hessian calculation failed

Dear Forum,

I have the following statement: Hessian calculation failed

This is the hessian [7,7]


0.0052221669 -0.00036374141 -3.3787134e-006 -9.0935352e-005 -0.00050014443 -0.00042057600 0.014322318
-0.00036374141 0.00060909852 -5.1967769e-005 0.0013197135 -0.0040606568 -0.0015227463 -0.0036545911
-3.3787134e-006 -5.1967769e-005 0.00032156226 6.7055186e-006 -4.0233112e-005 -3.1851213e-005 -1.6763797e-005
-9.0935352e-005 0.0013197135 6.7055186e-006 0.078675225 0.0015227463 -0.0012181970 -0.00050758210
-0.00050014443 -0.0040606568 -4.0233112e-005 0.0015227463 0.085273793 -0.0036545911 0.011268323
-0.00042057600 -0.0015227463 -3.1851213e-005 -0.0012181970 -0.0036545911 0.030251893 0.00091364778
0.014322318 -0.0036545911 -1.6763797e-005 -0.00050758210 0.011268323 0.00091364778 0.87446244

I have read that I can see linear dependencies and other stuff in the hessian.
Another comment that I read is that because almost all values are 0, this could be an error.

I show you now a calculated Hessian from a model running (with correct parameter values and everything fine). You can see that almost everything is also almost 0.

This hessian is [5,5].

0.0052538628 6.5190725e-009 1.1744363e-009 -5.9854253e-010 0.022445445
6.5190725e-009 0.032001262 -3.3672388e-009 3.8135282e-010 -0.013167274
1.1744363e-009 -3.3672388e-009 0.0067957326 -6.0526066e-010 -0.0020905364
-5.9854253e-010 3.8135282e-010 -6.0526066e-010 0.00032206954 6.0595038e-008
0.022445445 -0.013167274 -0.0020905364 6.0595038e-008 0.89788351

How can I get the Hessian to be calculated? 🙁

Thanks

1 Answer



0



Have you tried scaling your data? If you scale all your model variables so that their range is something like 0 to 1 or -1 to +1, this can often yield better results.

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