Originally posted by: anahana

Hi all,

I was wondering if Solver's Evolutionary Algorithms can be used to solve, or partly solve, problems modeled using Dispatcher of Scheduler? For example, if I use Dispatcher to model a TSP and use Dispatcher to find an initial solution, can I then use EA to improve that solution?

Regards,
#CPOptimizer
#DecisionOptimization

Originally posted by: GGR

Hi

IBM ILOG Solver gives an A.P.I. for implementing iterative improvement methods (IIM) but not offers a Evolutionist Algorithm per se.

That is you can develop on top of this A.P.I. your own method like a Local Search, a Large Neighborhood search of a EA (e.g. a GA) completed by your own learning method (e.g. some tabu for LS).

The distribution give you a set of example that you can inspire from. In the solver distribution you will find a implementation of an EA algorithms (rather a GA) or local search implementations: that are the examples loaded with the Microsoft Visual Studio solution examplesiim.sln or all examples whose name began by ea or ls.

IBM ILOG Scheduler extends the IIM A.P.I. for the case of scheduling problem. In the scheduler distribution, you will find examples that implement local search and large neighborhood search in altlns and altls examples (which demonstrate local search and large neighborhood search methods).

Hope that helps
#CPOptimizer
#DecisionOptimization

Originally posted by: anahana

Thanks, I did actually go through the examples provided (which are very good by the way), but the problem is that if I'm going to solve a scheduling problem or a routing problem, I'll need to create some sort of specialized operators for GA to improve whatever initial solution I get from Scheduler or Dispatcher solvers. Which considering the complexity of some of these problems, won't be an easy task.

I guess, I'll start with a simple problem and see how it goes.

Thanks gain for the reply.
#CPOptimizer
#DecisionOptimization

Originally posted by: rdumeur

Dear anahana,

Solver's iterative improvement methods provide Evolutionary Computation building blocks which operate on integer variable arrays. You should then be able to use this IIM evolutionary API by either :
- using integer variables present in your scheduler/dispatcher model
- creating integer variables and connecting them by constraints to expressions of your scheduler/dispatcher model.

I hope this helps.
Yours sincerely,

Renaud
#CPOptimizer
#DecisionOptimization

Originally posted by: anahana

Dear Renaud, thanks for getting back to me.

That was actually my major concern, if I have to represent the model I build with Dispatcher or Scheduler with a mathematical model, in order to use the GA classes, I would loose much of the modeling capabilities of Dispatcher or Scheduler as a lot of the classes they provide have to be re-written as mathematical expressions.

I guess I'll have to start with a simple problem and see what challenges I face.

Regards,
#CPOptimizer
#DecisionOptimization