Decision Optimization

Hello,

When you say "Ideally,...", you mean that, given two solutions with the same objective value, you would prefer the one with the ideal configuration, don't you?

You used search phases for trying to express preferences. Search phases are not intended to order the solutions of the problem by their preferences. They are a hint to the solver for finding a good solution quickly, in other words, they may order the solutions by the time they are found.

The way of expressing preferences is through the objective function. For example an additional term in the objective function. Or, as you mentioned, a second criterion in a multi-objective approach. Note that the multi-objective provided in CP Optimizer is lexicographic, i.e., the criteria are not "blended", and maybe, in your "ideal" ranking, they need to be.  

So from the description of your problem, you should try to modelize your needs with the objective function instead of a search phase.

Note that the order of variables inside a scheduling phase does not have any impact on search. Phases for scheduling are there for instatiating groups of variables before others. More details are here : https://www.ibm.com/docs/en/icos/22.1.2?topic=values-search-phases-scheduling.

Hi PhR
My mistake. I meant to say that the intervals in the sequence_var are sorted in order of penalty, where they do indeed matter
My apologies for saying they were in search .... again, my apologies
Ross

Hi Olivier
I understand that objective function is intended to define shape of the solution. I have been working with that for some time and thought that using search_phase would help give solver a 'hint' as to what a preferred solution might look like. I figure that one should consider/exploit all the tools available in the toolbox. 
But so far without joy.

I figure this is a VERY common requirement in production scheduling problems.
Hopefully someone with experience in this type of problem can help

Thanks in advance
Ross