Hello, I have noticed that with the latest CRAN version of robCompositions, the solve= 'eigen' and 'svd' options give inconsistent results when method="classical" is chosen for pcaCoDa.
Is this intentional? The results of the two solve methods can be brought closer together by changing line 114 of pcaCoDa.R (excepting the signs of the loadings and scores) to:
pcaIlr <- prcomp(xilr, scale = FALSE, center = TRUE)
Dear Andrew,
after careful revision: you are of course right. We changed the code and thank you very much. The github version includes the change, the CRAN version needs a bit to be uploaded.
Best
Matthias
Hello, I have noticed that with the latest CRAN version of robCompositions, the solve= 'eigen' and 'svd' options give inconsistent results when method="classical" is chosen for pcaCoDa.
Here is my session info:
Here's a script to illustrate the differences:
res1 and res2 are different:
Is this intentional? The results of the two solve methods can be brought closer together by changing line 114 of pcaCoDa.R (excepting the signs of the loadings and scores) to:
Thank you for reading this post and for contributing the useful R Package.
Best wishes,
Andrew