[Coin-ipopt] Checking number of function evaluations

[Andreas Waechter]

Hi Karsten,

> in my AMPL-Code, I define a function f
>
> var f  {i in ZEIT} = ..... ;
>
> which only occurs in the constraint:
>
> s.t. minmax {i in AZEIT} : f[i] <= zeta;
>
> Because the computation of f cost a lot of time, I would like to know, how
> many times this function is being evaluated by IPOPT. Is this possible by
> AMPL or directly with IPOPT ???

I'm not sure if I understand your question correctly.  If you are
interested in finding out how often the f functions have been evaluated
after an optimization run, you can check the Ipopt statistics output and
look for the number of constraint evaluations (that tells you how often
each f[i] for i in AZEIT has been evaluated).  Does that answer your
question?

> P.S at Andreas: IPOPT scheint in der Optimale-Steuerungs-Familie immer beliebter
> zu werden, allerdings haben die Leute große Probleme an AMPL zu kommen. Hast
> du da vielleicht einen Tipp, wo man das kaufen kann (oder kennst gar eine
> Alternative).

I'm very happy to hear that Ipopt gains popularity in the optimal control
community!

As modeling language/interface, AMPL is very convenient (for problems that
are not really large, i.e. less than say a few 10,000 variables).  There
exists a free student version of AMPL that can solve problems with up to a
few hundred variables/constraints, but if you want to model larger
problems with AMPL, you would have to get a licensed copy.  All this
information is available at the AMPL home page at www.ampl.com.

One way to solve large problems modeled in AMPL without having to obtain a
copy of the AMPL interpreter is via the NEOS solvers, see for example the
Ipopt NEOS webpage at - WELL!  I just tried to look for that website, but
at the moment Ipopt is no longer in the NLP solvers list!  I will have to
check on that....   Anyway, once this is up, you can send an AMPL model to
NEOS (e.g. via a webinterface), have it solved remotely, and have the
result of the optimization be sent back to you.

Beside AMPL, I think that Prof. Larry Biegler and his group at CMU have
been working on an AIMMS and GAMS interface, but I'm not sure what the
status is...  I'm copying Larry to this message, in case he can add some
more recent information.

Best regards,

Andreas