PLS-LDA

Description

LDA models using PLS latent variables as input space.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
  plslda(X, Y, grouping, comps = TRUE, ncomp = min(dim(X)),
    stripped = TRUE, ...)

  ## S3 method for class 'plslda'
 predict(object, newdata, ...)

  ## S3 method for class 'plslda'
 coef(object, ...)

  ## S3 method for class 'plslda'
 center(object, ...)

Arguments

X

input variate matrix

Y

matrix with class membership (classes correspond to columns). If missing, factor2matrix (grouping) is used.

grouping

factor with class membership. If missing, hardclasses (Y) is used.

comps

which latent variables should be used?

ncomp

how many latent variables should be calculated?

stripped

should the model be stripped to save memory? (Stripping is different from plsr's stripping.)

...

further parameters for plsr, importantly, the number of PLS variates to be used can be given via ncomp.

further arguments for link[MASS]{predict.lda} and link[pls]{predict.mvr}

object

the plslda model

newdata

matrix with new cases to be predicted.

Details

For the moment only kernelpls.fit (see kernelpls.fit for the original) is supported.

Value

object of class "plslda", consisting of the mvr object returned by plsr and the lda object returned by lda.

list with results from link[MASS]{predict.lda} plus the pls scores used for LDA prediction in element $scores

Author(s)

Claudia Beleites

See Also

plsr, lda

predict.lda, predict.mvr rotate for rotation of the LDA part of the model