Pipeline for (s)PLS: tune, model, perf?

Hello. Maybe @aljabadi or @kimanh.lecao an help me here.
Is this post still relevant? I don’t understand why the user here is running PLS before running sPLS. Moreover, perf in my case does not contain ncomp among the attributes.

On a more general level, is the right pipeline the following: tune (to tune the hyperparameters) → splsperf to compute the various metrics? I followed all the tutorials, although some of them are pretty old, but I still cannot understand how to properly use this package. Thanks.

Hi @lorenzoF92

The tune.spls function has changed since this post and now one can perform variable selection on Y as well. But one should use the tune function for tuning all hyperparameters. perf is only meant to evaluate the model performance.

It really depends on the study. For instance, you might want to compare the models and see if variable selection improves the model.


Hope it helps,