caret (version 6.0-73)

resamples: Collation and Visualization of Resampling Results

Description

These functions provide methods for collection, analyzing and visualizing a set of resampling results from a common data set.

Usage

resamples(x, ...)
"resamples"(x, modelNames = names(x), ...)
"sort"(x, decreasing = FALSE, metric = x$metric[1], FUN = mean, ...)
"summary"(object, metric = object$metrics, ...)
"as.matrix"(x, metric = x$metric[1], ...)
"as.data.frame"(x, row.names = NULL, optional = FALSE, metric = x$metric[1], ...)
modelCor(x, metric = x$metric[1], ...)
"print"(x, ...)

Arguments

x

a list of two or more objects of class train, sbf or rfe with a common set of resampling indices in the control object. For sort.resamples, it is an object generated by resamples.

modelNames

an optional set of names to give to the resampling results

decreasing

logical. Should the sort be increasing or decreasing?

metric

a character string for the performance measure used to sort or computing the between-model correlations

FUN

a function whose first argument is a vector and returns a scalar, to be applied to each model's performance measure.

object

an object generated by resamples

row.names, optional

not currently used but included for consistency with as.data.frame

...

only used for sort and modelCor and captures arguments to pass to sort or FUN.

Value

For resamples: an object with class "resamples" with elements elements For sort.resamples a character string in the sorted order is generated. modelCor returns a correlation matrix.

Details

The ideas and methods here are based on Hothorn et al. (2005) and Eugster et al. (2008).

The results from train can have more than one performance metric per resample. Each metric in the input object is saved.

resamples checks that the resampling results match; that is, the indices in the object trainObject$control$index are the same. Also, the argument trainControl returnResamp should have a value of "final" for each model.

The summary function computes summary statistics across each model/metric combination.

References

Hothorn et al. The design and analysis of benchmark experiments. Journal of Computational and Graphical Statistics (2005) vol. 14 (3) pp. 675-699

Eugster et al. Exploratory and inferential analysis of benchmark experiments. Ludwigs-Maximilians-Universitat Munchen, Department of Statistics, Tech. Rep (2008) vol. 30

See Also

train, trainControl, diff.resamples, xyplot.resamples, densityplot.resamples, bwplot.resamples, splom.resamples

Examples

data(BloodBrain)
set.seed(1)

## tmp <- createDataPartition(logBBB,
##                            p = .8,
##                            times = 100)

## rpartFit <- train(bbbDescr, logBBB,
##                   "rpart", 
##                   tuneLength = 16,
##                   trControl = trainControl(
##                     method = "LGOCV", index = tmp))

## ctreeFit <- train(bbbDescr, logBBB,
##                   "ctree", 
##                   trControl = trainControl(
##                     method = "LGOCV", index = tmp))

## earthFit <- train(bbbDescr, logBBB,
##                   "earth",
##                   tuneLength = 20,
##                   trControl = trainControl(
##                     method = "LGOCV", index = tmp))

## or load pre-calculated results using:
## load(url("http://caret.r-forge.r-project.org/exampleModels.RData"))

## resamps <- resamples(list(CART = rpartFit,
##                           CondInfTree = ctreeFit,
##                           MARS = earthFit))

## resamps
## summary(resamps)