*   Cluster1.R
###
###
###  489/589 ML Ch10 Lab 2 - Clustering Analysis
###
###
####################################################


###
###--------------- Simulated Data --------------------
###                       1. K-means Clustering
###                       2. Hierachical Clustering



###-------------------------------------------------
###--- 1. K-means Clustering


##-- Simulate Dataset
set.seed (2)
X=matrix(rnorm(50*2), 50, 2)
X[1:25, 1]=X[1:25, 1]+3
X[1:25, 2]=X[1:25, 2]-4
plot(X)


##-- K-means clustering with K=2
set.seed(4)
km.out=kmeans(X, centers=2, nstart=20)
km.out

plot(X, col=(km.out$cluster+1), xlab ="", ylab="", pch =20, cex =2,
     main="K-Means Clustering Results with K=2")


##-- K-means clustering with K=3
set.seed(4)
km.out=kmeans(X, centers=3, nstart=20)
km.out

plot(X, col=(km.out$cluster+1), xlab ="", ylab="", pch =20, cex =2,
     main="K-Means Clustering Results with K=3")





##-- Show it step by step
set.seed(234523)
km.1 <- kmeans(X, centers=5, iter.max=1)
km.2 <- kmeans(X, centers=km.1$centers, iter.max=1)
km.3 <- kmeans(X, centers=km.2$centers, iter.max=1)
km.4 <- kmeans(X, centers=km.2$centers, iter.max=1)


M <- cbind(km.1$cluster, km.2$cluster, km.3$cluster)
changing <- which(apply(M, 1, sd)>0)
changing

par(mfrow=c(1,3))

plot(X,col=km.1$cluster, pch=19, main="It1")
points(X[changing, c(1,2)],pch=21,cex=2)

plot(X,col=km.2$cluster, pch=19, main="It2")
points(X[changing, c(1,2)],pch=21,cex=2)

plot(X,col=km.3$cluster, pch=19, main="It3")
points(X[changing, c(1,2)],pch=21,cex=2)













##-- Effect of nstart  (Use nstart>20)
km.out =kmeans(X, 3, nstart=1)
km.out$tot.withinss

km.out =kmeans(X, 3, nstart=20)
km.out$tot.withinss







###-------------------------------------------------
###--- 2. Hierarchical Clustering


##-- Simulate Dataset
set.seed (2)
X=matrix(rnorm(50*2), 50, 2)
X[1:25, 1]=X[1:25, 1]+3
X[1:25, 2]=X[1:25, 2]-4
plot(X)


##-- Hierarchical Clustering
hc.co = hclust(dist(X), method="complete")
hc.av = hclust(dist(X), method="average")
hc.si = hclust(dist(X), method="single")


par(mfrow =c(1,3))
plot(hc.co, main="Complete Linkage", xlab="", sub ="", cex =.9)
plot(hc.av, main="Average Linkage",  xlab="", sub ="", cex =.9)
plot(hc.si, main="Single Linkage",   xlab="", sub ="", cex =.9)


cutree(hc.co, 2)
cutree(hc.av, 2)
cutree(hc.si, 2)

cutree (hc.si, 4)



X.sc=scale(X)
plot(hclust(dist(X.sc), method="complete"),
     main="Hierarchical Clustering with Scaled Features")




##-- Correlation based distance
X=matrix(rnorm(30*3), ncol =3)
dd=as.dist(1-cor(t(x)))
plot(hclust(dd, method =" complete"),
     main="Complete Linkage with Correlation-Based Distance",
     xlab="", sub ="")