pec (version 2023.04.12)

resolvesplitMethod: Resolve the splitMethod for estimation of prediction performance

Description

The function computes a matrix of random indices obtained by drawing from the row numbers of a data set either with or without replacement. The matrix can be used to repeatedly set up independent training and validation sets.

Usage

resolvesplitMethod(splitMethod, B, N, M)

Value

A list with the following components

name

the official name of the splitMethod

internal.name

the internal name of the splitMethod

index

a matrix of indices with B columns and either N or M rows, dependent on splitMethod

B

the value of the argument B

N

the value of the argument N

M

the value of the argument M

Arguments

splitMethod

String that determines the splitMethod to use. Available splitMethods are none/noPlan (no splitting), bootcv or outofbag (bootstrap cross-validation), cvK (K-fold cross-validation, e.g. cv10 gives 10-fold), boot632, boot632plus or boot632+, loocv (leave-one-out)

B

The number of repetitions.

N

The sample size

M

For subsampling bootstrap the size of the subsample. Note M<N.

Author

Thomas Alexander Gerds

Examples

  # BootstrapCrossValidation: Sampling with replacement   
  resolvesplitMethod("BootCv",N=10,B=10)

  # 10-fold cross-validation: repeated 2 times
  resolvesplitMethod("cv10",N=10,B=2)

  # leave-one-out cross-validation
  resolvesplitMethod("loocv",N=10)

  resolvesplitMethod("bootcv632plus",N=10,B=2)