[SCIP] Minlp doesn't finish running

[Mathieu Besançon]

Dear Pedro,

I took a look at the model, it seems to be fairly badly conditioned, with
huge and very small coefficients. There is also potentially a logarithmic
transformation that should be possible for the nonlinear expressions,
passing the second part of the addition of the nonlinear constraints to the
right of the equality, applying a log on both sides (should be safe to do
since all variables seem positive.

Best,
Mathieu

On Tue, Feb 20, 2024 at 9:39 PM Pedro Matias Carrancio <
pedro.carrancio at gmail.com> wrote:

> Hello all.
>
> I am working with a MINLP. It has N variables, a linear objective function
> (actually it is the sum of N 5-parameter logistics curves but by
> substitution of variables the non-linearities were taken to constraints)
> and N+1 constraints (N to linearize the function objective and 1 specific
> to the problem)
>
> I ran several instances of the same problem ok.  Around 90 cases, with gap
> <=1e-04 and obtaining maximum response times of 15 sec. Even some of them
> have more than 100 variables.
>
> But particularly this example of 85 variables is giving me problems to
> obtain a quality response and desirable time. It executes it with a time
> limit of 360,000 seconds (100 hours) and it stops for timeout with a gap of
> 2.5%, far from the gap that we consider to be good quality.
>
>
>
> I attach a .cip file with the example and the log.
>
> In Python it can be executed with this code.
>
> ######################################
>
> *import pyscipopt as pyscip*
>
> *import os *
>
> *if os.path.exists('log_scip_5pl.log'):*
>
> *    os.remove('log_scip_5pl.log')*
>
> *model=pyscip.Model()*
>
> *model.hideOutput(quiet=True)*
>
> *model.setParam('limits/gap', 1e-04)*
>
> *model.setParam('display/verblevel', 5)*
>
> *model.setParam('limits/time', 300)*
>
> *model.setLogfile('log_scip_5pl.log')*
>
> *model.setParam('numerics/feastol', 1e-04)*
>
> *model.setParam('numerics/dualfeastol', 1e-04)*
>
> *model.readProblem('model_5pl.cip')*
>
> *model.optimize()*
>
> ######################################
>
> Can you help me understand what alternatives I have to improve this
> condition?
> Thank you in advance for your help.
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>


-- 
Mathieu Besançon


-- 
Mathieu Besançon
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20240224/9b908ff9/attachment.html>