Savgolcv
Jump to navigation
Jump to search
Purpose
Cross-validation for Savitzky-Golay smoothing and differentiation.
Synopsis
- cumpress = savgolcv(x,y,lv,width,order,deriv,ind,rm,cvi,pre); %for x class "double"
- cumpress = savgolcv(x,y,lv,width,order,deriv,[],rm,cvi,pre); %for x class "dataset"
Description
SAVGOLCV performs cross-validation of Savitzky-Golay parameters: filter width, polynomial order, and derviative order.
Inputs
- x = M by N matrix of predictor variables with ROW vectors to be smoothed (e.g. spectra), and
- y = M by P matrix of predicted variables.
Optional Inputs
- ind = indices of columns of x to be used for calibration {default ind = [1:n] i.e. all x columns}.
The following are optional Savitzky-Golay parameters (calls SAVGOL). By entering a vector, instead of a scalar, these variables are cross-validated.
- width = number of points in filter {default width = [11 17 23]}.
- order = polynomial order {default order = [2 3]}.
- deriv = derivative order {default deriv = [0 1 2]}.
The following are optional cross-validation parameters (calls CROSSVAL).
- lv = maximum number of LVs {default lv = min(size(x))}.
- rm = regression method. Options are: rm = 'nip', PLS via NIPALS algorithm; rm = 'sim', PLS via SIMPLS algorithm {default}, and rm = 'pcr', uses PCR.
- cvi = cross-validation method. Options are: cvi = 'loo', leave-one-out, cvi = 'vet', venetian blinds {default}, cvi = 'con', contiguous blocks, and cvi = 'rnd', repeated random test sets.
- split = number of subsets to split the data into {default = 5} and is required for cvi = 'vet', 'con', or 'rnd'.
- iter = number of iterations {default = 5} and is required for cvi = 'rnd'.
- mc = 0 supresses mean centering of subsets {default mc = 1}.
Outputs
The output is a 4 dimensional array of cumulative Predictive Residual Error Sum of Squares (PRESS) value with each dimension corresponding to one of the directions cross-validated over.
- cumpress(i,:,:,:) = derivative dimension,
- cumpress(:,j,:,:) = latent variable dimension,
- cumpress(:,:,k,:) = window width dimension, and
- cumpress(:,:,:,l) = polynomial order dimension.
See Also
baseline, crossval, lamsel, mscorr, savgol, specedit, stdfir