rbCV: Rolling block cross validation

In ammarsahab/smoothCV: Wrappers and parameter optimization for time series smoothing methods

Rolling block cross validation

Description

The algorithm starts by splitting all observations into k folds. Specified portions of the fold will be split into training and test sets. Grid search is then committed using MA.Grid or ES.Grid. This occurs in all k folds.

Usage

rbCV(
  fullset,
  trainsplit,
  folds,
  type,
  start,
  end,
  dist,
  localoptim = F,
  alphrange,
  betarange
)

Arguments

fullset	A set of univariate time series data. Can be a vector (type double) or a data.table.
trainsplit	Proportion of observations to be used as training set in each fold. The remaining observations will automatically be allocated to the test set.
folds	Number of folds to divide the rest of the data into.
type	Type of smoothing. Can be "SMA", "DMA", "SES", "DES".
start	(if type="SMA" or type="DMA") Starting value for m (see sma.dt and dma.dt).
end	(if type="SMA" or type="DMA") Maximum value for m. Must be less than or equal to the length of the training set for SMA, or less than or equal to the length of the training set divided by two for DMA.
dist	(if type="SMA" or type="DMA") Distance between successive values for m. The vector of m values will be constructed as seq(start,end,dist).
localoptim	Whether to optimize in each training set. If true, there is no need to provide values for alpha and beta..
alphrange	(if type="SES" or type="DES") A vector of parameters for the level component. Can be created using seq, or by manually specifying a vector.
betarange	(if type="DES") A vector of parameters for the trend component. Can be created using seq, or by manually specifying a vector. Not used if type="SES"

Value

A data.table containing all parameter combinations during each iteration (fold) with their respective MSE and MAPE values. Can be grouped by parameter values to obtain the mean error for each parameter value.

Examples

fcCV(fullset=crudenow$Close, initialn=36, folds=12, type="DES", localoptim=F,
alphrange=seq(0.1,1,0.1), betarange=seq(0.1,1,0.1))
fcCV(fullset=crudenow$Close, initialn=36, folds=12, type="SMA", localoptim=F,
start=2, end=30, dist=3)

ammarsahab/smoothCV documentation built on April 18, 2022, 4:39 p.m.