MBM.cluster: MBMM clustering

In comato: Analysis of Concept Maps and Concept Landscapes

Description

MBM.cluster calculates a model based clustering using multivariate Bernoulli-mixtures as probabilistic model of the data. The quality of the clustering is judged using the AIC criterion.

Usage

MBM.cluster(data, min = 1, max = 10)

Arguments

data	A numeric matrix.
min	The minimal number of components that is tested.
max	The maximal number of components that is tested.

Value

A list with 3 elements. The first element is the minimal AIC value for each tested number of components. The second element is a vector of all AIC values. The third is the actual clustering as returned by the EM algorithm using the optimal number of components according to AIC. The element is again a list that contains the mixture coefficients, the actual parameters of the mutlivariate Benroulli distributions, the probability matrix of each observation (i.e. row if data) and component and the number of iterations that the EM algorithm needed to converge.

Examples

#Random data generation, 100 observations, 5 dimensions, dependencies within the dimensions
data = cbind(round(runif(100)), round(runif(100)), round(runif(100)))
data = cbind(data, data[,2], 1-data[,3])

#Noisy data:
s = round(runif(2, 1, 100))
data[s, c(4,5)] = 1 - data[s, c(4,5)]

#MBMM Clustering
res = MBM.cluster(data, 1,8)

comato documentation built on May 2, 2019, 6:52 a.m.