Loss of eigenvectors' sign information

[RahulBhalley]

Hi! Thanks for this code.

I find that the sign information is lost when the original vector elements are recovered because square absolute operation makes every real the complex number positive.

Is there any way we can recover the original eigenvectors' signs also without using original eigenvectors.sign() (from PyTorch) * square root of eigenvectors (deduced from eigenvalues)? Because that'd be really helpful!

Regards
Rahul Bhalley