ARGMAX: An Open Source Software for Structured Prediction

Introduction
ARGMAX is an open source implementation for structured prediction (SP). SP is a fundamental machine learning problem of generating outputs with complex internal structure in which the output variables are mutually dependent or constrained. ARGMAX is designed for a simple, customizable, and open source tool for natural language processing, machine learning and data mining.

Features
  • Linear-chain CRFs using LBFGS optimization
  • Can perform sparse forward-backward, tied potiential inference algorithms
  • Can use same feature templates of CRF++ and MALLET
News
2008-05-28: first beta version released

Download
Source code: ARGMAX (beta release)

Install
$ tar cvzf argmax-beta.tar.gz
$ cd argmax-beta
$ make

Usage
$ argmax [configuration file]

References
  • J. Lafferty, A. McCallum, and F. Pereira. Conditional random fields: Probabilistic models for segmenting and labeling sequence data, In Proc. of ICML, pp.282-289, 2001
  • C. Sutton and A. McCallum, An Introduction to Conditional Random Fields for Relational Learning, Book chapter in Introduction to Statistical Relational Learning. Edited by Lise Getoor and Ben Taskar. MIT Press., 2006
  • CRF++: Yet Another CRF toolkit
  • MALLET: A Machine Learning for Language Toolkit

Supported by
SourceForge.net Logo