# Read in and prepare the data - Chapter 22 Silhouette Analysis of IRIS dataset
i.data <- iris # iris is a built-in dataset
# Min-max normalization
i.data$SL <- (i.data$Sepal.Length - min(i.data$Sepal.Length))/
  (max(i.data$Sepal.Length) - min(i.data$Sepal.Length))
head(i.data)
tail(i.data)
i.data$SL
sort(i.data$SL)
i.data$SW <- (i.data$Sepal.Width - min(i.data$Sepal.Width))/
  (max(i.data$Sepal.Width) - min(i.data$Sepal.Width))
i.data$PL <- (i.data$Petal.Length - min(i.data$Petal.Length))/
  (max(i.data$Petal.Length) - min(i.data$Petal.Length))
i.data$PW <- (i.data$Petal.Width - min(i.data$Petal.Width))/
  (max(i.data$Petal.Width) - min(i.data$Petal.Width))
sort(i.data$SW)
sort(i.data$PL)
sort(i.data$PW)

# Silhouette values
install.packages("cluster")
library("cluster")
km1 <- kmeans(i.data[,6:9], 3)
km
km$cluster
km$centers
km$size
dist1 <- dist(i.data[,6:9], method="euclidean")
sil1 <- silhouette(km1$cluster, dist1)
plot(sil1,col = c("black", "red", "green"), main="Silhouette Plot: 3-Cluster
     K-Means Clustering of Iris Data")
# k-means (k=2)
km2 <- kmeans(i.data[,6:9],2)
dist2 <- dist(i.data[,6:9], method="euclidean")
sil2 <- silhouette(km2$cluster, dist2)
plot(sil2,col = c("black", "red"), main="Silhouette Plot: 2-Cluster
     K-Means Clustering of Iris Data")

# Plot silhouette values
silval1 <- ifelse(sil1[,3] <= 0.33,0,1)
silval1
plot(i.data$PL, i.data$PW, col = silval1 + 1, pch=16, main="Silhouette Valus, K=3",
     xlab = "Petal Length (min-max)", ylab = "Petal Width (min-max)")
legend("topleft",col=c(1,2),pch=16,legend=c("<=0.33",">0.33"))  

silval2 <- ifelse(sil2[,3] <= 0.33,0,1)
plot(i.data$PL, i.data$PW, col = silval2 + 1, pch=16, main="Silhouette Valus, K=2",
     xlab = "Petal Length (min-max)", ylab = "Petal Width (min-max)")
legend("topleft",col=c(1,2),pch=16,legend=c("<=0.33",">0.33"))

# Pseudo-F
# Requires package 'clusterSim'
n <- dim(i.data)[1]
psF1 <- index.G1(i.data[,6:9],cl=km1$cluster)
pf(psF1,2,n-2)
psF2 <- index.G1(i.data[,6:9],cl=km2$cluster)
pf(psF1,2,n-1)