Variational Bayesian inference for unsupervised clustering

vimix(X, K, prior, indep = F, init = "kmeans", select = F,
tol = 1e-09, maxiter = 2000, verbose = F)

Arguments

X NxD data matrix. (Maximum) number of clusters. Prior parameters (optional). Booleand indicator. If TRUE, the features are considered to be independent. Default is FALSE. Initialisation method (optional). If it is a vector, it is interpreted as the vector of initial cluster allocations. If it is a string, it is interpreted as the name of the clustering algorithm used for the initialisation (only "kmeans" and "random") available at the moment). Boolean flag. If TRUE, variable selection is used. Default is FALSE. Tolerance on lower bound. Default is 10e-20. Maximum number of iterations of the VB algorithm. Default is 2000. Boolean flag which, if TRUE, prints the iteration numbers. Default is FALSE.

Value

A list containing L, the lower bound at each step of the algorithm, label, a vector containing the cluster labels, model, a list containing the trained model structure.

References

Bishop, C.M., 2006. Pattern recognition and machine learning. Springer.

Examples

library(mvtnorm)
data <- rbind(rmvnorm(100,c(-3,0)), rmvnorm(100,c(3,0)))
output <- vimix(data, 2)