Numerical problem with the second derivative matrix

[Yuanjie Tu]

Hi Professor Bielaire,

Thanks for your help last time! 

My model utility contains a logsum:

I printed the messages of every iteration of the estimation process, and found the model met a numerical problem with the second derivative matrix (shown as below) multiple times, and then when the result comes out, it looks really weird. Besides, the model termination reason is that the "trust region is too small". Could you help me with it? Thanks so much!

 

Best,

Yuanjie Tu

[Bierlaire Michel]

Check the bounds on your parameters. Lambda1 appears at the denominator. So if it gets close to zero, the calculation will fail. 

Your specification looks to me like a nested logit. Why don’t you specify it as such.

 

On 22 Apr 2021, at 03:31, Yuanjie Tu <yua...@uw.edu> wrote:

Hi Professor Bielaire,

Thanks for your help last time! 

My model utility contains a logsum:

<Logsum.png>

I printed the messages of every iteration of the estimation process, and found the model met a numerical problem with the second derivative matrix (shown as below) multiple times, and then when the result comes out, it looks really weird. Besides, the model termination reason is that the "trust region is too small". Could you help me with it? Thanks so much!

 

<Warning.png>

<Small trust region.png>

Best,

Yuanjie Tu

--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/fd84486a-e5a8-4f09-97c0-607b3b43f122n%40googlegroups.com.
<Warning.png><Logsum.png><Small trust region.png>

[Yuanjie Tu]

Thank you! It works.