[Coin-ipopt] Results from derivative test

Andreas Waechter andreasw at watson.ibm.com
Mon Jan 7 13:29:29 EST 2008


Hi Fabian,

Thanks for the clarifications.

As you suggested, we could just add just a virtual method, probably in the 
TNLP, maybe that would be the easiest place to add it.  I could do that. 
Is that what you need?

Thanks

Andreas

On Mon, 7 Jan 2008, Fabian Wein wrote:

> I sent this answer by accident not to the list
>
> Hi Andreas,
>
>>> I'm not sure I understand what you mean by visualizing the relative
>>> error. Do you mean generating some sort of graphic which helps you to
>>> find which particular partial derivatives have how much error?
>
> Exactly. I optimize a Finite Element simulation with one design
> variable for each finite element (quadrilaterals, hexahedrons, ...)
> and hence color-mapped visualization shall give a good insight of what
> is going on.
>
>>>
>>> In either case, the code for the derivative checker is in
>>> Ipopt/src/Interfaces/IpTNLPAdapter.cpp
>>> in the method
>>> TNLPAdapter::CheckDerivatives
>
> Thanks for the hint!
>>>
>>> We could set things up so that another method is called when a new Ipopt
>>> option would be set, and of course it would be nice to include such a
>>> new method in the Ipopt distribution.
>
> You mean a virtual method which gives back the finite difference
> gradients,
> possibly again the user provided gradients and the iteration number?
>
> One could skip an additional feature, the virtual method is
> implemented or not
> (IMHO).
>
> Well, coding is easy - it's for me more the working through the class
> structure.
> What do you mean with "We could set things up..."? Do you mean to assist?
>
> Fabian
>
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-ipopt
>



More information about the Coin-ipopt mailing list