### (Dis)similarity matrix shading using concept trees
### CSE 8331 - Spring 2012 - Michael Hahsler


library(seriation)

data(iris)

### shuffle the data set
iris <- iris[sample(1:nrow(iris)),]

head(iris)
plot(iris[,-5], col=iris$Species)


### compute Eucliden distance (col 5 is the class attribute)
d <- dist(iris[,-5])

### matrix shading
pimage(d, lower.tri=FALSE)

### use a concept tree (a.k.a. decision tree)
library(party)
tree <- ctree(Species~., data=iris)
tree

plot(tree)

pclass <- predict(tree, iris)
table(iris$Species, pclass)

pimage(d, order(pclass))

### use clustering
hc <- hclust(d)
hc
plot(hc)

cl <- cutree(hc, k=3)
table(iris$Species, cl)

### just cluster information
pimage(d, order(cl))

### use leaf node order in dendrogram
pimage(d, hc$order)


### use seriation instead
order <- seriate(d)
order

pimage(d, order)