Why Matlab forces a sign convention while calculating the PCA coefficients?

Hi!

I am applying PCA on my data (Image sequences) but got different results when I used the PCA (Matlab Toolbox) command and doing it manually makind matrix and using SVD command.

After doing some search I found that matlab enforces some some sign conventions in the PCA code on coefficients calculation, Why is that?

Answer 1

[John D'Errico]

The eigenvectors computed are entirely arbitrary. You can arbitrarily change the sign and they will be as valid either way. Different schemes to compute the vectors can produce different signs, and in some cases, the sign might even be random.

The sign has ABSOLUTELY NO impact on the result, except for that factor of -1 or not.

So why are you worried about the sign? It is irrelevant.