Wavelet levels selection procedure

Share:

Description

A grouped backward variable selection procedure for selecting the most significant wavelet levels of a functional variable. The groups are the wavelet coefficients belonging to the same frequency level.

Usage

1
2
selectLevel(design, ydata, typeRF = ifelse(is.factor(ydata), "classif", "reg"), 
            verbose = TRUE, ntree = 500, ...)

Arguments

design

The design matrix of a functional variable.

ydata

The outcome data. Must be a factor for classification.

typeRF

The type of forest we want to construct, ‘classif’ for classification or ‘reg’ for regression.

verbose

Should the details be printed.

ntree

The number of trees in the forests (default: 500).

...

optional parameters to be passed to the ‘varImpGroup’ function.

Value

An object of class fRFE which is a list with the following components:

nselected

The number of selected wavelet levels.

selection

The selected wavelet levels.

selectionIndexes

The indexes of selected wavelet levels in the input matrix ‘design’.

error

The prediction error computed in each iteration of the backward procedure.

typeRF

The type of the forests, classification or regression.

ranking

The final ranking of the wavelet levels.

rankingIndexes

The final ranking indexes of the wavelet levels.

Author(s)

Baptiste Gregorutti

References

Gregorutti, B., Michel, B. and Saint Pierre, P. (2015). Grouped variable importance with random forests and application to multiple functional data analysis, Computational Statistics and Data Analysis 90, 15-35.

See Also

selectGroup,selectFunctional,varImpGroup

Examples

1
2
3
4
5
6
7
  data(toyRegFD)
  x <- toyRegFD$FDlist[[1]]
  y <- toyRegFD$Y

  design <- projectWavelet(xdata=x)
  summary(levSel <- selectLevel(design, y, ntree=100, verbose=TRUE))
  plot(levSel)