geneSignatureFinder (version 2014.02.17)

BICs: Compute the bayesian information criteria

Description

This function computes the bayesian information criteria (bic) under two hypotheses: 1) the data are drawn from a single gaussian; 2) the data are drawn from a mixture of two gaussians.

Usage

BICs(ddata, clusters, cutoff = 2.5)

Arguments

ddata

dataset on which the bayesian information criteria have to be computed

clusters

a list 0-1 of the same length of dataset indicating to which cluster a value belongs

cutoff

real value that controls the RLS steps for the robust estimates of the parameters of the gaussians

Value

• bics
  list of two values: the first element is the bic under the first hypothesis, the second element is the bic under the secod hypothesis
• bic1Parameters
  list of two values that are the estimates of center and deviation under the first hypothesis
• bic2Parameters
  list of five values that are the estimates of center and deviation for the first cluster (clusters == 0), center and deviation for the second cluster (clusters == 1), and the estimate of the mixing parameter of the two gaussians.

Details

The estimates of location and scale of the gaussians are obtained by carrying out reweighted least squares (RLS) steps on starting robust estimates. If center0 and deviation0 are the initial robust estimates of the gaussian, to each value in the dataset is assigned weigth = 1 if ((value - center0)/deviation0) < cutoff^2, weigth = 0 otherwise. The center and deviation estimates are updated with those value with weigth = 1.

The paremeter that controls the mixing of the two gaussians under the second hypothesis is given by mean(clusters).

Examples

data(geNSCLC)
ans <- classify(geNSCLC[, "STX1A"])
BICs(geNSCLC[, "STX1A"], ans$clusters)

# an example with missing values
data(geNSCLC)
ans <- classify(geNSCLC[, "CALCA"])
BICs(geNSCLC[!ans$missing, "CALCA"], ans$clusters[!ans$missing])