Originally posted by: SystemAdmin

[shadi said:]

hello all,
I have a column generation problem,
in the slave model i find all the duties for each trip...

somtimes my cp model slave cant find any column for a specific trip so the master model cant find solution for my problem.
how i can know which constraint i have to modify or to ignore in the slave model (cp model) in order to get at least one column for  a specific trip

Thank you
Kshieboun Shadi
#ConstraintProgramming-General
#DecisionOptimization

Originally posted by: SystemAdmin

[Philippe Refalo said:]

Shadi,

The general rule when a model does not have a solution is to try to remove some constraints that you suspect to be too strong and to try solve the resulting problem. This way you can identify the hardest parts of the problem.

Without any information about the structure of the subproblem it is difficult to be more precise.

If this happens in the process of finding a first solution to start column generation with, you can consider solving the problem by ignoring the dual values and just minimizing the reduced cost.

Philippe Refalo
prefalo@ilog.fr


#ConstraintProgramming-General
#DecisionOptimization

Originally posted by: SystemAdmin

[shadi said:]

Dear Refalo,

i understand what do you mean,but i want ask if there is a way to know which constraints i have to remove and to be sure that i will get a solution  (not to remove all my constraints :) becuase then sure i will get a solution).

thank you
Ksheiboun Shadi
#ConstraintProgramming-General
#DecisionOptimization

Originally posted by: SystemAdmin

[Philippe Refalo said:]

Shadi,

There is no functionality in CP Optimizer to get the set of conflicting constraint when no solution exists. You have to discover this yourself on the infeasible models you found by adding/removing constraints. You will determine the hard parts of your model this way.

If you want to relax these hard parts, you can relax these constraints by creating meta-constraints
and add them (by way of a variable if needed) to the objective function.

For instance you can relax x + y <= 7 by minimizing 1000 * (x+y <=7) or by minimizing 100 * z with the additional constraint x + y <= 7 + z and z >= 0.

Philippe
#ConstraintProgramming-General
#DecisionOptimization

Originally posted by: SystemAdmin

[faclon said:]

I think CP only help us build the Results and filter the lawless results by your constraints that you build in CP. So, If the constraints are too strong, it must be stength the lawless results, maybe, in the results pools there is nothing.

So, adjust the parameters of your constraints, or disable some filter rules, maybe,will get some results. Of course,the results is not mapping your exceptation. but, you can debug the parameters or filter rules.
#ConstraintProgramming-General
#DecisionOptimization