[gecode-users] Feasibility of using Gecode for non-linear optimization problem

[jam_bdn at yahoo.es]

Dear all,
   
    I have used Gecode in the past and I am considering it now to solve an optimization problem but I am not so sure about the feasibility.
    I would like your expert advise before starting a lot of coding that may not come to any good conclusion.

    The problem has around 40 parameters(floats bounded –20% to 20% let’s say) to be optimized( minimize or maximize, depending the case). The evaluation of the function to obtain the value takes 12 sec to be executed in the available environment ( no parallelization in place )
    The problem applies the 40 parameters that, can be seen as weights, to the data and obtains a list of one thousand values one for each subject in the study.
    The list of is sorted by the value of the variable calculated.
    The objective is to minimize the value of the element in the nth position (fixed) once sorted but the parameters change which element(subject) goes to the nth position.

    In addition there are other variables that are depending on the same set of 40 parameters, and that act as constraints: i.e. the value can not go beyond or below a certain limit. But the relationship is linear or quadratic on the parameters so I know this is easy for Gecode.

     The problem is on the first part of the problem, where I don’t see how to express the constraint.
     Any hints on how to approach that problem would be appreciated.
     I am in parallel evaluating changing the problem to be solved and try to optimize instead the sum of the one thousand values what I think would simplify the problem, and make it much faster. Would that make the problem solvable with Gecode?

      The problem being feasible, the second point is about viability. Taking each computation 12 secs, the computation can be dramatically slow. Where should I look at, in the documentation, for ways to improve speed?

         Thanks in advance

             Jose Magana
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gecode.org/pipermail/users/attachments/20141116/b4177dae/attachment.html>