COMMUNAL (version 1.0)

plotRange3D: Visualize cluster quality

Description

Given COMMUNAL outputs for a dataset, summarize the suitability of a range of cluster numbers according to validation measures. Input can be the output of clusterRange.

Usage

plotRange3D(test_range, ks, goodAlgs, goodMeasures, filename=NULL, ...)

Arguments

test_range

list of 2 items, where first element is list of "COMMUNAL" objects, and second is a vector of the number of data points clustered for each element in the list. Can be the output of

ks

range of cluster number k to plot. Must be subset of ks supplied to COMMUNAL.

goodAlgs

which algorithms to use in summarizing validation measures. Must be subset of algorithms supplied to COMMUNAL.

goodMeasures

which validation measures to use in summarizing validation measures. Must be subset of measures supplied to COMMUNAL.

filename

optionally specify filename to save a snapshot of the 3D image.

...

other arguments to pass down to 2D plot of mean z-score against k.

Value

• A 3D plot is generated, using the package "rgl". A matrix of the plotted values is returned. A 2D plot of average metric against k is also generated.

Details

A summarized validation measure value is computed for each value of k, for each dataset. This is done by first subsetting the data to the measures, ks, and algorithms of interest, and then computing averages of the measures for each dataset and k (number of clusters). For some validation measures, a lower value implies better clustering, and for others a higher value is better. Prior to averaging, measures that favor a lower value are multiplied by negative one. Furthermore, each measure is scaled to have zero mean and unit variance across all the datasets prior to averaging, so each measure has equal weight, and we can compare the plot across datasets. If the wrong number of clusters was identified

Examples

data(BRCA.results) # clusterRange output for breast cancer dataset
goodAlgs <- c("hierarchical", "kmeans", "model", "agnes", "som")
goodMeasures <- c('wb.ratio', 'avg.silwidth', 'dunn')
ks <- 2:8
(values <- plotRange3D(BRCA.results, ks, goodAlgs, goodMeasures))