Provides an implementation of Conditional Random Fields (CRF) for use in sequential classification tasks. This package can be used independent of the other packages in this distribution.

This implementation of CRF is as described in the following two papers.

John Lafferty and Andrew McCallum and Fernando Pereira, Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data, "Proceedings of the International Conference on Machine Learning (ICML-2001)",2001
F. Sha and F. Pereira, Shallow parsing with conditional random fields, "In Proceedings of HLT-NAACL", "2003"

The code relies on a sparse matrix operations available from the COLT distribution and an implementation of the Quasi-Newton optimization algorithm (LBFGS) available under the package name riso.numerical

The basic package is intentionally kept barebones without any code for data input/output and feature design. Before you can start using the package you need to provide implementations of the FeatureGenerator and DataIter classes. The best way to learn how to use this code is to examples in the package @see iitb.usingCRFs.Segment for Sequence annotations and @see iitb.usingCRFs.MaxentClassifier for a basic maximum entropy based classifier. @author Sunita Sarawagi, IIT Bombay (sunita@iitb.ac.in)