Block sPLSDA error (DIABLO) and multi block sPLS


I would like to know how I get the overall error rate of my tuned block.splsda model.
When I do:
MyPerf.diablo ← perf(tuned.diablo, validation = ‘Mfold’, folds = 5,
nrepeat = 10,
dist = ‘centroids.dist’)

And then look at MyPerf.diablo$WeightedVote.error.rate

I get an error rate per component. In the paper you talk about the overall error rate across all components, is that an average of the weighted vote error rate per component? Or do you take the last component?

My other question is around tuning the number of keepX in the block sparsePLS (continuous outcome). I’ve seen forum post saying that you don’t yet have a tuning function available for this, but the posts date from 2 years ago so I was wondering if there was any updates on that.


Best wishes,

Would you mind linking what paper you’re referring to? In mixOmics, overall error rate (ER) refers to the basic form of ER, such that class proportions are not used. Balanced Error Rate (BER) uses class proportions to weight each classes ER. Averaging error rates across all components doesn’t really make sense to do unless in specific circumstances.

In regards to a block sPLS tuning function, unfortunately it hasn’t been developed. We’ll make sure to let you know when it is