SSKmeans: SSKmeans

In gatorBiostat/SSKmeans: Smoothed Sparse Kmeans

Description

Smoothed sparse Kmeans

Usage

SSKmeans(x, K = NULL, lambda1 = 0, lambda2 = 0, E = NULL, nstart = 20,
  silent = FALSE, maxiter = 20)

Arguments

x	a data matrix of dimension n * p, where n is number of samples to be clustered, p is number of features.
K	pre-specified number of clusters
lambda1	tuning parameter for l1 norm lasso penalty. Large lambda1 will induce more feature weights to be 0.
E	network structure. E should be a m by 2 matrix, where m is total number of connections (edges) in the graph. For example, if feature 2 and feature 3 are connected in the graph, E[j,] <- c(2, 3). j = 1, ..., m.
nstart	number of initialization for Kmeans.
silent	if print out progress.
maxiter	max number of iterations.
lambda1	tuning parameter for the smoothness of feature selection. Large lambda will induce feature weights to be similar.

Details

Perform sparse Kmeans to perform sample clustering and feature selection. In feature selection, we also want to incorporate spatio information such that adjacent voxels have similar coefficient.

Value

a list. The following items are included in the list.

ws	weight for each feature. Zero weight means the feature is not selected.
Cs	Cluster Assignment
wcss	within cluster sum of square
crit	objective value
E	network structure

Author(s)

Caleb

Examples

set.seed(11)
x <- matrix(rnorm(50*70),ncol=70)
x[1:25,1:20] <- x[1:25,1:20]+1
x <- scale(x, TRUE, TRUE)
# choose tuning parameter
lambda1 <- 1
lambda2 <- 1
E <- cbind(1:69,2:70)
K=2
nstart=20
silent=FALSE
maxiter=6

km.out <- SSKmeans(x,K=2,lambda1=1, lambda2=1, E=E)

gatorBiostat/SSKmeans documentation built on May 26, 2019, 11:59 p.m.