fastFM System

A fastFM System is a factorization machine-based system that implements an FM algorithm to solve an FM-based prediction task.

• Context:
  • It can implement Factorization Solver Algorithms, such as: ALS alg., MCMC, and SGD.
  • It can support Factorization Loss Functions, such as: Square Loss (for regression), Probit and Sigmoid (for classification) and BPR alg. (for ranking).
• Example(s):
  • fastFM v0.2.5 (2016-05-25), which supports OpenBLAS).
  • …
• Counter-Example(s):
  • libFM.
• See: Factorization Algorithm.

---

---

References

2017

• https://github.com/ibayer/fastFM
  • QUOTE: This repository allows you to use Factorization Machines in Python (2.7 & 3.x) with the well known scikit-learn API. All performence critical code as been written in C and wrapped with Cython. fastFM provides stochastic gradient descent (SGD) and coordinate descent (CD) optimization routines as well as Markov Chain Monte Carlo (MCMC) for Bayesian inference. The solvers can be used for regression, classification and ranking problems. Detailed usage instructions can be found in the online documentation and on arXiv. ...

  Task 	Solver 	Loss
Regression 	als, mcmc, sgd 	Square Loss
Classification 	als, mcmc, sgd 	Probit(Map), Probit, Sigmoid
Ranking 	sgd 	BPR

2016

• (Bayer, 2016) ⇒ Immanuel Bayer. (2016). “fastFM: A Library for Factorization Machines.” In: The Journal of Machine Learning Research, 17(1).
  • QUOTE: Factorization Machines (FM) are currently only used in a narrow range of applications and are not yet part of the standard machine learning toolbox, despite their great success in collaborative filtering and click-through rate prediction. However, Factorization Machines are a general model to deal with sparse and high dimensional features. Our Factorization Machine implementation (fastFM) provides easy access to many solvers and supports regression, classification and ranking tasks. Such an implementation simplifies the use of FM for a wide range of applications. Therefore, our implementation has the potential to improve understanding of the FM model and drive new development.

---