evclust (version 2.0.3)

expandlink: Expansion of must-link and cannot-link constraints

Description

expandlink returns an expanded set of must-link and cannot-link constraints using the k nearest neighbors of each observation.

Usage

expandlink(link, ind, distan)

Value

A list with two attributes:

ML

The new matrix of must-link constraints.

CL

The new matrix of cannot-link constraints.

Arguments

link

A list with two attributes: a matrix ML containing nbML x 2 must-link constraints and a matrix CL containing nbCL x 2 cannot-link constraints.

ind

An n*k matrix containing the k nearest neighbor indices.

distan

An n*k matrix containing the k nearest neighbor distances.

Author

Feng Li and Thierry Denoeux.

Details

Using the algorithm described in Li et al (2018), expandlink generates new must-link and cannot-link constraints from existing ones, using the k nearest neighbors of each observations. The extended constraint list can be used by constrained clusetring algorithms such as cecm and kcevclus.

References

F. Li, S. Li and T. Denoeux. k-CEVCLUS: Constrained evidential clustering of large dissimilarity data. Knowledge-Based Systems (142):29-44, 2018.

See Also

kcevclus,cecm,create_MLCL, bananas

Examples

if (FALSE) {
data<-bananas(200)
link<-create_MLCL(data$y,10)
nml<-nrow(link$ML)
plot(data$x,col=data$y)
for(k in 1:nml) lines(data$x[link$ML[k,],1],data$x[link$ML[k,],2],lwd=2,col="red")
ncl<-nrow(link$CL)
for(k in 1:ncl) lines(data$x[link$CL[k,],1],data$x[link$CL[k,],2],lwd=2,col="blue")
library(FNN)
nn<-get.knn(data$x,5)
link1<-expandlink(link,ind=nn$nn.index,distan=nn$nn.dist)
nml<-nrow(link1$ML)
for(k in 1:nml) lines(data$x[link1$ML[k,],1],data$x[link1$ML[k,],2],lwd=1,lty=2,col="red")
ncl<-nrow(link1$CL)
for(k in 1:ncl) lines(data$x[link1$CL[k,],1],data$x[link1$CL[k,],2],lwd=1,lty=2,col="blue")
}