Originally posted by: SystemAdmin

[Sylvain said:]

I read in the documentation that there exists a "ILOG Hybrid Cooperating Optimizers library".
Where can I find more information about this library ?
#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Philippe Refalo said:]

Hi Sylvain,

Hybrid optimizer was a library that permitted to use ILOG CPLEX and Ilog Solver
together. CPLEX was handled as a global constraint of Solver.

The last time we released this library was for Solver 6.0 and CPLEX 9.0.
We no longer deliver it so any reference to it in the doumentation
of ILOG CP 1.x is a mistake.

Philippe

#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Sylvain said:]

Thanks for the answer.

What is the recommended way of doing hybrid optimization with Cplex 10/11 and CP 1.x ?
From Cplex, using CP inside callbacks or goals ?
or from CP, using Cplex inside goals ?
#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Sylvain said:]

To follow-up on the subject, I would like to know what is the best way to solve linear relaxation at each node when using Ilog Solver.
Should I write a specific constraint to handle the linear system (knowing that I would like to add cuts to the linear system) ?
Or should I use a IloGoal ?
#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Philippe Refalo said:]

Sylvain,

The way I would recommend in your case is to begin with a CP application
of you problem and then to modify it the following way.

First create a CPLEX instance that you fill with the initial model (a relaxation
for instance). Then, in the search goal you can add constraints and you need
to call CPLEX optimization. The result is used to fail or succeed. The important
feature to code is the removal of localy valid constraints. For this you need
to set up a reversible action when adding the constraints that need to be
removed at backtracking.

Be aware that this requires quite a good knowledge of both CP and CPLEX.

Philippe



#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Sylvain said:]

Thanks again.

Now I have another question.
Is it possible to perform a "branch & bound" type search using a node selection rule based on the objective function estimation ?

Sylvain
#CPOptimizer
#DecisionOptimization

Originally posted by: SystemAdmin

[Philippe Refalo said:]

Sylvain,

You can select the next node to be searched by using IlcApply and
IlcNodeSelector. Have a look at the reference manual for more details
on these functions.

Used in conjunction with linear relaxations, this approach can be difficult
to master and requires a detailed knowledge of CPLEX. The main reason is
that to be efficient you will probably have to restart from an advanced basis
to add and remove constraints. Therefore, CPLEX will not be in the exact
same state when jumping back to a node while Solver will.

Philippe


#CPOptimizer
#DecisionOptimization