LIBFFM: A Library for Field-aware Factorization Machines

Machine Learning Group at National Taiwan University
Contributors

Introduction

LIBFFM is an open source tool for field-aware factorization machines (FFM). For the formulation of FFM, please see this paper. It has been used to win the top-3 in recent click-through rate prediction competitions (Criteo, Avazu, Outbrain, and RecSys 2015).

It supports

l2-regularized logistic loss

Main features include

using SSE instructions to accelerate vector operations
on-disk learning that can handle data larger than the memory capacity

Download

Warning: FFM is prone to overfitting. See README in the package before using.

Please download LIBFFM at github.

Branches and Interfaces

Description	Version	Developer	Link
A branch for regression problems	1.12	LIBFFM authors	zip or tar.gz
A Python wrapper to run LIBFFM on SFrames	N/A	Chris DuBois	link
A branch that includes linear and bias terms	1.14	LIBFFM authors	zip or tar.gz
A branch that implements standard FM	1.22	LIBFFM authors	zip or tar.gz

Please contact Yuchin Juan for questions, comments, feature requests, or bug report. For Windows related questions please send your email to Wei-sheng Chin.