cluster_est_Mbinder_norm: Point estimate of the partition using a modified Binder loss function

Description

Get a point estimate of the partition using a modified Binder loss function for Gaussian components

cluster_est_Mbinder_norm(c, Mu, Sigma, lambda = 0, a = 1, b = a, logposterior)

a list of vector of length n. c[[j]][i] is the cluster allocation of observation i=1...n at iteration j=1...N.

is a list of length n composed of p x l matrices. Where l is the maximum number of components per partition.

Sigma

is list of length n composed of arrays containing a maximum of l p x p covariance matrices.

lambda

is a nonnegative tunning parameter allowing further control over the distance function. Default is 0.

nonnegative constant seen as the unit cost for pairwise misclassification. Default is 1.

nonnegative constant seen as the unit cost for the other kind of pairwise misclassification. Default is 1.

logposterior

vector of logposterior correponding to each partition from c used to break ties when minimizing the cost function

a list:

c_est: a vector of length n. Point estimate of the partition
cost: a vector of length N. cost[j] is the cost associated to partition c[[j]]
similarity: matrix of size n x n. Similarity matrix (see similarityMat)
opt_ind: the index of the optimal partition among the MCMC iterations.

Note that he current implementation only allows Gaussian components.

The modified Binder loss function takes into account the distance between mixture components using #'the Bhattacharyya distance.

JW Lau, PJ Green, Bayesian Model-Based Clustering Procedures, Journal of Computational and Graphical Statistics, 16(3):526-558, 2007.

DA Binder, Bayesian cluster analysis, Biometrika 65(1):31-38, 1978.