compute.kclust: Gram similarity matrix checker

In sClust: R Toolbox for Unsupervised Spectral Clustering

Description

Function which select the number of cluster to compute thanks to a selected method

Usage

compute.kclust(
  eigenValues,
  method = "default",
  Kmax = 20,
  tolerence = 1,
  threshold = 0.9,
  verbose = FALSE
)

Arguments

eigenValues	The eigenvalues of the laplacian matrix.
method	The method that will be used. "default" to let the function choose the most suitable method. "PEV" for the Principal EigenValue method. "GAP" for the GAP method.
Kmax	The maximum number of cluster which is allowed.
tolerence	The tolerance allowed for the Principal EigenValue method.
threshold	The threshold to select the dominant eigenvalue for the GAP method.
verbose	To output the verbose in the terminal.

Value

a vector which contain the number of cluster to compute.

Author(s)

Emilie Poisson Caillault and Erwan Vincent

Examples

### Example 1: 2 disks of the same size
n<-100 ; r1<-1
x<-(runif(n)-0.5)*2;
y<-(runif(n)-0.5)*2
keep1<-which((x*2+y*2)<(r1*2))
disk1<-data.frame(x+3*r1,y)[keep1,]
disk2 <-data.frame(x-3*r1,y)[keep1,]
sameTwoDisks <- rbind(disk1,disk2)
W <- compute.similarity.ZP(scale(sameTwoDisks))
W <- checking.gram.similarityMatrix(W)
eigVal <- compute.laplacian.NJW(W,verbose = TRUE)$eigen$values
K <- compute.kclust(eigVal, method="default", Kmax=20, tolerence=0.99, threshold=0.9, verbose=TRUE)

### Example 2: Speed and Stopping Distances of Cars
W <- compute.similarity.ZP(scale(cars))
W <- checking.gram.similarityMatrix(W)
eigVal <- compute.laplacian.NJW(W,verbose = TRUE)$eigen$values
K <- compute.kclust(eigVal, method="default", Kmax=20, tolerence=0.99, threshold=0.9, verbose=TRUE)

sClust documentation built on Aug. 24, 2021, 1:06 a.m.