0
$\begingroup$

I have been reading various articles/papers on PCA and some of the authors mention it as a Eigen values whereas others go by singular values. From whetever remnants of Bachelor's algebra in my memory, I believe that EVD is not the same as SVD (although they are closely related).

My question is - While doing PCA, what should I use - EVD or SVD? I was trying it on Matlab and they have implementation of both, but the PCA function in Matlab seems to be using Eigen values. It would be nice if someone can clear this.

Thanks.

$\endgroup$
  • $\begingroup$ Relevant $\endgroup$ – jkn Nov 19 '13 at 20:58
  • $\begingroup$ @jkn thanks for the link...that helps..but can't help wondering why Matlab's implementation of PCA is based on EVD if SVD is more precise...mathworks.se/help/stats/princomp.html $\endgroup$ – user295338 Nov 19 '13 at 21:03
0
$\begingroup$

SVD is a popular choice for the reasons below:

  1. It does not require input as a squarer matrix. SVD accepts any size, whereas EVD requires square matrix.
  2. SVD returns sorted singular values and vectors, whereas EVD does not. To apply PCA, you'll have to sort eigenvalues and eigenvectors before using them.
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.