In MickyDowns/mine_algorithms: What the package does (short line)

KMeans

Clustering cluster.csv

Use kmeans() w/ default values to find the k=2 solution for the 2-dimensional data cluster.csv

setwd("./data")
data<-read.csv("cluster.csv",header=F)
setwd("../")

Plot initial data.

plot(data)

Cluster: kmeans() produces centers, cluster assignments, etc.

fit<-kmeans(data,2)
fit
fit$centers

Plot: Key here is to use kmeans() output rather than moving data around.

plot(data$V1,data$V2,col=fit$cluster)
points(fit$centers,col=c("black","red"),pch=19)

Clustering sonar

Use kmeans() w/ default values to find the k=2 solution for the 2-dimensional sonar data.

setwd("./data")
train<-read.csv("sonar_train.csv",header=F)
test<-read.csv("sonar_test.csv",header=F)
setwd("../")

Plot just the first two columns of the sonar data.

plot(train[,1:2])

Cluster: kmeans() can use as many attributes as you want. But, let's look at the clusters created by the first two.

fit<-kmeans(train[,1:2],2)
fit

Plot: Key here is to use kmeans() output rather than moving data around.

plot(train[,1:2],col=fit$cluster)
points(fit$centers,col="blue",pch=19)

Compare sonar clusters to actual class labels??

plot(train[,1:2],pch=19,xlab=expression(x[1]),
     ylab=expression(x[2]))
## get your y labels
y<-train[,61]
## re-plot points with color based on class labels.
points(train[,1:2],col=2+2*y,pch=19)

Compute the misclass error

What if we used kmeans() to classify. What would our misclass error be?

## transform cluster labels (1's and 2's) to -1s and 1s
sum(fit$cluster*2-3==y)/length(y)

Try it for all 60 columns

fit<-kmeans(train[,1:60],2)
sum(fit$cluster*2-3==y)/length(y)
sum(fit$cluster*2-3!=y)/length(y)

Try w/ more centroids. Disaster.

fit<-kmeans(train[,1:60],10)
sum(fit$cluster*2-3==y)/length(y)
sum(fit$cluster*2-3!=y)/length(y)

Gist: kmeans() is a good clustering tool. Not a good prediction tool.

What is kmeans doing?

First code it manually.

x<-c(1,2,3,5,6,7,8)
center1<-1
center2<-2

for (k in 2:10) {
     cluster1<-x[abs(x-center1[k-1])<=abs(x-center2[k-1])]
     ## Put in cluster1 all x's where distance to c1<= distance to c2.
     cluster2<-x[abs(x-center1[k-1])>abs(x-center2[k-1])]
     ## Put in c2 all x's where distance to c1>distance to c2

     center1[k]<-mean(cluster1)
     center2[k]<-mean(cluster2)
     ## apparently mean() will take the mean between of all values in a cluster.
     ## set k=2. Decrement it 1 to control iteration. Also use it to track the updates clusters. 
}

center1
center2
cluster1
cluster2

Compare to kmeans()

x<-c(1,2,3,5,6,7,8)

fit<-kmeans(x,2)

plot(x,col=fit$cluster)

Calc distances

x1<-c(2,2)
x2<-c(5,7)

data<-matrix(c(x1,x2),nrow=2,byrow=T)
data

dist(data)

MickyDowns/mine_algorithms documentation built on May 8, 2019, 10:49 a.m.