Re: [biogeme] Type of cause termination

[Bierlaire Michel]

The documentation about the parameters is here:
http://biogeme.epfl.ch/sphinx/algorithms.html#biogeme.algorithms.simpleBoundsNewtonAlgorithm

It means that the value of the parameters to not change much from one iteration to the next. The algorithm many not have converged. You need to check the value of the gradient to check convergence.


> On 13 Apr 2023, at 10:12, Damar Dewayanto <damarde...@gmail.com> wrote:
>
> Dear Prof. Bierlaire,
>
> I hope this finds you well.
>
> I was wondering if you had any idea about type of cause termination such as: relative change that appears on estimation report. Does this mean the model have converged yet?
> Have you any references of those regarding issue (type cause of termination)?
>
> Thank you for the help.
>
> --
> 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/e94d2757-7bf2-4292-80bb-a1f5d4f41971n%40googlegroups.com.

Michel Bierlaire
Transport and Mobility Laboratory
School of Architecture, Civil and Environmental Engineering
EPFL - Ecole Polytechnique Fédérale de Lausanne
http://transp-or.epfl.ch
http://people.epfl.ch/michel.bierlaire

[Bierlaire Michel]

> On 14 Apr 2023, at 08:40, damarde...@gmail.com wrote:
>
> Thank you for your responses.
>
> In regard to your answer, I have a several questions:
> 1. Does this mean the model and the estimation still acceptable, even though the cause of termination is relative change?

Probably, yes. It depends on the gradient.

> 2. How do we check the value of the gradient to check convergence? Are there any codes to do that?

It is reported in the html file.

[azis hakim]

Dear Prof. Bierlaire

Sorry for suddenly appeared in this conversation..

Actually , I also have similar question about cause of termination...
1. based on your answer,  even though the cause of termination is relative change... it could be an acceptable depending on the gradient...
So,  what kind of gradient values can be accepted in the model?

2. In the estimation report, how to check the convergence in the model ? sorry I am still learning how to read the estimation report...

3. based on the estimation report below, is it an acceptable model ? I am running the logit model....

Thank you so much for your attention...

[Bierlaire Michel]

> On 19 Apr 2023, at 11:29, azis hakim <azis....@gmail.com> wrote:
>
> Dear Prof. Bierlaire
>
> Sorry for suddenly appeared in this conversation..
>
> Actually , I also have similar question about cause of termination...
> 1. based on your answer, even though the cause of termination is relative change... it could be an acceptable depending on the gradient...

Yes.

> So, what kind of gradient values can be accepted in the model?

The relative gradient should be close to zero. How close is a subjective matter. Computer work in finite arithmetic, and reaching zero is simply not possible.
Usually, we consider that the algorithm has converged in the relative gradient is lesser than 1.0e-6.

>
> 2. In the estimation report, how to check the convergence in the model ? sorry I am still learning how to read the estimation report...

The stopping criterion is always reported. At the end of the report, there is also information about the eigenvalues. If the model is not identified, it is reported there.

>
> 3. based on the estimation report below, is it an acceptable model ? I am running the logit model....

I would say no. For logit, it should be possible to reach a better level of convergence. Check the eigenvalue information at the end of the html file.

>
> Thank you so much for your attention...

> To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/9fcf9346-ae52-457f-a79a-88f3e519a536n%40googlegroups.com.
> <estimation report.JPG>

[Bierlaire Michel]

It means that ASC_CS is completely confounded with B

> On 9 May 2023, at 05:29, azis....@gmail.com wrote:
>
> Thank you for your answer.. It is really helpful for me...
> However  I want to ask about eigenvalue at the end of the html file... what does it mean?
> The html shows following result:
>

> --
> Best Regards,
>
> Azis Hakim Sjafruddin
> Institut Teknologi Bandung