inlr: Blocks for INLR

In mlesnoff/rnirs: Dimension reduction, Regression and Discrimination for Chemometrics

Blocks for INLR

Description

Function inlr builds matrix blocks useable for "implicit non-linear latent variable regression (INLR)" (Berglund & Wold 1997). INLR is in the same idea of direct non linear kernel regression and discrimination methods (see kgram).

Given an input reference (=training) data matrix X = [ x_ij ] and a degree d > 1, the function builds a new matrix resulting from the concatenation of the d blocks [ (x_ij)^k ] ; k = 1, ..., d corresponding to the successive polynoms (without the cross-product terms).

The same is done for an eventual matrix of new data to predict (= test matrix). If there is a block scaling for the training matrix (see argument scale.blocks), the same scaling values are used for the test matrix.

Usage

inlr(Xr, Xu = NULL, degree = 2, scale.blocks = TRUE)

Arguments

Xr	A n x p matrix or data frame of reference (= training) observations.
Xu	A m x p matrix or data frame of new (= test) observations to predict.
degree	The degree used for the concatenation. Must be > 1.
scale.blocks	If TRUE (default), the blocks are internally scaled by function blockscal.

Value

A list of outputs, see the examples.

References

Berglund, A., Wold, S., 1997. INLR, implicit non-linear latent variable regression. Journal of Chemometrics 11, 141-156. https://doi.org/10.1002/(SICI)1099-128X(199703)11:2<141::AID-CEM461>3.0.CO;2-2

Examples

n <- 5
p <- 3
m <- 3
set.seed(1)
X <- matrix(rnorm(n * p, mean = 10), ncol = p)
Xu <- matrix(rnorm(m * p, mean = 10), ncol = p)
y <- rnorm(nrow(X))
set.seed(NULL)

inlr(X, degree = 2)
inlr(X, X[1:2, ], degree = 2)

res <- inlr(X, degree = 3, scale.blocks = TRUE)
res
blockscal(res$Xr, colblocks = res$colblocks)$xdisptot

res <- inlr(X, Xu, degree = 3)
fm <- plsr(res$Xr, y, res$Xu, ncomp = 2)
fm

mlesnoff/rnirs documentation built on April 24, 2023, 4:17 a.m.