Conditional Random Field (CRF) Toolbox for Matlab

This package is a set of Matlab functions for chain-structured conditional random fields (CRFs) with categorical features. The code implements decoding (with the Viterbi algorithm), inference (with the forwards-backwards algorithm), sampling (with the forwards-filter bacwards-sample algorithm), and parameter estimation (with a limited-memory quasi-Newton algorithm) in these models. Several of the functions have been implemented in C as mex files to speed up calculations.