infer.lbp {CRF}

Inference method using loopy belief propagation

Description

Computing the partition function and marginal probabilities

Usage

infer.lbp(crf, max.iter = 10000, cutoff = 1e-04, verbose = 0, maximize = FALSE)

Arguments

crf

The CRF

max.iter

The maximum allowed iterations of termination criteria

cutoff

The convergence cutoff of termination criteria

verbose

Non-negative integer to control the tracing informtion in algorithm

maximize

Logical variable to indicate using max-product instead of sum-product

Details

Approximate inference using sum-product loopy belief propagation

Value

This function will return a list with components:

node.bel

Node belief. It is a matrix with crf$n.nodes rows and crf$max.state columns.

edge.bel

Edge belief. It is a list of matrices. The size of list is crf$n.edges and the matrix i has crf$n.states[crf$edges[i,1]] rows and crf$n.states[crf$edges[i,2]] columns.

logZ

The logarithmic value of CRF normalization factor Z.

Examples

library(CRF)
data(Small)
i <- infer.lbp(Small$crf)