predict.rock: Rock Clustering

In cba: Clustering for Business Analytics

Rock Clustering

Description

Classify the rows of a data matrix using the Rock classifier.

Usage

## S3 method for class 'rock'
predict(object, x, drop = 1, ...)

## S3 method for class 'rock'
fitted(object, ...)

Arguments

object	an object of class rock.
x	a data matrix containing test or new samples.
drop	cluster size threshold.
...	further (unused) arguments.

Details

Provides a wrapper function to the Rock classifier: cluster memberships of the (row) samples are determined by majority vote using the size (of the cluster) weighted number of links.

The classifier uses random tie-breaking and assigns samples with zero link counts to NA. This allows for detection of possible outliers, or interesting patterns.

In the case a drop value greater than zero is specified, all clusters with size equal or less than this value are removed from the classifier. Especially, fitted uses a threshold of one because for singleton clusters the neighborhood is empty.

Note that for the training data the predicted memberships need not necessarily be the same as those obtained from the cluster algorithm.

Value

A list with the following components:

cl	a factor of cluster memberships of the samples.
size	a vector of cluster sizes.

See Also

rockCluster for obtaining a rock classifier and gknn for k-nearest neighbor classification.

Examples

### example from paper
data(Votes)
x <- as.dummy(Votes[-17])
rc <- rockCluster(x, n=2, theta=0.73, debug=TRUE)
rp <- predict(rc, x)
table(rp$cl)

cba documentation built on Sept. 11, 2024, 9:32 p.m.