Agglomerative Nesting (Hierarchical Clustering)

Description

Computes agglomerative hierarchical clustering of the dataset.

Usage

agnes(x, diss = inherits(x, "dist"), metric = "euclidean",
      stand = FALSE, method = "average", par.method,
      keep.diss = n < 100, keep.data = !diss, trace.lev = 0)

Arguments

Details

agnes is fully described in chapter 5 of Kaufman and Rousseeuw (1990). Compared to other agglomerative clustering methods such as hclust, agnes has the following features: (a) it yields the agglomerative coefficient (see agnes.object) which measures the amount of clustering structure found; and (b) apart from the usual tree it also provides the banner, a novel graphical display (see plot.agnes).

The agnes-algorithm constructs a hierarchy of clusterings.
At first, each observation is a small cluster by itself. Clusters are merged until only one large cluster remains which contains all the observations. At each stage the two nearest clusters are combined to form one larger cluster.

For method="average", the distance between two clusters is the average of the dissimilarities between the points in one cluster and the points in the other cluster.
In method="single", we use the smallest dissimilarity between a point in the first cluster and a point in the second cluster (nearest neighbor method).
When method="complete", we use the largest dissimilarity between a point in the first cluster and a point in the second cluster (furthest neighbor method).

The method = "flexible" allows (and requires) more details: The Lance-Williams formula specifies how dissimilarities are computed when clusters are agglomerated (equation (32) in K&R(1990), p.237). If clusters C_1 and C_2 are agglomerated into a new cluster, the dissimilarity between their union and another cluster Q is given by

D(C_1 \cup C_2, Q) = α_1 * D(C_1, Q) + α_2 * D(C_2, Q) + β * D(C_1,C_2) + γ * |D(C_1, Q) - D(C_2, Q)|,

where the four coefficients (α_1, α_2, β, γ) are specified by the vector par.method, either directly as vector of length 4, or (more conveniently) if par.method is of length 1, say = α, par.method is extended to give the “Flexible Strategy” (K&R(1990), p.236 f) with Lance-Williams coefficients (α_1 = α_2 = α, β = 1 - 2α, γ=0).
Also, if length(par.method) == 3, γ = 0 is set.

Care and expertise is probably needed when using method = "flexible" particularly for the case when par.method is specified of longer length than one. Since cluster version 2.0, choices leading to invalid merge structures now signal an error (from the C code already). The weighted average (method="weighted") is the same as method="flexible", par.method = 0.5. Further, method= "single" is equivalent to method="flexible", par.method = c(.5,.5,0,-.5), and method="complete" is equivalent to method="flexible", par.method = c(.5,.5,0,+.5).

The method = "gaverage" is a generalization of "average", aka “flexible UPGMA” method, and is (a generalization of the approach) detailed in Belbin et al. (1992). As "flexible", it uses the Lance-Williams formula above for dissimilarity updating, but with α_1 and α_2 not constant, but proportional to the sizes n_1 and n_2 of the clusters C_1 and C_2 respectively, i.e,

α_j = α'_j * n_1/(n_1 + n_2),

where α'_1, α'_2 are determined from par.method, either directly as (α_1, α_2, β, γ) or (α_1, α_2, β) with γ = 0, or (less flexibly, but more conveniently) as follows:

Belbin et al proposed “flexible beta”, i.e. the user would only specify β (as par.method), sensibly in

-1 ≤ β < 1,

and β determines α'_1 and α'_2 as

α'_j = 1 - β,

and γ = 0.

This β may be specified by par.method (as length 1 vector), and if par.method is not specified, a default value of -0.1 is used, as Belbin et al recommend taking a β value around -0.1 as a general agglomerative hierarchical clustering strategy.

Note that method = "gaverage", par.method = 0 (or par.method = c(1,1,0,0)) is equivalent to the agnes() default method "average".

Value

an object of class "agnes" (which extends "twins") representing the clustering. See agnes.object for details, and methods applicable.

BACKGROUND

Cluster analysis divides a dataset into groups (clusters) of observations that are similar to each other.

Hierarchical methods

like agnes, diana, and mona construct a hierarchy of clusterings, with the number of clusters ranging from one to the number of observations.

Partitioning methods

like pam, clara, and fanny require that the number of clusters be given by the user.

Author(s)

Method "gaverage" has been contributed by Pierre Roudier, Landcare Research, New Zealand.

References

Kaufman, L. and Rousseeuw, P.J. (1990). (=: “K&R(1990)”) Finding Groups in Data: An Introduction to Cluster Analysis. Wiley, New York.

Anja Struyf, Mia Hubert and Peter J. Rousseeuw (1996) Clustering in an Object-Oriented Environment. Journal of Statistical Software 1. doi: 10.18637/jss.v001.i04

Struyf, A., Hubert, M. and Rousseeuw, P.J. (1997). Integrating Robust Clustering Techniques in S-PLUS, Computational Statistics and Data Analysis, 26, 17–37.

Lance, G.N., and W.T. Williams (1966). A General Theory of Classifactory Sorting Strategies, I. Hierarchical Systems. Computer J. 9, 373–380.

Belbin, L., Faith, D.P. and Milligan, G.W. (1992). A Comparison of Two Approaches to Beta-Flexible Clustering. Multivariate Behavioral Research, 27, 417–433.

See Also

agnes.object, daisy, diana, dist, hclust, plot.agnes, twins.object.

Examples

data(votes.repub)
agn1 <- agnes(votes.repub, metric = "manhattan", stand = TRUE)
agn1
plot(agn1)

op <- par(mfrow=c(2,2))
agn2 <- agnes(daisy(votes.repub), diss = TRUE, method = "complete")
plot(agn2)
## alpha = 0.625 ==> beta = -1/4  is "recommended" by some
agnS <- agnes(votes.repub, method = "flexible", par.meth = 0.625)
plot(agnS)
par(op)

## "show" equivalence of three "flexible" special cases
d.vr <- daisy(votes.repub)
a.wgt  <- agnes(d.vr, method = "weighted")
a.sing <- agnes(d.vr, method = "single")
a.comp <- agnes(d.vr, method = "complete")
iC <- -(6:7) # not using 'call' and 'method' for comparisons
stopifnot(
  all.equal(a.wgt [iC], agnes(d.vr, method="flexible", par.method = 0.5)[iC])   ,
  all.equal(a.sing[iC], agnes(d.vr, method="flex", par.method= c(.5,.5,0, -.5))[iC]),
  all.equal(a.comp[iC], agnes(d.vr, method="flex", par.method= c(.5,.5,0, +.5))[iC]))

## Exploring the dendrogram structure
(d2 <- as.dendrogram(agn2)) # two main branches
d2[[1]] # the first branch
d2[[2]] # the 2nd one  { 8 + 42  = 50 }
d2[[1]][[1]]# first sub-branch of branch 1 .. and shorter form
identical(d2[[c(1,1)]],
          d2[[1]][[1]])
## a "textual picture" of the dendrogram :
str(d2)

data(agriculture)

## Plot similar to Figure 7 in ref
## Not run: plot(agnes(agriculture), ask = TRUE)


data(animals)
aa.a  <- agnes(animals) # default method = "average"
aa.ga <- agnes(animals, method = "gaverage")
op <- par(mfcol=1:2, mgp=c(1.5, 0.6, 0), mar=c(.1+ c(4,3,2,1)),
          cex.main=0.8)
plot(aa.a,  which.plot = 2)
plot(aa.ga, which.plot = 2)
par(op)


## Show how "gaverage" is a "generalized average":
aa.ga.0 <- agnes(animals, method = "gaverage", par.method = 0)
stopifnot(all.equal(aa.ga.0[iC], aa.a[iC]))