[SciPy-User] How to calculate Yulewalk with scipy.optimize.leastsq

Sturla Molden sturla@molden...
Sun Jan 22 07:05:18 CST 2012

Den 22.01.2012 13:12, skrev David Cournapeau:
> We are not talking about the same algorithm here. Because the
> correlation matrix has a very specific structure (toeplitz), it can be
> inverted in O(N^2) instead of O(N^3), this is what the Levinson Durbin
> algorithm is all about. You cannot easily implement Levinson-Durbin in
> numpy, because of its recursive nature.
> I think you can also reasonably expect talkbox author to know one
> thing or two about numpy ;)

Sure :)

But whenever I have used autoregression, it seems the expensive part is 
estimating the covariance, not inverting it.

Levinson-Durbin would belong in scipy.linalg though, it's not just for 


More information about the SciPy-User mailing list