ClassifyR (version 1.6.2)

distribution: Get Frequencies of Feature Selection and Sample Errors

Description

There are two modes. For aggregating feature selection results, the function counts the number of times each feature was selected in all cross validations. For aggregating classification results, the error rate for each sample is calculated. This is useful in identifying outlier samples that are difficult to classify.

Usage

"distribution"(result, dataType = c("features", "samples"), plotType = c("density", "histogram"), summaryType = c("percentage", "count"), plot = TRUE, xMax = NULL, xLabel = "Percentage of Cross-validations", yLabel = "Density", title = "Distribution of Feature Selections", fontSizes = c(24, 16, 12), ...)

Arguments

result

An object of class ClassifyResult.

dataType

Whether to calculate sample-wise error rate or the number of times a feature was selected.

plotType

Whether to draw a probability density curve or a histogram.

summaryType

Whether to summarise the feature selections as a percentage or count.

plot

Whether to draw a plot of the frequency of selection or error rate.

xMax

Maximum data value to show in plot.

xLabel

The label for the x-axis of the plot.

yLabel

The label for the y-axis of the plot.

title

An overall title for the plot.

fontSizes

A vector of length 3. The first number is the size of the title. The second number is the size of the axes titles. The third number is the size of the axes values.

...

Further parameters, such as colour and fill, passed to geom_histogram or stat_density, depending on the value of plotType.

Value

If type is "features", a vector as long as the number of features that were chosen at least once containing the number of times the feature was chosen in cross validations or the percentage of times chosen. If type is "samples", a vector as long as the number of samples, containing the cross-validation error rate of the sample. If plot is TRUE, then a plot is also made on the current graphics device.

Examples

  if(require(curatedOvarianData) && require(sparsediscrim))
  {
    data(TCGA_eset)
    badOutcome <- which(pData(TCGA_eset)[, "vital_status"] == "deceased" & pData(TCGA_eset)[, "days_to_death"] <= 365)
    goodOutcome <- which(pData(TCGA_eset)[, "vital_status"] == "living" & pData(TCGA_eset)[, "days_to_death"] >= 365 * 5)
    TCGA_eset <- TCGA_eset[, c(badOutcome, goodOutcome)]
    classes <- factor(rep(c("Poor", "Good"), c(length(badOutcome), length(goodOutcome))))
    pData(TCGA_eset)[, "class"] <- classes
    result <- runTests(TCGA_eset, "Ovarian Cancer", "Differential Expression", resamples = 2, fold = 2)
    sampleDistribution <- distribution(result, "samples", xLabel = "Sample Error Rate",
                                       title = "Distribution of Error Rates")
    featureDistribution <- distribution(result, "features", summaryType = "count", plotType = "histogram",
                                        xLabel = "Number of Cross-validations", yLabel = "Count",
                                        binwidth = 1)
    print(head(sampleDistribution))
    print(head(featureDistribution))
  }