DMwR (version 0.4.1)

join: Merging several compExp class objects

Description

This function can be used to join several compExp class objects into a single object. The merge is carried out assuming there there is something in common between the objects (e.g. all use the same learners on different data sets), and that the user specifies which property should be used for the merging process.

Usage

join(..., by = "datasets")

Arguments

...

The compExp class object names separated by commas

by

The dimension of the compExp class objects that should be used for the merge. All objects should have the same values on this dimension.

Value

The result of this function is a compExp object.

Details

The objects of class compExp (type "class?compExp" for details) contain several information on the results of an experimental comparison between several prediction models on several data sets. These experiments are carried out with the function experimentalComparison(). One of the "slots" of the objects of class compExp contains the actual results of the experiment on the different repetitions that were carried out. This slot is an array with four dimensions: "iterations","statistics","variants","datasets", in this order. This function allows the user the merge several objects of this class according to one of these four dimensions. Example uses of this function is a user that carries out a similar experiment (i.e. with the same experimental settings) on the same data sets twice, each time with a different set of learners being compared. This user might be interested in mergind the two compExp objects resulting from these experiments into a single object for comparing the results across all learners. This use should then use this function to join the two objects by "variants". Another example would be a set up where the same experiment with a set of learners was repeated with different sets of data sets. All the resulting objects vould be merged by "datasets" to obtain a single results object.

You should note that the merging is only possible if all objects share the same experimental settings. Obviously, it only makes sense to merge several objects into a single one by some dimension "x" if all other dimensions are equal.

References

Torgo, L. (2010) Data Mining using R: learning with case studies, CRC Press (ISBN: 9781439810187).

http://www.dcc.fc.up.pt/~ltorgo/DataMiningWithR

See Also

experimentalComparison, compExp, subset

Examples

## Run some experiments with the swiss data and tow different prediction models
data(swiss)

## First the user defined functions for obtaining the two models
cv.rpart <- function(form, train, test, ...) {
    model <- rpartXse(form, train, ...)
    preds <- predict(model, test)
    regr.eval(resp(form, test), preds,
              stats=c('mae','nmse'), train.y=resp(form, train))
}
cv.lm <- function(form, train, test, ...) {
    model <- lm(form, train, ...)
    preds <- predict(model, test)
    regr.eval(resp(form, test), preds,
              stats=c('mae','nmse'), train.y=resp(form, train))
}
## Now the evaluation of the two models, which will be done separately 
## just to illustrate the use of the join() function afterward
exp1 <- experimentalComparison(
               c(dataset(Infant.Mortality ~ ., swiss)),
               c(variants('cv.rpart',se=c(0,0.5,1))),
               cvSettings(1,10,1234))
exp2 <- experimentalComparison(
               c(dataset(Infant.Mortality ~ ., swiss)),
               c(variants('cv.lm')),
               cvSettings(1,10,1234))

## Now the examples of the join

## joining the two experiments by variants (i.e. models)
all <- join(exp1,exp2,by='variants')
bestScores(all)   # check the best results

## an example including also subsetting
bestScores(join(subset(exp1,stats='mae'),subset(exp2,stats='mae'),
                by='variants'))