10.01.2015 Views

Package 'fpc' - open source solution for an Internet free, intelligent ...

Package 'fpc' - open source solution for an Internet free, intelligent ...

Package 'fpc' - open source solution for an Internet free, intelligent ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

22 cluster.stats<br />

Usage<br />

cluster.stats(d = NULL, clustering, alt.clustering = NULL,<br />

noisecluster=FALSE,<br />

silhouette = TRUE, G2 = FALSE, G3 = FALSE,<br />

wgap=TRUE, sepindex=TRUE, sepprob=0.1,<br />

sepwithnoise=TRUE,<br />

compareonly = FALSE,<br />

aggregateonly = FALSE)<br />

Arguments<br />

d<br />

clustering<br />

a dist<strong>an</strong>ce object (as generated by dist) or a dist<strong>an</strong>ce matrix between cases.<br />

<strong>an</strong> integer vector of length of the number of cases, which indicates a clustering.<br />

The clusters have to be numbered from 1 to the number of clusters.<br />

alt.clustering <strong>an</strong> integer vector such as <strong>for</strong> clustering, indicating <strong>an</strong> alternative clustering.<br />

If provided, the corrected R<strong>an</strong>d index <strong>an</strong>d Meila’s VI <strong>for</strong> clustering vs.<br />

alt.clustering are computed.<br />

noisecluster<br />

silhouette<br />

logical. If TRUE, it is assumed that the largest cluster number in clustering<br />

denotes a ’noise class’, i.e. points that do not belong to <strong>an</strong>y cluster. These points<br />

are not taken into account <strong>for</strong> the computation of all functions of within <strong>an</strong>d<br />

between cluster dist<strong>an</strong>ces including the validation indexes.<br />

logical. If TRUE, the silhouette statistics are computed, which requires package<br />

cluster.<br />

G2 logical. If TRUE, Goodm<strong>an</strong> <strong>an</strong>d Kruskal’s index G2 (cf. Gordon (1999), p. 62)<br />

is computed. This executes lots of sorting algorithms <strong>an</strong>d c<strong>an</strong> be very slow (it<br />

has been improved by R. Fr<strong>an</strong>cois - th<strong>an</strong>ks!)<br />

G3<br />

wgap<br />

sepindex<br />

sepprob<br />

sepwithnoise<br />

compareonly<br />

aggregateonly<br />

logical. If TRUE, the index G3 (cf. Gordon (1999), p. 62) is computed. This<br />

executes sort on all dist<strong>an</strong>ces <strong>an</strong>d c<strong>an</strong> be extremely slow.<br />

logical. If TRUE, the widest within-cluster gaps (largest link in within-cluster<br />

minimum sp<strong>an</strong>ning tree) are computed.<br />

logical. If TRUE, a separation index is computed, defined based on the dist<strong>an</strong>ces<br />

<strong>for</strong> every point to the closest point not in the same cluster. The separation index<br />

is then the me<strong>an</strong> of the smallest proportion sepprob of these. This allows to<br />

<strong>for</strong>malise separation less sensitive to a single or a few ambiguous points. The<br />

output component corresponding to this is sindex, not separation!<br />

numerical between 0 <strong>an</strong>d 1, see sepindex.<br />

logical. If TRUE <strong>an</strong>d sepindex <strong>an</strong>d noisecluster are both TRUE, the noise<br />

points are incorporated as cluster in the separation index computation. Also they<br />

are taken into account <strong>for</strong> the computation <strong>for</strong> the minimum cluster separation<br />

regardless of sepindex.<br />

logical. If TRUE, only the corrected R<strong>an</strong>d index <strong>an</strong>d Meila’s VI are computed<br />

<strong>an</strong>d given out (this requires alt.clustering to be specified).<br />

logical. If TRUE (<strong>an</strong>d not compareonly), no clusterwise but only aggregated<br />

in<strong>for</strong>mation is given out (this cuts the size of the output down a bit).

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!