fastText Python Bindings

0.9.3 · active · verified Sat Apr 11

fastText is a library for efficient learning of word representations and sentence classification. Developed by Facebook AI Research, it's particularly good for large-scale text processing tasks. The current version is 0.9.3, with releases focusing on new features, performance, and API stability rather than a fixed cadence.

Warnings

breaking Version 0.9.1 merged the previously separate official 'fastText' (from GitHub) and unofficial 'fasttext' (from PyPI) Python modules. This involved significant API changes, especially for users who were previously installing directly from GitHub.
Fix: Ensure you are importing `fasttext` (lowercase) and refer to the v0.9.1+ documentation for updated API calls, particularly `fasttext.load_model()` and training parameters. Uninstall any old `fastText` installations before reinstalling `fasttext`.
gotcha Installation on some operating systems (e.g., Windows, or macOS without specific tools) can fail due to C++ compilation requirements. fastText is a C++ library with Python bindings.
Fix: Ensure you have a C++ compiler (e.g., GCC, Clang for Linux/macOS, MSVC for Windows) installed and correctly configured in your PATH. On Windows, this often means installing 'Build Tools for Visual Studio'.
gotcha The `fasttext.load_model()` function expects a `.bin` model file, which contains both word vectors and classification information. It cannot directly load standalone `.vec` (vector) files.
Fix: Always save your trained fastText models using `model.save_model('model.bin')` and load them with `fasttext.load_model('model.bin')`. If you only have `.vec` files, you'll need to retrain or find the original `.bin` model.
gotcha Training data for supervised classification (e.g., `train_supervised`) must adhere to a specific format: each line should contain the label prefixed with `__label__`, followed by the text, for example: `__label__positive This is a great product.`
Fix: Pre-process your training data to match the fastText input format. Each line is a document, and each document starts with its label(s) prefixed by `__label__`.

Install

pip install fasttext Install fasttext

Imports

fasttext
```
import fasttext
```
Prior to v0.9.1, the official GitHub module was 'fastText', while PyPI had an unofficial 'fasttext'. They merged, and the official import is now 'fasttext'.

Quickstart

This quickstart demonstrates how to train a basic supervised text classification model and make predictions using fastText. The training data must be in the specific fastText format with `__label__` prefixes.

import fasttext
import os

# Create a dummy training file for demonstration
training_data_path = 'train.txt'
with open(training_data_path, 'w') as f:
    f.write('__label__positive This is a good movie.\n')
    f.write('__label__negative This movie was terrible.\n')
    f.write('__label__positive I love this film.\n')

# Train a supervised model
model = fasttext.train_supervised(input=training_data_path)

# Predict a label
text_to_predict = 'This is an excellent film.'
predictions = model.predict(text_to_predict)
print(f"Text: '{text_to_predict}'")
print(f"Prediction: {predictions[0][0]}, Probability: {predictions[1][0]:.4f}")

# Clean up dummy file
os.remove(training_data_path)

view raw JSON →