DendSer: Dendrogram Seriation: Ordering for Visualisation

Description

Re-arranges a dendrogram to optimize visualisation-based cost functions. The methods implemented here are desribed in Advances in Dendrogram Seriation for Application to Visualization, JCGS (2015) D. Earle and C.B. Hurley doi:10.1080/10618600.2013.874295.

Author(s)

Maintainer: Catherine Hurley catherine.hurley@mu.ie

Authors:

• Catherine B. Hurley

• Denise Earle

Implements dendrogram seriation

Description

Implements dendrogram seriation.

Usage

DendSer(
  h,
  ser_weight,
  cost = costBAR,
  node_op = NULL,
  costArg = NULL,
  maxloops = NULL,
  saveinfo = FALSE,
  direction = NULL,
  GW = NULL,
  ...
)

Arguments

Details

costED uses the Gruvaeus and Wainer 1972 algorithm, as provided by package gclus.

Value

Numeric vector giving an optimal dendrogram order

Author(s)

Catherine Hurley & Denise Earle

References

Gruvaeus, G. & Wainer, H. (1972), “Two additions to hierarchical cluster analysis”, British Journal of Mathematical and Statistical Psychology, 25, 200-206.

See Also

dser

Examples

	 			

d<- dist(iris[,-5])
h <- hclust(d,method="average")
ob<- DendSer(h,d)
opl<- DendSer(h,d,cost=costPL)
plotAsColor(d,ob)

w <- rowSums(iris[,-5])
ow <- DendSer(h,w,cost=costLS) # arranges cases by size, within hclust
stars(iris[ow,-5],labels=NULL, col.stars=cutree(h,3)[ow]) # and color by cluster
#stars(iris[ow,-5],labels=NULL, col.stars=iris[ow,5]) # or by species

Cost functions for DendSer Each of these functions evaluates the cost of an ordering.

Description

Cost functions for DendSer Each of these functions evaluates the cost of an ordering.

Usage

costLS(sw, o, target = seq_along(sw), ...)

costARc(sw, o, target = nrow(sw) - 1, ...)

costED(sw, o, node, se, ...)

costPL(sw, o, ...)

costLPL(sw, o, target = (nrow(sw) - 1):1, ...)

costBAR(sw, o, target = max(2, floor(nrow(sw)/5)), ...)

Arguments

Value

Result of cost

Author(s)

Catherine Hurley & Denise Earle

Cost function from seriation criterion

Description

Returns a cost function from seriation criterion (package seriation)

Usage

crit2cost(crit)

Arguments

Examples

	 			

d<- dist(iris[1:20,-5])
h <- hclust(d)
DendSer(h,d,cost=function(x,o,...) seriation::criterion(as.dist(x),o,method="AR_deviations"))
DendSer(h,d,cost=crit2cost("AR_deviations")) # short version of above
DendSer(h,d,cost=crit2cost("ME"))

# use DendSer methods via seriate
# seriation::get_order(seriate(d,method="Dendser", control=list(hclust=h,cost=costARc))) 
DendSer(h,d, cost=costARc)

Implements dendrogram seriation. Interface to DendSer.

Description

Implements dendrogram seriation. Interface to DendSer.

Usage

dser(x, ser_weight, cost = costBAR, ...)

## S3 method for class 'data.frame'
dser(x, ser_weight, cost = costBAR, ...)

## S3 method for class 'matrix'
dser(x, ser_weight, cost = costBAR, scale = TRUE, dmethod = "euclidean", ...)

## S3 method for class 'dist'
dser(x, ser_weight, cost = costBAR, hmethod = "average", ...)

## S3 method for class 'hclust'
dser(x, ser_weight, cost = costBAR, ...)

Arguments

Details

When x is a matrix or data.drame, forms a dist of rows using function dist with method = dmethod. When x is a dist, forms a hclust with method = hmethod which is then reordered.

Value

Numeric vector giving an optimal dendrogram order

Methods (by class)

• dser(data.frame): dser method

• dser(matrix): dser method

• dser(dist): dser method

• dser(hclust): dser method

Author(s)

Catherine Hurley & Denise Earle

Examples

	 			

iriss <- scale(iris[,-5])
plotAsColor(iriss,order.row=dser(iriss))



w <- prcomp(iris[,-5],scale=TRUE)$x[,1]
h<- hclust(dist(iriss))
h$order <- ow <- dser(h,w,cost=costLS) # arranges cases along first PC, within dendrogram


# compare re-rordered dendrogram to PC scores, w
dev.new(width=10,height=5)
par(mar=c(0,2,1,1))
layout(matrix(1:2, nrow = 2), heights = c(4,1.5) )
par(cex=.7)
plot(h,main="",xlab="",hang=-1,labels=FALSE)
u <- par("usr")
par(mar=c(1,2,0,1))
 

plot.new()
par(usr=c(u[1:2],min(w),max(w)))

x<- 1:length(w)
rect(x-.5,0,x+.5,w[ow],col=cutree(h,3)[ow]+1)

Function that plots a matrix as a color image

Description

Function that plots a matrix as a color image, in matrix order.

Usage

plotAsColor(
  m,
  order = NULL,
  order.col = order,
  order.row = order,
  rank = FALSE,
  border.color = "grey70",
  labels = FALSE,
  x = 1:ncol(d),
  y = 1:nrow(d),
  ...
)

Arguments

See Also

See Also as image.

Examples

	 			

d<- dist(scale(iris[,-5]))
plotAsColor(d,dser(d,hmethod="average"))