understanding PCA results and SVD

1 view (last 30 days)
Virginia on 5 Jun 2013
Hello, I have been recently reading a lot and trying to understand the princomp() and svd() functions from matlab. So I read the threads and recommended papers, but it seems I am not able to find the answer to my problem. I have the concept, but cannot get it numerically. So, here it goes: Let's say I have m signals (6), which are varying with time, which is discrete. So, I have n samples (20) of this m signals. This signals are somehow linear combinations of some primary signals, which is what I want to find. Therefore, for example, let's say I ant to find the 2 signals that leinarly combined will give me this 6 signals. So, I costruct the matrix A(n,m). Here comes the first problem , according to matlab or depending if is svd or pca I should introduce A(n,m) or A'(m,n). The more I read the more confused I get about how should I introduce them. Then I delete the mean value of each signal and perfomr [U S V]=svd(A,0) and [coeff,score,latent] = princomp(X); From here I get several signals, but, for what I read, from SVD, V should be the principal components, but for me, the principal components are 2 signals of 20 samples each...and what I get is a 6x6 matrix... For PCA is more or less the same. So, putting it more clear, what I am not able to retieve is something that will make me possible to do: A=loads*pc; I want to know loads and pc from svd of pca. Also know which method is better. And, final question is, if one of my origianl signals does have a mean value, whay sould I delete it? If I delete it I will never be able to recompute the original signals... I know is quite long, but I hope I was able to express myself. Thanks in advance for your help

Answers (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!