criteria {Rankcluster}

Criteria estimation

Description

This function estimates the loglikelihood of a mixture of multidimensional ISR model, as well as the BIC and ICL model selection criteria.

Usage

criteria(data, proportion, pi, mu, m, Ql = 500, Bl = 100, IC = 1, nb_cpus = 1)

Arguments

data

a matrix in which each row is a rank (partial or not; for partial rank, missing elements of a rank are put to 0).

proportion

a vector (which sums to 1) containing the K mixture proportions.

pi

a matrix of size K*p, where K is the number of clusters and p the number of dimension, containing the probabilities of a good comparison of the model (dispersion parameters).

mu

a matrix of size K*sum(m), containing the modal ranks. Each row contains the modal rank for a cluster. In the case of multivariate ranks, the reference rank for each dimension are set successively on the same row.

m

a vector containing the size of ranks for each dimension.

Ql

number of iterations of the Gibbs sampler used for the estimation of the log-likelihood.

Bl

burn-in period of the Gibbs sampler.

IC

number of run of the computation of the loglikelihood.

nb_cpus

number of cpus for parallel computation

Value

a list containing:

ll

the estimated log-likelihood.

bic

the estimated BIC criterion.

icl

the estimated ICL criterion.

Author(s)

Quentin Grimonprez

Examples

data(big4)
res <- rankclust(big4$data, m = big4$m, K = 2, Ql = 100, Bl = 50, maxTry = 2)
if (res@convergence) {
  crit <- criteria(big4$data, res[2]@proportion, res[2]@pi, res[2]@mu,
                  big4$m, Ql = 200, Bl = 100)
}