CRF (version 0.3-6)

CRF-package: CRF - Conditional Random Fields

Description

Library of Conditional Random Fields model

Details

CRF is R package for various computational tasks of conditional random fields as well as other probabilistic undirected graphical models of discrete data with pairwise and unary potentials. The decoding/inference/sampling tasks are implemented for general discrete undirected graphical models with pairwise potentials. The training task is less general, focusing on conditional random fields with log-linear potentials and a fixed structure. The code is written entirely in R and C++. The initial version is ported from UGM written by Mark Schmidt.

Decoding: Computing the most likely configuration

• decode.exactExact decoding for small graphs with brute-force search
• decode.chainExact decoding for chain-structured graphs with the Viterbi algorithm
• decode.treeExact decoding for tree- and forest-structured graphs with max-product belief propagation
• decode.conditionalConditional decoding (takes another decoding method as input)
• decode.cutsetExact decoding for graphs with a small cutset using cutset conditioning
• decode.junctionExact decoding for low-treewidth graphs using junction trees
• decode.sampleApproximate decoding using sampling (takes a sampling method as input)
• decode.marginalApproximate decoding using inference (takes an inference method as input)
• decode.lbpApproximate decoding using max-product loopy belief propagation
• decode.trbpApproximate decoding using max-product tree-reweighted belief propagtion
• decode.greedyApproximate decoding with greedy algorithm
• decode.icmApproximate decoding with the iterated conditional modes algorithm
• decode.blockApproximate decoding with the block iterated conditional modes algorithm
• decode.ilpExact decoding with an integer linear programming formulation and approximate using LP relaxation

Inference: Computing the partition function and marginal probabilities

• infer.exactExact inference for small graphs with brute-force counting
• infer.chainExact inference for chain-structured graphs with the forward-backward algorithm
• infer.treeExact inference for tree- and forest-structured graphs with sum-product belief propagation
• infer.conditionalConditional inference (takes another inference method as input)
• infer.cutsetExact inference for graphs with a small cutset using cutset conditioning
• infer.junctionExact decoding for low-treewidth graphs using junction trees
• infer.sampleApproximate inference using sampling (takes a sampling method as input)
• infer.lbpApproximate inference using sum-product loopy belief propagation
• infer.trbpApproximate inference using sum-product tree-reweighted belief propagation

Sampling: Generating samples from the distribution

• sample.exactExact sampling for small graphs with brute-force inverse cumulative distribution
• sample.chainExact sampling for chain-structured graphs with the forward-filter backward-sample algorithm
• sample.treeExact sampling for tree- and forest-structured graphs with sum-product belief propagation and backward-sampling
• sample.conditionalConditional sampling (takes another sampling method as input)
• sample.cutsetExact sampling for graphs with a small cutset using cutset conditioning
• sample.junctionExact sampling for low-treewidth graphs using junction trees
• sample.gibbsApproximate sampling using a single-site Gibbs sampler

Training: Given data, computing the most likely estimates of the parameters

• train.crfTrain CRF model
• train.mrfTrain MRF model

Tools: Tools for building and manipulating CRF data

• make.crfGenerate CRF from the adjacent matrix
• make.featuresMake the data structure of CRF features
• make.parMake the data structure of CRF parameters
• duplicate.crfDuplicate an existing CRF
• clamp.crfGenerate clamped CRF by fixing the states of some nodes
• clamp.resetReset clamped CRF by changing the states of clamped nodes
• sub.crfGenerate sub CRF by selecting some nodes
• mrf.updateUpdate node and edge potentials of MRF model
• crf.updateUpdate node and edge potentials of CRF model

References

J. Lafferty, A. McCallum, and F. Pereira. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In the proceedings of International Conference on Machine Learning (ICML), pp. 282-289, 2001.

Mark Schmidt. UGM: Matlab code for undirected graphical models. http://www.di.ens.fr/~mschmidt/Software/UGM.html

Examples

library(CRF)
data(Small)
decode.exact(Small$crf)
infer.exact(Small$crf)
sample.exact(Small$crf, 100)