EVOLUTION-MANAGER

Edit File: agnes.html

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><title>R: Agglomerative Nesting (Hierarchical Clustering)</title>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<link rel="stylesheet" type="text/css" href="R.css" />
</head><body>

<table width="100%" summary="page for agnes {cluster}"><tr><td>agnes {cluster}</td><td style="text-align: right;">R Documentation</td></tr></table>

<h2>Agglomerative Nesting (Hierarchical Clustering)</h2>

<h3>Description</h3>

Computes agglomerative hierarchical clustering of the dataset.

<h3>Usage</h3>

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

<h3>Arguments</h3>

data matrix or data frame, or dissimilarity matrix, depending on the
value of the <code>diss</code> argument.

In case of a matrix or data frame, each row corresponds to an observation,
and each column corresponds to a variable. All variables must be numeric.
Missing values (NAs) are allowed.

In case of a dissimilarity matrix, <code>x</code> is typically the output of
<code><a href="daisy.html">daisy</a></code> or <code><a href="../../stats/html/dist.html">dist</a></code>.
Also a vector with length n*(n-1)/2 is allowed (where n is the number
of observations), and will be interpreted in the same way as the
output of the above-mentioned functions. Missing values (NAs) are not
allowed.

</td></tr>
<tr valign="top"><td><code>diss</code></td>
<td>

logical flag: if TRUE (default for <code>dist</code> or
<code>dissimilarity</code> objects), then <code>x</code> is assumed to be a
dissimilarity matrix. If FALSE, then <code>x</code> is treated as
a matrix of observations by variables.

</td></tr>
<tr valign="top"><td><code>metric</code></td>
<td>

character string specifying the metric to be used for calculating
dissimilarities between observations.
The currently available options are <code>"euclidean"</code> and <code>"manhattan"</code>.
Euclidean distances are root sum-of-squares of differences, and
manhattan distances are the sum of absolute differences.
If <code>x</code> is already a dissimilarity matrix, then this argument will
be ignored.

</td></tr>
<tr valign="top"><td><code>stand</code></td>
<td>

logical flag: if TRUE, then the measurements in <code>x</code> are
standardized before calculating the dissimilarities. Measurements
are standardized for each variable (column), by subtracting the
variable's mean value and dividing by the variable's mean absolute
deviation. If <code>x</code> is already a dissimilarity matrix, then this
argument will be ignored.

</td></tr>
<tr valign="top"><td><code>method</code></td>
<td>

character string defining the clustering method. The six methods
implemented are
<code>"average"</code> ([unweighted pair-]group [arithMetic] average method, aka &lsquo;UPGMA&rsquo;),
<code>"single"</code> (single linkage), <code>"complete"</code> (complete linkage),
<code>"ward"</code> (Ward's method),
<code>"weighted"</code> (weighted average linkage, aka &lsquo;WPGMA&rsquo;), its generalization
<code>"flexible"</code> which uses (a constant version of)
the Lance-Williams formula and the <code>par.method</code> argument, and
<code>"gaverage"</code> a generalized <code>"average"</code> aka &ldquo;flexible
UPGMA&rdquo; method also using the Lance-Williams formula and <code>par.method</code>.

The default is <code>"average"</code>.

</td></tr>
<tr valign="top"><td><code>par.method</code></td>
<td>

If <code>method</code> is <code>"flexible"</code> or <code>"gaverage"</code>, a numeric
vector of length 1, 3, or 4, (with a default for <code>"gaverage"</code>), see in
the details section.

</td></tr>
<tr valign="top"><td><code>keep.diss, keep.data</code></td>
<td>
logicals indicating if the dissimilarities
and/or input data <code>x</code> should be kept in the result. Setting
these to <code>FALSE</code> can give much smaller results and hence even save
memory allocation time.
</td></tr>
<tr valign="top"><td><code>trace.lev</code></td>
<td>
integer specifying a trace level for printing
diagnostics during the algorithm. Default <code>0</code> does not print
anything; higher values print increasingly more.
</td></tr>
</table>

<h3>Details</h3>

<code>agnes</code> is fully described in chapter 5 of Kaufman and Rousseeuw (1990).
Compared to other agglomerative clustering methods such as <code>hclust</code>,
<code>agnes</code> has the following features: (a) it yields the
agglomerative coefficient (see <code><a href="agnes.object.html">agnes.object</a></code>)
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 <code><a href="plot.agnes.html">plot.agnes</a></code>).

The <code>agnes</code>-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 <code>method="average"</code>, 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 <code>method="single"</code>, we use the smallest dissimilarity between a
point in the first cluster and a point in the second cluster (nearest
neighbor method).
 
When <code>method="complete"</code>, we use the largest dissimilarity
between a point in the first cluster and a point in the second cluster
(furthest neighbor method).

The <code>method = "flexible"</code> allows (and requires) more details:
The Lance-Williams formula specifies how dissimilarities are
computed when clusters are agglomerated (equation (32) in K&amp;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) = &alpha;_1 * D(C_1, Q) + &alpha;_2 * D(C_2, Q) +
 &beta; * D(C_1,C_2) + &gamma; * |D(C_1, Q) - D(C_2, Q)|,

where the four coefficients (&alpha;_1, &alpha;_2, &beta;, &gamma;)
are specified by the vector <code>par.method</code>, either directly as vector of
length 4, or (more conveniently) if <code>par.method</code> is of length 1,
say = &alpha;, <code>par.method</code> is extended to
give the &ldquo;Flexible Strategy&rdquo; (K&amp;R(1990), p.236 f) with
Lance-Williams coefficients (&alpha;_1 = &alpha;_2 = &alpha;, &beta; =
 1 - 2&alpha;, &gamma;=0). 
Also, if <code>length(par.method) == 3</code>, &gamma; = 0 is set.

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

The <code>method = "gaverage"</code> is a generalization of <code>"average"</code>, aka
&ldquo;flexible UPGMA&rdquo; method, and is (a generalization of the approach)
detailed in Belbin et al. (1992). As <code>"flexible"</code>, it uses the
Lance-Williams formula above for dissimilarity updating, but with
&alpha;_1 and &alpha;_2 not constant, but proportional to
the sizes n_1 and n_2 of the clusters C_1 and
C_2 respectively, i.e,


 &alpha;_j = &alpha;'_j * n_1/(n_1 + n_2),

where &alpha;'_1, &alpha;'_2 are determined from <code>par.method</code>,
either directly as (&alpha;_1, &alpha;_2, &beta;, &gamma;) or
(&alpha;_1, &alpha;_2, &beta;) with &gamma; = 0, or (less flexibly,
but more conveniently) as follows:

Belbin et al proposed &ldquo;flexible beta&rdquo;, i.e. the user would only
specify &beta; (as <code>par.method</code>), sensibly in

-1 &le; &beta; &lt; 1,

and &beta; determines &alpha;'_1 and &alpha;'_2 as

&alpha;'_j = 1 - &beta;,
 and &gamma; = 0.

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

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

<h3>Value</h3>

an object of class <code>"agnes"</code> (which extends <code>"twins"</code>)
representing the clustering. See <code><a href="agnes.object.html">agnes.object</a></code> for
details, and methods applicable.

<h3>BACKGROUND</h3>

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

<dl>
<dt>Hierarchical methods</dt><dd>like
<code>agnes</code>, <code><a href="diana.html">diana</a></code>, and <code><a href="mona.html">mona</a></code>
construct a hierarchy of clusterings, with the number of clusters
ranging from one to the number of observations.
</dd>
<dt>Partitioning methods</dt><dd>like
<code><a href="pam.html">pam</a></code>, <code><a href="clara.html">clara</a></code>, and <code><a href="fanny.html">fanny</a></code>
require that the number of clusters be given by the user.
</dd>
</dl>

<h3>Author(s)</h3>

Method <code>"gaverage"</code> has been contributed by Pierre Roudier, Landcare
Research, New Zealand.

<h3>References</h3>

Kaufman, L. and Rousseeuw, P.J. (1990). (=: &ldquo;K&amp;R(1990)&rdquo;)
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: <a href="http://doi.org/10.18637/jss.v001.i04">10.18637/jss.v001.i04</a>

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

Lance, G.N., and W.T. Williams (1966).
A General Theory of Classifactory Sorting Strategies, I. Hierarchical
Systems.
Computer J. 9, 373&ndash;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&ndash;433.

<code><a href="agnes.object.html">agnes.object</a></code>, <code><a href="daisy.html">daisy</a></code>, <code><a href="diana.html">diana</a></code>,
<code><a href="../../stats/html/dist.html">dist</a></code>, <code><a href="../../stats/html/hclust.html">hclust</a></code>, <code><a href="plot.agnes.html">plot.agnes</a></code>,
<code><a href="twins.object.html">twins.object</a></code>.

<h3>Examples</h3>

<pre>
data(votes.repub)
agn1 &lt;- agnes(votes.repub, metric = "manhattan", stand = TRUE)
agn1
plot(agn1)

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

## "show" equivalence of three "flexible" special cases
d.vr &lt;- daisy(votes.repub)
a.wgt  &lt;- agnes(d.vr, method = "weighted")
a.sing &lt;- agnes(d.vr, method = "single")
a.comp &lt;- agnes(d.vr, method = "complete")
iC &lt;- -(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 &lt;- 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  &lt;- agnes(animals) # default method = "average"
aa.ga &lt;- agnes(animals, method = "gaverage")
op &lt;- 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 &lt;- agnes(animals, method = "gaverage", par.method = 0)
stopifnot(all.equal(aa.ga.0[iC], aa.a[iC]))
</pre>

<hr /><div style="text-align: center;">[Package cluster version 2.0.8 <a href="00Index.html">Index</a>]</div>
</body></html>