multi-start module

[John]

Dear Guido,

I didn't find the article wrote by you about the multi-start. Could
you please tell me where I can find it or give me a link about it?

Another question is that when I call the multi-start program, the
color of the generated mathematical program is red and the error says
"PhA has not been declared." However, I've declared this identifier as
a Mp in
the program. The two procedures used to call the multi-start program
are listed below:

gmpa := GMP::Instance::Generate(PhA);
MulStart::DoMultiStart(gmpa,10,5);

Thank you.
Best regards,
John

[Marcel Hunting]

Hi John,

 

Attached is a small example that demonstrates the multi-start algorithm. You should create a new AIMMS project and load this .aim file. If you do a normal solve (run MainExecution) then the solution found has an objective of 5.64 (minimization problem; I used CONOPT 3.14V) and if you use multi-start (run UseMultiStart) then two other solutions are found with an objective of 5.33 and 4.98; the latter being the global optimum and returned by the multi-start algorithm.

 

The multi-start algorithm can be controlled by several parameters as listed in the MultiStart Control Declarations section of the multi-start module. The most important ones are IterationLimit which controls the number of iterations used by the multi-start algorithm and UsePresolver which controls whether the multi-start algorithm should start by preprocessing the nonlinear model (switched off by default).

 

More information about the multi-start algorithm can be found in chapter 17.2 in the Language Reference (AIMMS 3.12).

 

About the error, did you by chance declare math program PhA inside a module?

 

Best regards,

 

Marcel Hunting
AIMMS Software Developer

[John]

Dear Marcel,

Thank you for the model you gave. It is a very good example to start
with. I will read the literature review about multi-start algorithm.

As for the error, I didn't notice a very small typing error. Now, I've
corrected it.

Thank you very much for your help. One more question, do you know
where I can find other local optima results after using multi-start
algorithm?

Thank you very much.

Best regards,
John

>  globopt_m28.aim
> 3KViewDownload

[Marcel Hunting]

Hi John,

 

If you want the multi-start algorithm to save more solutions then you have to set, e.g.,

 

   MulStart::NumberOfBestSolutions := 5;

 

before calling the MulStart::DoMultiStart procedure. The algorithm will then store the 5 best solutions in the so-called solution repository of the GMP. To send a solution from the solution repository to the model identifiers you have to use the procedure GMP::Solution::SendToModel. See the attached example.

[John]

Dear Marcel,

Thank you so much for your model. I think it's really a good idea to
illustrate one concept by starting with an easy example. It's very
helpful and my problem is solved.

I want to say Thanks a lot for your warm help.

I have a following question about the multistart algorithm. I wrote
one problem with AIMMS, in which several nonlinear MPs are needed to
solve (independently). They are solved one by one or simultaneously
and then the results are passed to another nonlinear MP to do the
final solve.

My question is that since they are all nonlinear MPs, to ensure the
accuracy of results, I will use multistart algorithm to each of MPs.
However, this may take a little bit longer time. I know there are
parallel /distributed solver sessions that can be used to shorten this
solving time. Could you please give me some guidance to this area? If
there is any example available, that'll be the best.

Appreciate it for your time and help. Have a good day.

Best regards,
John

>  get_best_sol1.aim
> 4KViewDownload

[Guido Diepen]

Hi John,

the article is not yet available on the AIMMS blog (http://blog.aimms.com). It is still one of the draft articles there, which I will try to make available as soon as possible. It will contain the same / similar example as the one provided by Marcel.

Guido Diepen

AIMMS Specialist