prediction of new sample by Diablo with several components

[Former user]

Hello,

In the Diablo article p21, it is stated that the prediction of class membership of a new sample is made thanks to the component t(k,i) where k is the block and i is the sample, and then some majority vote or averaging is done over the blocks. However, how is the prediction done when there's more than 1 component per block ?

It looks like a similar question was asked here for PLS-DA, but was left without answer : https://bitbucket.org/klecao/package-mixomics/issues/2/website-predict-function-with-spls-da

Thanks in advance, Arnaud

PS : sorry, created the same issue twice but logging in with my name the 2nd time ! Please delete the 1st one ...

[Florian Rohart]

Hi Arnaud,

A prediction is made for a block k and all its components, independently from the other blocks: the loadings/variates of the block k are used to calculate the variates for each component. Then, if you want to have a DIABLO prediction on component ncomp, the predictions of all the blocks on component ncomp are combined (majority vote or averaging).

The way we obtain predictions for all components for a specific block k is the same as a prediction with a PLS.

Each block