Originally posted by: mremn

Hi everybody,

While searching in the constraint programming, I need to prune a node if it does not satisfy some conditions. I'm not sure I should use a gold or "writing a constraint". In the latter case, I can use modifiers to tell the solver that the current solution is infeasible  (e.g. x.setValue(-1)  for a non-negative variable x). But I'm not sure this makes x equal to -1 for other nodes. 

Thank you for your help,

Ha Minh Hoang

Originally posted by: GGR

Hi

You actually seem needing a custom constraint. Beware a constraint does is not a global cut. It is local to the node you apply the pruning. If a constraint hold on a variable x domain, the pruning algorithm (the propagate member of an instance of IlcConstraintI) is executed to reduve the actual domain of the involved constraint. The domain reduction is undid when the solver backtrack through the node.

Hope that helps

Originally posted by: mremn

Thank you GGR for your answer,

So I really need a custom constraint to prune a node. However, I need to know the objective value of the current best integer solution so far. How could I retrieve this information while writing a constraint?

Originally posted by: GGR

Hi

You should have a look to the function IloCP::next(). After each succcessful iteration, the solver is in a state that is the best so far incumbent.

IloEnv env();
IloModel model(env);
/*..*/ //fill the model
IloExpr obj = /*..*/; the objective expression
IloGoal goal = /*..*/; // if you do not use automatic search
IloConstraint ct = /*..*/; the user constraint
IloMinimize(model, obj);
IloCP cp(model);
cp.startNewSearch(goal); // ignore goal argument in automatic search
while(cp.next()) {
    IloNum bestSoFar = cp.getObjValue();
    /*..*/ store bestSofar in your data structure
}

Hope that helps