Hi!
I’d like to tune my mulitblock spls model using cross validation, however, I am uncertain what evaluation metric to choose? Do you have any recommendations? I could unfortunately not find anything in this regard on the website/ the code.
Additionally, I was wondering whether you have published the generalised sPLS approach? On the website you are only referencing the generalised CCA algorithm by Tenenhaus.
Thank you!
Best,
Clara
hi @clarasophie.v,
Regarding the metric, I assume you mean the prediction distances?
You can have a look at this mixOmics: An R package for ‘omics feature selection and multiple data integration and the supplemental material that describes those metrics.
You can also have a look at this page: Distance Metrics | mixOmics
and also the tutorial for sPLS-DA on how to use / make those decisions: sPLSDA SRBCT Case Study() | mixOmics
Regarding the generalised PLS, we have block.pls and block.spls, but maybe you are after wrapper.sgcca (the name from Tenenhaus is a bit misleading, it actually does call a PLS).
Hopefully this graphic helps on which method is appropriate for your data
Kim-Anh
