Use of maximize with time_out

Question

Sicstus 4.3.0 has a new elegant feature to enumerate all solutions in an optimization problem, minimize(:Goal,?X,+Options).

When testing out a new search heuristics it is often not possible to prove a global optimum due to large search space which means that the search takes a long time.

But often (especially for industrial problems) it is enough to find a 'good enough' solution.

So to the problem:

I have a problem where I find a 'good enough' solution pretty fast, but I'm not able to prove global optimum. So now I want to use maximize to see how my search heuristics behaves during the search, but not waiting for global optimum.

With the new 4.3.0 version I can do like this:

findall(L,     
(
   minimize(labeling([],VarsList), Es, [all]),
   L = Es
),
Ls)

I will then end up with a list Ls containing all solution found on the trace for the global optimum.

But I don't want to wait for a global optimum so I add a time_out like this:

findall(L,
(
  minimize(labeling([time_out(MaxLabelingTime,LabelingResult)],VarsList), Es, [all]),
  L = [Es,LabelingResult]
),
Ls),

This should find all solutions, but not spending more time then MaxLabelingTime between each backtrack. But this way will fail when the time_out kicks in since Es then will be uninstantiated:

Instantiation error in argument 2 of user:minimize/2
! goal:  minimize(user:labeling([time_out(500,_201)],[_203,_207....]),_179)

My solution / workaround is as follows:

findall(L,
(
   minimize(my_labeling(MaxLabelingTime,LabelingResult,VarsList,Es), Es, [all]),
   L = [Es,LabelingResult]
),
Ls)

my_labeling(MaxLabelingTime, LabelingResult, VarsList,Es):-
   labeling([time_out(MaxLabelingTime,LabelingResult)], VarsList),
   (  LabelingResult=time_out
   -> Es=0
   ;  true
   ).

This seams to work as expected, but I'm unsure if this is a good approach?

Use of maximize with time_out

0 Answers0