vector fit crashes when npoles >= npoints/2

[PinkMushroom]

vector fit crashes when npoles >= npoints/2

[Vinc0110]

Thanks for reporting this. Of course, this case should be caught and the code should not crash.

I guess the reason nobody has run into this is that the setup does not make much sense. Each complex conjugate pole introduces 2 unknowns to solve for. So, for 2 * n_poles_cmplx > npoints, the system is underdetermined.

[PinkMushroom]

yes, but the samples are complex, so it should all work out.

Intuitively shouldn't the order of my polynomial be valid up to the number of samples points ?

If were to use something like Levy/S-K (the original Levy/S-K) I'm pretty sure you could compute a polynomial with an order equal to the number of sample points.

I don't have access to my Levy code right now, but I'll give that a try when I do.

Unless you convince me I'm wrong. and then I'll still give it a try ;-)

edit:
aha. ok, my brain is definitely in the slow lane today. if the polynomial is degree n then it only generates n/2 poles which occur as complex conjugates.
the good news is this is an easy fix. Just tell the user "don't do that !" :-)

[Vinc0110]

Intuitively shouldn't the order of my polynomial be valid up to the number of samples points ?

Your intuition might the right. It also does not work if I add multiple frequency responses with 2 * n_poles_cmplx samples each. Then, the system is definitely not underdetermined anymore.

This seems to be a real bug in the code. I'll check it in the next days.