buster: Buster

In SimonRaper/buster: Bagging on hierarchical clustering

Description

Performs bagging on hierarchical clusters

Usage

buster(dist, n = 100, k, size = 0.66, method = "ward", pct.exc = 0.1)

Arguments

dist	A distance object
n	The number of times the data should be resampled
k	The number of clusters
size	The percentage of the original data in each resample
method	The linkage method to be passed to hclust
pct.exc	Setting this to x excludes or highlights the top x example if we set this to 0.1 then the top decile of unstable observations are excluded or highlighted.
low.mem	Setting this to true uses a slower but less memory intensive way of calculating the co-occurrences

Value

An an object of class buster which includes

• bhclust An hclust object on the co-ocurrences

• bhclust.ex An hclust object on the co-ocurrences exclduing unstable observations

• hclust An hclust object on the original distance measure

• obs.eval An evalution of the stability of the observations

Author(s)

Simon Raper

Examples

#Testing on the iris data set
iris.dist<-dist(iris[,1:4])
bhc<-buster(iris.dist, n=250, k=3, size=0.66, method='ward', pct.exc=0.07)
plot(bhc)

#We see the unstable observations in pink.

cluster<-bhc$obs.eval$cluster[order(bhc$obs.eval$obs.ind)]
plot(iris[,1:4], col=6-cluster, pch = rep(15:17, each=50))

#Another simple test

x.1<-rnorm(50, 10, 3)
y.1<-rnorm(50, 10, 3)
x.2<-rnorm(50, 20, 3)
y.2<-rnorm(50, 10, 3)
x.3<-rnorm(50, 13, 3)
y.3<-rnorm(50, 20, 3)

test.data<-data.frame(group=rep(1:3, each=50), x=c(x.1, x.2, x.3), y=c(y.1, y.2, y.3))
names<-c(paste0("group 1: ", 1:50), paste0("group 2: ", 1:50), paste0("group 3: ", 1:50))
rownames(test.data)<-names
dist<-dist(test.data[,-1])

bhc<-buster(dist, n=200, k=3, size=0.66, method='ward', pct.exc=0.1)

plot(bhc)

#Append the assigned clusters
graph.data<-cbind(test.data, bhc$obs.eval[order(bhc$obs.eval$obs.ind, decreasing=FALSE),])
plot(graph.data$x, graph.data$y, xlim=c(0,30), ylim=c(0, 30), pch = graph.data$group, col=graph.data$cluster+1)

max.co<-bhc$obs.eval$max.co[order(bhc$obs.eval$obs.ind)]
alpha<-(max.co-min(max.co))/(max(max.co)-min(max.co))
cols <- hsv(0,0,0,alpha)
plot(graph.data$x, graph.data$y, xlim=c(0,30), ylim=c(0, 30), pch = 19, col=cols)

SimonRaper/buster documentation built on May 9, 2019, 1:32 p.m.