rminer (version 1.1)

crossvaldata: Computes k-fold cross validation for rminer models.

Description

Computes k-fold cross validation for rminer models.

Usage

crossvaldata(x, data, theta.fit, theta.predict, ngroup = n, 
             order = FALSE, model, task, feature = "none",
             ...)

Arguments

x

See fit for details.

data

See fit for details.

theta.fit

fitting function

theta.predict

prediction function

ngroup

number of folds

order

if TRUE then a static ordered sampling cross-validation is adopted (e.g. useful for time series data), else the normal random sampling is adopted.

model

See fit for details.

task

See fit for details.

feature

See fit for details.

...

Additional parameters sent to theta.fit (e.g. search, mpar, scale, transform)

Value

• Returns a list with:
  • $cv.fit -- all predictions (factor iftask="class", matrix iftask="prob"or numeric iftask="reg");
  • $mpar -- matrix with the mpar for each fold;
  • $attributes -- the selected attributes for each fold if a feature selection algorithm was adopted;
  • $ngroup -- the number of folds;
  • $leave.out -- the computed size for each fold (=nrow(data)/ngroup);
  • $groups -- vector list with the indexes of each group;
  • $call -- the call of this function;

Details

Standard k-fold cross-validation but adopted for rminer models. For classification tasks ("class" or "prob") a stratified sampling is used (the class distributions are identical for each fold).

References

Check the crossval function of the bootstrap library.

See Also

holdout, fit, mining and predict.fit.

Examples

data(iris)
 M=crossvaldata(Species~.,iris,fit,predict,ngroup=3,model="mlpe",
                task="prob",search=4,mpar=c(3,100,"holdout",2/3,"AUC"))