cv.wpls: Cross-validation for segmented spectral regions of the...

In windows.pls: Segmentation Approaches in Chemometrics

Cross-validation for segmented spectral regions of the original spectra.

Description

Computes and stores cross-validation metrics for one of the three possible modes ‘wpls’, ‘epls’, ‘swpls’.

Usage

cv.wpls(
  xblock = NULL,
  yblock = NULL,
  windows = 3,
  window.size = 30,
  increment = 10,
  cv = 10,
  scale = FALSE,
  ncp = 10,
  mode = "wpls"
)

Arguments

xblock	A matrix containing one spectra for each observation.
yblock	A vector containing the concentration associated to each spectra in the xblock matrix.
windows	Parameter used when either ‘wpls’ or ‘ewpls’ is chosen. Points out how many windows the user wants to divide the spectra in.
window.size	Parameter used when ‘swpls’ is chosen. Indicates the width of the window that slides along the spectra.
increment	Parameter used when ‘swpls’ is chosen. Indicates how many steps the window slides forward.
cv	Number of segments used for cross-validation.
scale	logical, asks to perform standardization.
ncp	Maximum number of principal components to be computed for each model.
mode	'wpls','ewpls' or 'swpls', see Details for more.

Details

NIR and Vis-NIR technologies are used to obtain spectra which might contain helpful information about the content of the samples the user is investigating. Since this method has been combined with multivariate statistical methods, researchers have been questioning the importance of using spectra in its entirety or if it might be a better solution to divide it in smaller regions which can guarantee higher performance in terms of predictions. Several methods have been proposed, from selecting only some regions to selecting combinations of those which are performing the best. This function provides three possibilities:

1. ‘wpls’, which stands for Window PLS, divides the original spectra into several windows, computes PLS and stores metrics of interest such as RMSE and R2 for calibration and cross-validation both.

2. ‘ewpls’, which stands for Evolving Window PLS, divides the original spectra into several windows, but each new window incorporates the previous ones, so that we are comparing smaller windows with the entire spectra.

3. ’swpls’, which stands for Sliding Window Window PLS, ,asks the width of the window that will be used to compute the model and the step that the window will make forward in the spectra so that a new model is calculated. In this way the window slides along spectra and computes several models, which will be compared with metrics.

This function proposes a simpler version of iPLS, that can be found in the mdatools package, which divides the spectra in smaller segments and tries to find the combination with the lowest RMSE in cross-validation.

Value

Returns a list containing:

xblock	Matrix containing spectra used to train the model.
yblock	Vector containing values of the dependent variable.
cal	List containing RMSE and R2 of calibratrion.
cv	List containing RMSE and R2 of cross-validation.
ncp	Number of components used to compute the model.
scale	Contains logical condition used for standardization.
cv.segment	Number of segments used for cross-validation.

References

1. Chen, J., Yin, Z., Tang, Y. et al. Vis-NIR spectroscopy with moving-window PLS method applied to rapid analysis of whole blood viscosity. Anal Bioanal Chem 409, 2737–2745 (2017).

2. Y.P. Du, Y.Z. Liang, J.H. Jiang, R.J. Berry, Y. Ozaki, Spectral regions selection to improve prediction ability of PLS models by changeable size moving window partial least squares and searching combination moving window partial least squares, Analytica Chimica Acta, Volume 501, Issue 2, 2004, Pages 183-191,

3. mdatools package, https://github.com/svkucheryavski/mdatools

Examples

data(beer)
conc=beer[,1]
sp=beer[,2:ncol(beer)]
names(sp)=convert.names.wl(1100,2250,2)
conc=unlist(conc)
mywpls=cv.wpls(sp, conc,mode='wpls', windows = 5)

windows.pls documentation built on Aug. 10, 2023, 1:07 a.m.