README

PHP port of the LightFM Python library using FFI bindings to C implementation for hybrid recommender systems.

Installation

Install via Composer:

composer require lightfm/lightfm-php

Requirements

PHP 8.0 or higher
FFI extension enabled
The bundled C library (liblightfm.so)

Usage

<?php

require_once 'vendor/autoload.php';

// Create a LightFM model
$model = new LightFM([
    'no_components' => 30,
    'loss' => 'warp',
    'learning_rate' => 0.05
]);

// Fit the model with interaction data
$interactions = new CSRMatrix($data, $rows, $cols);
$model->fit($interactions);

// Get recommendations
$predictions = $model->predict($userIds, $itemIds);

Running the Examples

The repository includes a comprehensive demo script that showcases all major features of the library.

Prerequisites

Install dependencies:

composer install

Ensure the C library is compiled and available:

# The liblightfm.so file should be in the project root
ls -la liblightfm.so

Run the Demo

Execute the demo script from the project root:

php test/main.php

The demo will demonstrate:

Loading synthetic MovieLens-like data
Training models with WARP loss function
Making predictions and generating recommendations
Evaluating model performance with precision metrics
Using user and item features
Working with custom interaction data

Expected Output

The demo will display:

Dataset statistics (users, items, interactions)
Training progress and timing
Model evaluation metrics (precision@5)
Top recommendations for sample users
Performance comparison with and without features
Custom dataset predictions

Note: The current C implementation is simplified, so prediction scores may appear as zeros. This is expected behavior for the initial implementation.

Features

Hybrid Recommender Systems: Supports both collaborative filtering and content-based filtering
Multiple Loss Functions: WARP, BPR, and logistic loss functions
Fast C Implementation: Uses FFI bindings to a compiled C library for performance
CSR Matrix Support: Efficient sparse matrix operations
Evaluation Metrics: Built-in precision, recall, and AUC metrics

Classes

LightFM: Main model class for training and prediction
CSRMatrix: Sparse matrix implementation for efficient data handling
Evaluation: Evaluation metrics for model performance
MovieLensDataset: Dataset loader for MovieLens data
LightFMFFI: FFI bindings to the C library

License

Apache-2.0

lightfm / lightfm-php

Maintainers

Package info

Statistics

Security