Table of contents

• Requirements

• Installation

• Usage overview

• Word representation model

• Text classification model

• IMPORTANT: Preprocessing data / encoding conventions

• More examples

• API

• train_unsupervised parameters

• train_supervised parameters

• model object

Word representation model

In order to learn word vectors, as described here, we can use fasttext.train_unsupervised function like this:

import fasttext

# Skipgram model :
model = fasttext.train_unsupervised('data.txt', model='skipgram')

# or, cbow model :
model = fasttext.train_unsupervised('data.txt', model='cbow')

where data.txt is a training file containing utf-8 encoded text.

The returned model object represents your learned model, and you can use it to retrieve information.

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

model.save_model("model_filename.bin")

model = fasttext.load_model("model_filename.bin")

Text classification model

In order to train a text classifier using the method described here, we can use fasttext.train_supervised function like this:

import fasttext

model = fasttext.train_supervised('data.train.txt')

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__

Once the model is trained, we can retrieve the list of words and labels:

print(model.words)
print(model.labels)

To evaluate our model by computing the precision at 1 (P@1) and the recall on a test set, we use the test function:

def print_results(N, p, r):
    print("N\t" + str(N))
    print("P@{}\t{:.3f}".format(1, p))
    print("R@{}\t{:.3f}".format(1, r))

print_results(*model.test('test.txt'))

We can also predict labels for a specific text :

model.predict("Which baking dish is best to bake a banana bread ?")

By default, predict returns only one label : the one with the highest probability. You can also predict more than one label by specifying the parameter k:

model.predict("Which baking dish is best to bake a banana bread ?", k=3)

If you want to predict more than one sentence you can pass an array of strings :

model.predict(["Which baking dish is best to bake a banana bread ?", "Why not put knives in the dishwasher?"], k=3)

Of course, you can also save and load a model to/from a file as in the word representation usage.

For more information about text classification usage of fasttext, you can refer to our text classification tutorial.

# with the previously trained `model` object, call :
model.quantize(input='data.train.txt', retrain=True)

# then display results and save the new model :
print_results(*model.test(valid_data))
model.save_model("model_filename.ftz")

IMPORTANT: Preprocessing data / encoding conventions

In general it is important to properly preprocess your data. In particular our example scripts in the root folder do this.

fastText assumes UTF-8 encoded text. All text must be unicode for Python2 and str for Python3. The passed text will be encoded as UTF-8 by pybind11 before passed to the fastText C++ library. This means it is important to use UTF-8 encoded text when building a model. On Unix-like systems you can convert text using iconv.

fastText will tokenize (split text into pieces) based on the following ASCII characters (bytes). In particular, it is not aware of UTF-8 whitespace. We advice the user to convert UTF-8 whitespace / word boundaries into one of the following symbols as appropiate.

• space

• tab

• vertical tab

• carriage return

• formfeed

• the null character

The newline character is used to delimit lines of text. In particular, the EOS token is appended to a line of text if a newline character is encountered. The only exception is if the number of tokens exceeds the MAX_LINE_SIZE constant as defined in the Dictionary header. This means if you have text that is not separate by newlines, such as the fil9 dataset, it will be broken into chunks with MAX_LINE_SIZE of tokens and the EOS token is not appended.

The length of a token is the number of UTF-8 characters by considering the leading two bits of a byte to identify subsequent bytes of a multi-byte sequence. Knowing this is especially important when choosing the minimum and maximum length of subwords. Further, the EOS token (as specified in the Dictionary header) is considered a character and will not be broken into subwords.

More examples

In order to have a better knowledge of fastText models, please consider the main README and in particular the tutorials on our website.

You can find further python examples in the doc folder.

As with any package you can get help on any Python function using the help function.

For example

+>>> import fasttext
+>>> help(fasttext.FastText)

Help on module fasttext.FastText in fasttext:

NAME
    fasttext.FastText

DESCRIPTION
    # Copyright (c) 2017-present, Facebook, Inc.
    # All rights reserved.
    #
    # This source code is licensed under the MIT license found in the
    # LICENSE file in the root directory of this source tree.

FUNCTIONS
    load_model(path)
        Load a model given a filepath and return a model object.

    tokenize(text)
        Given a string of text, tokenize it and return a list of tokens
[...]

train_unsupervised parameters

input             # training file path (required)
model             # unsupervised fasttext model {cbow, skipgram} [skipgram]
lr                # learning rate [0.05]
dim               # size of word vectors [100]
ws                # size of the context window [5]
epoch             # number of epochs [5]
minCount          # minimal number of word occurences [5]
minn              # min length of char ngram [3]
maxn              # max length of char ngram [6]
neg               # number of negatives sampled [5]
wordNgrams        # max length of word ngram [1]
loss              # loss function {ns, hs, softmax, ova} [ns]
bucket            # number of buckets [2000000]
thread            # number of threads [number of cpus]
lrUpdateRate      # change the rate of updates for the learning rate [100]
t                 # sampling threshold [0.0001]
verbose           # verbose [2]

train_supervised parameters

input             # training file path (required)
lr                # learning rate [0.1]
dim               # size of word vectors [100]
ws                # size of the context window [5]
epoch             # number of epochs [5]
minCount          # minimal number of word occurences [1]
minCountLabel     # minimal number of label occurences [1]
minn              # min length of char ngram [0]
maxn              # max length of char ngram [0]
neg               # number of negatives sampled [5]
wordNgrams        # max length of word ngram [1]
loss              # loss function {ns, hs, softmax, ova} [softmax]
bucket            # number of buckets [2000000]
thread            # number of threads [number of cpus]
lrUpdateRate      # change the rate of updates for the learning rate [100]
t                 # sampling threshold [0.0001]
label             # label prefix ['__label__']
verbose           # verbose [2]
pretrainedVectors # pretrained word vectors (.vec file) for supervised learning []

model object

train_supervised, train_unsupervised and load_model functions return an instance of _FastText class, that we generaly name model object.

This object exposes those training arguments as properties : lr, dim, ws, epoch, minCount, minCountLabel, minn, maxn, neg, wordNgrams, loss, bucket, thread, lrUpdateRate, t, label, verbose, pretrainedVectors. So model.wordNgrams will give you the max length of word ngram used for training this model.

In addition, the object exposes several functions :

get_dimension           # Get the dimension (size) of a lookup vector (hidden layer).
                        # This is equivalent to `dim` property.
get_input_vector        # Given an index, get the corresponding vector of the Input Matrix.
get_input_matrix        # Get a copy of the full input matrix of a Model.
get_labels              # Get the entire list of labels of the dictionary
                        # This is equivalent to `labels` property.
get_line                # Split a line of text into words and labels.
get_output_matrix       # Get a copy of the full output matrix of a Model.
get_sentence_vector     # Given a string, get a single vector represenation. This function
                        # assumes to be given a single line of text. We split words on
                        # whitespace (space, newline, tab, vertical tab) and the control
                        # characters carriage return, formfeed and the null character.
get_subword_id          # Given a subword, return the index (within input matrix) it hashes to.
get_subwords            # Given a word, get the subwords and their indicies.
get_word_id             # Given a word, get the word id within the dictionary.
get_word_vector         # Get the vector representation of word.
get_words               # Get the entire list of words of the dictionary
                        # This is equivalent to `words` property.
is_quantized            # whether the model has been quantized
predict                 # Given a string, get a list of labels and a list of corresponding probabilities.
quantize                # Quantize the model reducing the size of the model and it's memory footprint.
save_model              # Save the model to the given path
test                    # Evaluate supervised model using file given by path
test_label              # Return the precision and recall score for each label.

The properties words, labels return the words and labels from the dictionary :

model.words         # equivalent to model.get_words()
model.labels        # equivalent to model.get_labels()

The object overrides __getitem__ and __contains__ functions in order to return the representation of a word and to check if a word is in the vocabulary.

model['king']       # equivalent to model.get_word_vector('king')
'king' in model     # equivalent to `'king' in model.get_words()`

Join the fastText community

• Facebook page

• Stack overflow

• Google group

• GitHub