testing if kmeans clustering is deterministic or not

kmeans.md

kmeans testing

Andrew Barr
September 19, 2015

library(ggplot2)

groups <- factor(rep(c("A", "B"), 100))
x <- rnorm(200) 
y <- rnorm(200) 
clusters <- kmeans(as.matrix(x, y), 2)
clusters2 <- kmeans(as.matrix(x, y), 2)
clusters3 <- kmeans(as.matrix(x, y), 2)
clusters4 <- kmeans(as.matrix(x, y), 2)

table(clusters$cluster)

## 
##   1   2 
## 115  85

table(clusters2$cluster)

## 
##   1   2 
## 115  85

table(clusters3$cluster)

## 
##   1   2 
##  85 115

table(clusters4$cluster)

## 
##   1   2 
## 115  85

forPlot <- data.frame(x, y)
qplot(x, y, color=as.factor(clusters$cluster), size=I(4), data=forPlot, main="First Clustering") + guides(color=FALSE)

qplot(x, y, color=as.factor(clusters2$cluster),size=I(4), data=forPlot, main="Second Clustering") + guides(color=FALSE)

qplot(x, y, color=as.factor(clusters3$cluster),size=I(4), data=forPlot, main="Third Clustering") + guides(color=FALSE)

qplot(x, y, color=as.factor(clusters4$cluster),size=I(4), data=forPlot, main="Fourth Clustering") + guides(color=FALSE)