Empirically Deriving Design matrix for mixDiablo

Create issue
Issue #120 resolved
Former user created an issue

With respect to mixDiablo, it is suggested that the design matrix can be empirically populated by running PLS in a pairwise fashion and assigning a nonzero constant to all those pairs for which the correlation between the first PLS component is greater than some value (e.g., >.8).

Maybe this is a silly question, but does this refer to the correlation between the loadings on the first component or the variates on the first component? Furthermore, is this correlation coefficient nothing more than a simple Pearson correlation coefficient?

Best, Tom

Comments (2)

  1. Florian Rohart

    Hi Tom,

    This refers to the correlation between the variates (the coordinates of each sample in the different block of data), these are the only parameters of same dimension across the block (#variates = #samples).

    It is a simple Pearson correlation coefficient, but once the variates have been calculated for each pairwise block of data. I'm not sure how you would calculate Pearson correlation on 2 datasets with different numbers of variables otherwise?

  2. Log in to comment