popls: Parallel and Orthogonalised Partial Least Squares - PO-PLS

Description

This is a basic implementation of PO-PLS with manual and automatic component selections.

Usage

popls(
  X,
  Y,
  commons = 2,
  auto = TRUE,
  auto.par = list(explVarLim = 40, rLim = 0.8),
  manual.par = list(ncomp = rep(0, length(X)), ncommon = list())
)

Value

A multiblock object with block-wise, local and common loadings and scores. Relevant plotting functions: multiblock_plots

and result functions: multiblock_results.

Arguments

X: list of input blocks
Y: matrix of response variable(s)
commons: numeric giving the highest number of blocks to combine when calculating local or common scores.
auto: logical indicating if automatic choice of complexities should be used.
auto.par: named list setting limits for automatic choice of complexities. See Details.
manual.par: named list for manual choice of blocks. The list consists of ncomp which indicates the number of components to extract from each block and ncommon which is the corresponding for choosing the block combinations (local/common). For the latter, use unique_combos(n_blocks, commons) to see order of local/common blocks. Component numbers will be reduced if simpler models give better predictions. See example.

Details

PO-PLS decomposes a set of input data blocks into common, local and distinct components through a process involving pls and gca. The rLim parameter is a lower bound for the GCA correlation when building common components, while explVarLim is the minimum explained variance for common components and unique components.

References

I Måge, BH Mevik, T Næs. (2008). Regression models with process variables and parallel blocks of raw material measurements. Journal of Chemometrics: A Journal of the Chemometrics Society 22 (8), 443-456
I Måge, E Menichelli, T Næs (2012). Preference mapping by PO-PLS: Separating common and unique information in several data blocks. Food quality and preference 24 (1), 8-16

Examples

Run this code

data(potato)

# Automatic analysis
pot.po.auto <- popls(potato[1:3], potato[['Sensory']][,1])
pot.po.auto$explVar

# Manual choice of up to 5 components for each block and 1, 0, and 2 blocks,
# respectively from the (1,2), (1,3) and (2,3) combinations of blocks.
pot.po.man <- popls(potato[1:3], potato[['Sensory']][,1], auto=FALSE, 
                manual.par = list(ncomp=c(5,5,5), ncommon=c(1,0,2)))
pot.po.man$explVar

# Score plot for local (2,3) components
plot(scores(pot.po.man,3), comps=1:2, labels="names")

Run the code above in your browser using DataLab

The Learning Leader's Guide to AI Literacy