GSVD as opposed to SVD -- How to apply this?
in [Matlab]
|
Prev: Aero Animation movable object
Next: image steganography
From: Pat Finder on 5 Jan 2010 09:46 I'm just learning about SVD (as opposed to PCA) for clustering and pattern recognition. A Professor recommended I use GSVD instead of svd when I have more than one cluster center. While I read the help on gsvd, but it is not particularly illuminating on how to apply gsvd. Not wanting to embarrass myself in front of the Professor, I am opting to embarrass myself in front of thousands of Matlab newsgroup readers worldwide.(grin) For good reason -- the Prof is at another University, so I cannot easily ask him, etc... Help gsvd says... "[U,V,X,C,S] = gsvd(A,B) returns unitary matrices U and V, a (usually) square matrix X, and non-negative diagonal matrices C and S so that A = U*C*X' B = V*S*X' C'*C + S'*S = I"... THE QUESTIONS THE DOC FILE DOES NOT HELP WITH: How do I get from "unitary matrices" (I know what they are) to math that helps me classify new data points? How does this find the cluster centers? Okay, so suppose I have two classes of data, with (x,y) coordinates. I put one class in A and one class in B. I run gsvd, then 1. How do I interpret the results? 2. How do I use the results to classify new data points? 3. Does anyone have a working example they can share? Thank you all for your kind help and understanding!
From: Rune Allnor on 5 Jan 2010 10:04 On 5 Jan, 15:46, "Pat Finder" <pfin...(a)netacc.net> wrote: > I'm just learning about SVD (as opposed to PCA) for clustering and pattern recognition. > A Professor recommended I use GSVD instead of svd when I have more than one cluster center. While I read the help on gsvd, but it is not particularly illuminating on how to apply gsvd. > > Not wanting to embarrass myself in front of the Professor, I am opting to embarrass myself in front of thousands of Matlab newsgroup readers worldwide.(grin) For good reason -- the Prof is at another University, so I cannot easily ask him, etc... I suppose emails work just as well at your prof's uni as everywhere else...? Just write him an email and ask. A professor is nothing more than a human being - presumably as yourself - and thus might be sloppy, imprecise or plain wrong. You have looked up the docs for GSVD and found no clues as to why it wpuld be better than the plain old SVD. Tell him as much. That way you 1) Indicate that you have mad an effort to follow his advice 2) Found no clues as to why his advice would be better That way you show initiative (which reflects good on you) while at the same time you need his advice (which reflects good on him). A win-win situation. Rune
From: Pat Finder on 5 Jan 2010 11:22 This Professor receives literally hundreds of e-mails a day. He ignores 99% of them, starting with the ones that come from other universities. (i.e. mine). Follow up question: I've found 100's of papers on the net related to gsvd, but none with example code. Can anyone recommend ONE paper with an example showing how to use GSVD for classification (dimensional reduction) in Matlab?
From: Rune Allnor on 5 Jan 2010 11:38 On 5 Jan, 17:22, "Pat Finder" <pfin...(a)netacc.net> wrote: > This Professor receives literally hundreds of e-mails a day. Then call him on the phone. > He ignores 99% of them, starting with the ones that come from other universities. (i.e. mine). If he ignores the phone call - find a new professor. Rune
From: Doug Schwarz on 5 Jan 2010 11:52
Pat Finder wrote: > This Professor receives literally hundreds of e-mails a day. He > ignores 99% of them, starting with the ones that come from other > universities. (i.e. mine). > > Follow up question: > I've found 100's of papers on the net related to gsvd, but none with > example code. > Can anyone recommend ONE paper with an example showing how to use GSVD > for classification (dimensional reduction) in Matlab? Hi "Pat", I don't know anything about GSVD, but here's a link to the Matlab Toolbox for Dimensionality Reduction. Perhaps GSVD is in there somewhere. <http://ict.ewi.tudelft.nl/~lvandermaaten/Matlab_Toolbox_for_Dimensionality_Reduction.html> -- Doug Schwarz dmschwarz&ieee,org Make obvious changes to get real email address. |