eigen value of cov matrix about minimal-entropy-correlation-alignment HOT 3 CLOSED

Hi,
yes, in principle a covariance matrix should be positive definite.
However it can happen that negative eigenvalues pop out due to numerical approximations.
A you say, this can result in nan values.
For this reason it is common practice to regularize the covariance matrix as I commented in

coral_loss is suboptimal, but of course is a viable option if you are stuck with numerical issues.

Best,
P.

from minimal-entropy-correlation-alignment.

Thank you for your answer!

I will try the regularization for numerical stability!

Best,
Yoo

from minimal-entropy-correlation-alignment.

Hope it works,
let me know.

Best,
P.

from minimal-entropy-correlation-alignment.