FastText Deployment and Usage Guide

Prerequisites

• Compiler: C++11 compatible compiler (g++-4.7.2 or newer, or clang-3.3 or newer)
• Build Tools: GNU Make (for make-based build) or CMake 2.8.9 or newer (for cmake-based build)
• Operating System: Modern Mac OS or Linux distributions (tested on Debian jessie and newer)
• Python Dependencies (for Python bindings):
  • Python 2.7 or >=3.4
  • NumPy & SciPy
  • pybind11

Installation

Option 1: Using Make (Recommended)

# Download the latest stable release
wget https://github.com/facebookresearch/fastText/archive/v0.9.2.zip

# Extract the archive
unzip v0.9.2.zip

# Navigate to the extracted directory
cd fastText-0.9.2

# Build the project
make

Option 2: Using CMake

# Clone the repository (master branch for latest features)
git clone https://github.com/facebookresearch/fastText.git

# Navigate to the project directory
cd fastText

# Create a build directory and navigate to it
mkdir build && cd build

# Configure with CMake
cmake ..

# Build and install
make && make install

Option 3: Python Bindings

# Clone the repository (master branch)
git clone https://github.com/facebookresearch/fastText.git

# Navigate to the project directory
cd fastText

# Install Python package
pip install .

Configuration

• Compiler Selection: If you want to use a compiler other than the default system-wide compiler, update the CC and INCLUDES macros at the beginning of the Makefile.
• Word Vector Parameters: FastText uses character n-grams from 3 to 6 characters by default. You can modify these parameters when running the training commands.

Build & Run

Local Development

After building with make, you can run fastText commands directly:

# Word representation learning (skipgram model)
./fasttext skipgram -input data.txt -output model

# Word representation learning (cbow model)
./fasttext cbow -input data.txt -output model

# Text classification
./fasttext supervised -input train.txt -output model

Production Usage

For production deployments, consider using the CMake build to generate shared libraries:

# After cmake build
./fasttext supervised -input train.txt -output model -epoch 25 -lr 1.0

Deployment

FastText is a C++ library that can be deployed in various environments:

Web Applications

• Docker: Containerize your application with the fastText binary and required dependencies
• Cloud Functions: Deploy as serverless functions (AWS Lambda, Google Cloud Functions) with the binary included

API Services

• Flask/FastAPI: Create REST APIs wrapping fastText functionality
• Node.js: Use child processes to call the fastText binary from Node applications

Machine Learning Pipelines

• Apache Spark: Integrate fastText as a transformer in ML pipelines
• Kubeflow: Deploy as part of MLOps workflows

Troubleshooting

Common Issues and Solutions

1. Compilation Errors

Issue: "error: 'to_string' is not a member of 'std'" Solution: Your compiler doesn't support C++11. Upgrade to g++-4.7.2 or newer, or clang-3.3 or newer.

2. Missing Dependencies

Issue: "make: *** No rule to make target 'fasttext', needed by 'all'. Stop." Solution: Ensure you have GNU Make installed and run make from the correct directory.

3. Python Bindings Not Working

Issue: "ImportError: No module named 'fasttext'" Solution: Verify that you ran pip install . from the correct directory and that your Python environment has the required dependencies (NumPy, SciPy, pybind11).

4. Memory Issues During Training

Issue: "std::bad_alloc" or system running out of memory Solution: Reduce the training data size, decrease the number of dimensions, or use a machine with more RAM.

5. Text Encoding Problems

Issue: Characters not displaying correctly or training failing Solution: Ensure your input text files are UTF-8 encoded, as fastText expects UTF-8 input.

6. CMake Configuration Issues

Issue: CMake configuration fails Solution: Ensure you have CMake 2.8.9 or newer installed, and that you're running cmake from within a build directory.

For additional help, visit the fastText FAQ or join the fastText community.