Skip to contents

Select the optimal ncomp from sPLS cross-validation results

Source: R/spls.R
spls_get_optim_ncomp.Rd

Select the optimal number of components to compute in a sPLS run from the cross-validation results obtained with mixOmics::perf() on a PLS or sPLS result, using the the mean Q2.total values. Note that this function is experimental, the corresponding diagnostic plots should be considered when selecting the optimal number of components to use.

Usage

spls_get_optim_ncomp(spls_perf, thr = 0.0975, min_ncomp = 1)

Arguments

spls_perf

List, the result of mixOmics::perf().

thr

Numeric, the threshold to be used for the Q2 values. Default value is 0.0975.

min_ncomp

Integer, the minimum ncomp value to be returned. Default value is 1, i.e. this argument does not play a role in selecting the comp value. Can be useful if we want at least 2 latent components for final plots.

Value

An integer, the optimal number of components to use for the sPLS run.

Details

The selection is made as follows:

  • If all Q2 values are below the threshold specified with thr, the number of components yielding the highest Q2 value is selected.

  • If all Q2 values are above the threshold, the number of components yielding the lowest Q2 value is selected.

  • If the Q2 values are increasing, the number of components n is selected such that n+1 is the smallest number of components with a Q2 value above the threshold.

  • If the Q2 values are decreasing, the number of components n is selected such that n+1 is the smallest number of components with a Q2 value below the threshold.