I've been using Scilab's PCA implementation:
http://www.scilab.org/product/man/index.php?module=statistics&page=pca.htm.
I tried several other PCA implementations, some of which stated they
were SVD-based, and SciLab gave me the "cleanest" results.
I wasn't sure what implementation Scilab uses, but before trusting it I
did validate it against published results generated using a different
PCA implementation (native Fortran). The Scilab results not only
replicated the prior results, but they were preferable to me due to
significantly lower levels of noise artifacts.
Well, I just looked at the Scilab PCA implementation, and I'm not enough
of a Scilab (or math) guru to really tell what it is doing! The lower
noise could be due more to the implementation of Scilab's other
statistical functions, and not due solely to the PCA implementation itself.
I suppose that makes me more like a tool-using monkey: I can find many
sticks to use as tools to get termites, and I can test them to find the
one that works best, and I can even learn the characteristics of a "good
stick", but I don't understand the plant genetics and environmental
factors that caused that stick or form to be ideal for this use.
I'm just slogging my way through the data, and when I'm done, I want it
to look good! Hence my interest in FC, another tool of great utility.
-BobC
Christopher Barker wrote:
Getting OT a bit here...
Robert Cunningham wrote:
I was experimenting with using a
PCA-based approach to extract unified data from 4 sensors: The SVD
approach outlined in the paper may be significantly faster.
If you mean Principle Component Analysis -- I've done PCA using SVD (for
an oceanographic application), so it may be the same thing. However, SVD
is a pretty efficient way to compute a PCA, which may be what you mean.
-Chris
--
Robert Cunnigham
Chief Engineer, SDTI
[EMAIL PROTECTED]
Ph: 858-332-0700 ext. 118
Fax:858-332-0709
SD Technologies, Inc.
C/O Space Micro, Inc.
10401 Roselle Street Suite 400
San Diego, CA 92121
begin:vcard
fn:Robert Cunningham
n:Cunningham;Robert
org:SD Technologies, Inc.
adr;dom:;;10401 Roselle Street, Suite 400;San Diego;CA;92121
email;internet:[EMAIL PROTECTED]
title:Chief Engineer
tel;work:858-332-0700 x118
tel;fax:858-332-0709
version:2.1
end:vcard
_______________________________________________
FloatCanvas mailing list
[email protected]
http://mail.mithis.com/cgi-bin/mailman/listinfo/floatcanvas
