ibayer

fastFM: Factorization Machines for Python

fastFM brings factorization machines to Python through a scikit-learn-style API, with C and Cython implementations of solvers for regression, classification, and ranking. It suits machine-learning practitioners who need factorization models, but the repository has been inactive for years.

Stars
1.1k
Forks
203
Stars, last 30 days
-1
Last commit
4.2 years ago
Contributors
15

Our take

Inactiveno commits in over a year

Health score

30/100

How it is scored
Activity
19
Community
61
Issues
21
Pull requests
20

-1 stars in the last 30 days

fastFM offers a focused Python interface to factorization machines, but its long inactivity makes it a cautious choice for new production dependencies.

Good fit if

  • You specifically need factorization machines for regression, classification, or ranking.
  • You can work with its Python and native-extension build requirements and accept a project with no recent maintenance activity.
  • You are conducting research and can cite the project’s 2016 JMLR paper.

Look elsewhere if

  • Your team needs recent releases, active issue handling, or ongoing maintenance guarantees.
  • You need a project with clearly asserted license metadata: the README says BSD, while the supplied repository metadata reports NOASSERTION.
  • Your problem does not call for factorization machines or the solver and loss combinations this library provides.
All health signals
Last commit2022-07-17 (4.2 years ago)
Commits, last 90 days0
Releases, last 12 months0 (latest 0.2.11, 2017-11-22)
Contributors15 (top contributor: 36% of commits)
Issues closed, last 90 days0
Pull requests merged, last 90 days0
Project age11 years

Checked on 2026-10-15 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

1,084

Forks

203

Change since Nov 17, 2026: 0 stars·0 forks

Overview

fastFM is a Python library for using factorization machines, a model family for learning from feature interactions. It follows the familiar scikit-learn API and uses C for performance-critical code, wrapped with Cython. The project targets Python users building regression, classification, or ranking systems.

Its solver options include stochastic gradient descent (SGD), coordinate descent (CD), and Markov Chain Monte Carlo (MCMC) for Bayesian inference. The repository is not archived, but it has had no commits in over a year and no recent releases, so teams should weigh its inactivity against its fit for their modeling needs.

Key Features

  • Python interface with a scikit-learn-style fit and predict workflow.
  • C and Cython implementation for performance-critical operations.
  • SGD, coordinate descent, and MCMC optimization methods.
  • Regression with square loss.
  • Classification with probit and sigmoid losses.
  • Ranking with Bayesian Personalized Ranking (BPR) using SGD.
  • Supports Python 2.7 and Python 3.x according to the README.

Use Cases

  • A machine-learning practitioner can try factorization machines for regression on sparse or interaction-rich feature data.
  • A recommender-system developer can evaluate factorization-based ranking with the provided BPR solver.
  • A researcher can compare SGD, coordinate descent, and Bayesian inference approaches in one Python-facing library.
  • A team using scikit-learn-style workflows can integrate factorization-machine models into an existing Python modeling process.

What you need

Detected in the repository

  • Python (from setup.py)

Getting Started

For a binary install on 64-bit systems:

pip install fastFM

For source builds, the README lists Cython, NumPy, pandas, SciPy, and scikit-learn dependencies, plus CBLAS and Python development headers on Linux. See the README and online documentation for build instructions and usage details.

License in Plain Words

No standard license was detected. Without a license, the default copyright rules apply: you may read the code, but not reuse it. Check the repository or ask the authors.

Alternatives

  • spotlight: Spotlight focuses on recommendation models, including matrix factorization and sequence-based ranking, rather than fastFM’s factorization-machine solvers for regression, classification, and ranking.
ProjectLanguageLicenseStarsStatus
fastFMPython–1.1kInactive
spotlightPythonMIT3kInactive

Considerations

The latest recorded commit is from 2022, with no commits in over a year and no releases in the past year. The project is not archived, and its contributor base is not concentrated in one person, but the available health data shows no recent merged pull requests or closed issues. Factorization-machine operations rely on native C code and Cython, so source installation may require platform-specific build dependencies; the README says Linux and OS X were the actively tested systems and describes binary installation as 64-bit only. The README identifies the license as BSD, but the repository metadata does not assert a license, so confirm the applicable terms before adoption. The README also directs users to a separate repository, fastFM-core, for the C implementation and its tests.

Found this useful?

Share it with someone who would like fastFM.

Comparisons