#Iris data from randomForest help file is used!

## Classification:
##data(iris)
set.seed(71)
iris.rf <- randomForest(Species ~ ., data=iris, importance=TRUE,
                        proximity=TRUE)

#Default MDSplot
MDSplot(iris.rf, iris$Species)

#Figure Supervised RF
MDSplot(iris.rf, iris$Species,
bg=rainbow(length(levels(iris$Species)))[iris$Species],
pch=c(21:23), palette=rep("black",length(levels(iris$Species))))
legend("topleft",levels(iris$Species), pch=c(21:23), col = rep("black",length(levels(iris$Species))), 
pt.bg = rainbow(length(levels(iris$Species))), cex=.8)
title(main="Supervised Random Forest MDS Plot of Iris Data")

#Now Unsupervised RF!

## The `unsupervised' case:
set.seed(17)
iris.urf <- randomForest(iris[, -5])

#Let's pretend we KNOW there are 3 groups!

#Figure Unsupervised RF with 
pamData <- pam(1 - iris.urf$proximity, k = 3, diss = TRUE)
MDSplot(iris.urf, pamData$clustering, bg=rainbow(length(levels(as.factor(pamData$clustering))))[unclass=pamData$clustering], 
pch=c(21:23)[unclass=pamData$clustering], palette=rep("black",3))

legend("bottomleft",c("Group 1", "Group 2", "Group 3"), 
pch=c(21:23), col = rep("black", 3), pt.bg = rainbow(3), cex=.8)
title(main="Unsupervised Random Forest MDS Plot of Iris Data \n PAM Groupings, k=3")

#Let's Look at the Silhouette Widths and Average Silhoutte Width to Decide on the number of groups!

#We could examine the silhouette plots for k=2,3,...
#Let's use our k=3 from above
plot(pamData)

#OR even better, let's plot the average over a range of groups!
pamSilhouetteWidth.fun(1 - iris.urf$proximity) # found in pamFunctions.R
title(main="PAM Silhouette Width, Unsupervised Random Forest Dissimilarity")

#3 or 4 Groups, it's pretty close!