KullbackLeiblerSelection: Selection of Differential Distributions with Kullback Leibler Distance

Description

Ranks features by largest Kullback-Leibler distance and chooses the features which have best resubstitution performance.

Usage

"KullbackLeiblerSelection"(expression, classes, ...) "KullbackLeiblerSelection"(expression, datasetName, trainParams, predictParams, resubstituteParams, ..., selectionName, verbose = 3)

Arguments

expression

Either a matrix or ExpressionSet containing the training data. For a matrix, the rows are features, and the columns are samples.

classes

A vector of class labels.

datasetName

A name for the dataset used. Stored in the result.

trainParams

A container of class TrainParams describing the classifier to use for training.

predictParams

A container of class PredictParams describing how prediction is to be done.

resubstituteParams

An object of class ResubstituteParams describing the performance measure to consider and the numbers of top features to try for resubstitution classification.

...

Variables passed to getLocationsAndScales.

selectionName

A name to identify this selection method by. Stored in the result.

verbose

A number between 0 and 3 for the amount of progress messages to give. This function only prints progress messages if the value is 3.

Value

An object of class SelectResult or a list of such objects, if the classifier which was used for determining resubstitution error rate made a number of prediction varieties.

Details

The distance is defined as $0.5 * (location1 - location2)^2 / scale1^2 + (location1 - location2)^2 / scale2^2 + scale1^2 / scale2^2 + scale2^2 / scale1^2 $ The subscripts denote the group which the parameter is calculated for.

Examples

Run this code

  if(require(sparsediscrim))
  {
    # First 20 features have bimodal distribution for Poor class. Other 80 features have normal distribution for
    # both classes.
    genesMatrix <- sapply(1:25, function(sample) c(rnorm(20, sample(c(8, 12), 20, replace = TRUE), 1), rnorm(80, 10, 1)))
    genesMatrix <- cbind(genesMatrix, sapply(1:25, function(sample) rnorm(100, 10, 1)))
    classes <- factor(rep(c("Poor", "Good"), each = 25))
    KullbackLeiblerSelection(genesMatrix, classes, "Example",
                             trainParams = TrainParams(naiveBayesKernel, FALSE, doesTests = TRUE),
                             predictParams = PredictParams(function(){}, FALSE, getClasses = function(result) result),
                             resubstituteParams = ResubstituteParams(nFeatures = seq(10, 100, 10), performanceType = "balanced", better = "lower")
                             )
  }