[R] generalized procrustes analysis without translation
meyners.m at pg.com
Fri Dec 9 10:13:57 CET 2011
I’m not aware of any, but if you really want this, it should be possible to modify the code of any of the functions you propose and delete the part doing the translation. I’m not sure that this is a good idea, though; either your matrices are *truly* centered, then it doesn’t make a difference, or they are only centered in theory but not in practice due to random error, then the rotation and scaling will depend heavily on the deviation of the center of the matrix from the origin. Optimal scaling and rotation in GPA are independent from each other so might be (de-)selected depending on needs, but they depend on the translation, so it seems to make sense that you cannot deselect that part. You wouldn’t do scaling of two variables without centering them around 0 either, would you? You may transform them back if required, but scaling without centering first does not appeal to me, same for rotations.
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Mark Heckmann
Sent: Wednesday, December 07, 2011 19:09
To: r-help at r-project.org
Subject: [R] generalized procrustes analysis without translation
I am searching for a GPA function/algorithm that allows to omit translation of matrices.
procGPA in the shapes package does not have this option.
I need scaling, rotation and reflection but no translation as my matrices are naturally centered.
I am not sure if GPA from FactoMineR might be used for this.
There are no arguments allowing to switch on the above mentioned arguments except for scaling.
Frankly, I also do not understand how to set up the data for GPA properly, especially how to use the group argument.
Does someone know another GPA function or has some code available?
Thanks in advance
[[alternative HTML version deleted]]
R-help at r-project.org mailing list
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
More information about the R-help