate   # Rate of updates for the learning rate
model.t                # Value of sampling threshold
model.get_vector(word) # Get the vector of specified word
model[word]            # Get the vector of specified word

Supervised model

Train & load the classifier

classifier = fasttext.supervised(params)

Load pre-trained classifier

File .bin that previously trained or generated by fastText can be loaded using this function.

./fasttext supervised -input train.txt -output classifier -label 'some_prefix'
classifier = fasttext.load_model('classifier.bin', label_prefix='some_prefix')

Test classifier

This is equivalent as fasttext(1) test command. The test using the same model and test set will produce the same value for the precision at one and the number of examples.

precision_at_one, nexamples = classifier.test(test_file)

Predict the most-likely label of texts

This interface is equivalent as fasttext(1) predict command.

texts is an array of string

labels = classifier.predict(texts)

Attributes and methods for the classifier

Classifier have the following atributes & methods

classifier.labels         # List of labels
classifier.label_prefix   # Prefix of the label
classifier.dim            # Size of word vector
classifier.ws             # Size of context window
classifier.epoch          # Number of epochs
classifier.min_count      # Minimal number of word occurences
classifier.neg            # Number of negative sampled
classifier.word_ngrams    # Max length of word ngram
classifier.loss_name      # Loss function name
classifier.bucket         # Number of buckets
classifier.minn           # Min length of char ngram
classifier.maxn           # Max length of char ngram
classifier.lr_update_rate # Rate of updates for the learning rate
classifier.t              # Value of sampling threshold
classifier.test(filename) # Test the classifier
classifier.predict(texts) # Predict the most likely label

Params

List of available params and their default value:

For Skipgram, CBOW and Supervised model
input          training file path
output         output file path
lr             learning rate [0.05]
lr_update_rate change the rate of updates for the learning rate [100]
dim            size of word vectors [100]
ws             size of the context window [5]
epoch          number of epochs [5]
min_count      minimal number of word occurences [1]
neg            number of negatives sampled [5]
word_ngrams    max length of word ngram [1]
loss           loss function {ns, hs, softmax} [ns]
bucket         number of buckets [2000000]
minn           min length of char ngram [3]
maxn           max length of char ngram [6]
thread         number of threads [12]
t              sampling threshold [0.0001]
silent         disable the log output from the C++ extension [1]

For Supervised model only
label_prefix   Prefix of the label name [__label__]

References

Enriching Word Vectors with Subword Information

[1] P. Bojanowski*, E. Grave*, A. Joulin, T. Mikolov, *Enriching Word Vectors with Subword Information*

@article{bojanowski2016enriching,
  title={Enriching Word Vectors with Subword Information},
  author={Bojanowski, Piotr and Grave, Edouard and Joulin, Armand and Mikolov, Tomas},
  journal={arXiv preprint arXiv:1607.04606},
  year={2016}
}

Bag of Tricks for Efficient Text Classification

[2] A. Joulin, E. Grave, P. Bojanowski, T. Mikolov, *Bag of Tricks for Efficient Text Classification*

@article{joulin2016bag,
  title={Bag of Tricks for Efficient Text Classification},
  author={Joulin, Armand and Grave, Edouard and Bojanowski, Piotr and Mikolov, Tomas},
  journal={arXiv preprint arXiv:1607.01759},
  year={2016}
}

(* These authors contributed equally.)

Join the fastText community

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page