On 07/29/2013 01:35 PM, John Hayes wrote:
I have noticed that when I changed the order of the data groups the calculated eigenvalues change when the first group on the list changes. The changes are very small and not significant, but based on my (very limited) knowledge of PCA, it would seem that the order of the data would not matter.
John, I agree that order should not matter in terms of the useful output of a PCA. However, it seems plausible that the order of input vectors could result in small numerical differences in the eigenvectors. It even seems plausible that input vectors in a different order could result in eigenvectors that are rotated by some non-negligible angle. In truth, I have neither thought about this nor investigated it before. I'll put it on my to do list, but I suspect that the bottom line is one of the two things I suggested. BTW, the specific problem you reported will certainly be fixed in the next release. I got that working this morning. B -- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Science Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 Homepage: http://xafs.org/BruceRavel Software: https://github.com/bruceravel