RPGMMClu: Random Projection Ensemble Clustering Algorithm

In RPEClust: Random Projection Ensemble Clustering Algorithm

Description

This function allows to run the RPEClu algorithm.

Usage

RPGMMClu(x, true.cl = NULL, g, d = NULL, c = 10, B = 1000,
  B.star = 100, modelNames = NULL, diagonal = FALSE,
  ensmethod = "DWH", seed = 101, verb = FALSE)

Arguments

x	A numeric high-dimensional matrix where rows correspond to observations and columns correspond to variables.
true.cl	A vector containing the true cluster membership. If supplied, the Adjusted Rand Index (ARI) of the predicted clustering is also returned. By default is set to NULL.
g	The number of clusters.
d	The dimension of the projected space. If is NULL (default option), then d = ceil(c* log(g)).
c	The constant which governs the dimension of the projected space if d is not provided. The default is set to 10.
B	The number of generated projections; the default is 1000.
B.star	The number of base models to retain in the final ensemble; the default is 100.
modelNames	A vector of character strings indicating the models to be fitted in the EM phase of the GMM. The help file for Mclust describes the available options.
diagonal	A logical value indicating whether the conditional covariate matrix has a restricted form, i.e. it is diagonal. The default is FALSE.
ensmethod	A character string specifying the method for computing the clustering consensus. See the cl_consensus help file for available options. The default is DWH.
seed	A single value indicating the initializing seed for random generations.
verb	A logical controlling if the progress number of projections is displayed during the fitting procedure. By default is FALSE.

Value

The output components are as follows:

ensemble	A list including: label.vec The vector of labels predicted by the ensemble of size B.star. ari The corresponding ARI (if true.cl is not NULL).
individual	A list including: label.vec The vectors of labels predicted by each base model. ari The corresponding ARI (if true.cl is not NULL). bic The BIC associated to each base model. bic.GMM The BIC associated to the Gaussian mixture fitted on each projected data. bic.reg The BIC for the linear regression of the (p-d) last columns of Y* on the first d ones.

References

Anderlucci, Fortunato, Montanari (2019) <arXiv:1909.10832>

Examples

data(Meat)
out.clu <- RPGMMClu(Meat$x, Meat$y, g=5, B=1000, B.star=100, verb=TRUE)

data <- sim_normal(n = rep(100, 2), p = 100, rho = rep(0.1, 2), delta = 0.5, sigma2 = 1, seed = 106)
out.clu <- RPGMMClu(data$x, data$y, g=2, B=10, B.star=5, verb=TRUE)

RPEClust documentation built on Nov. 6, 2019, 5:08 p.m.