How to scale my data to avoid Hessian calculation failure?

Hi all,

I'm getting the problem "Hessian calculation failed" in using Maxlik for estimation. I have checked the R matrix as on page 8 of the paper: https://www.aptech.com/wp-content/uploads/2013/05/qnewton.pdf, and it turns out there are not zero elements in the diagonals of R, which means no linear dependency problem, right? But the Hessian diagonal elements have somewhat different magnitudes (from 0.01 to 5.5),  so I believe it is likely to be a scaling problem. Besides, there are also a couple of zeros in the last few columns of Hessian.

I've been reading the forum but no one talks about how to properly scale the data. Should I make all variables be at the same magnitude for the variances and means? Or is experimenting the only way to get it right? If so, is there recommended steps for doing experiments? 

Any advice is highly appreciated!

 

Regards,

Ziwei

1 Answer



0



Generally things go best with the variables scaled to be between 0 and 1 or -1 and +1. The GAUSS rescale function provides some scaling options you can use.

aptech

1,423

Your Answer


You must login to post answers.

Have a Specific Question?

Get a real answer from a real person

Need Support?

Get help from our friendly experts.

Try GAUSS for 30 days for FREE

See what GAUSS can do for your data

© Aptech Systems, Inc. All rights reserved.

Privacy Policy