Get @ProblemFactCollectionProperty data in ConstraintProvider

[jgazegers]

I am currently working on a scheduling tool and one of the constraints I am trying to implement is to penalize each @PlanningEntity based on how many days away it is from the first possible date in the @ValueRangeProvider on the @PlanningSolution.

Example @PlanningSolution

@PlanningSolution
public class Schedule {
    @ValueRangeProvider
    @ProblemFactCollectionProperty
    private Collection<LocalDate> dates;

    @PlanningEntityCollectionProperty
    private Collection<TrainingAssignment> trainingAssignments;
}

Example Constraint

    private Constraint scheduleTrainingAsEarlyAsPossible(ConstraintFactory constraintFactory) {
        LocalDate date = LocalDate.now(); // This should be the earliest date possible

        return constraintFactory
                .forEach(TrainingAssignment.class)
                .filter(ta -> ta.getDate() != null)
                .penalize(
                        HardSoftScore.ONE_HARD,
                        (assignment) -> (int) ChronoUnit.DAYS.between(assignment.getDate(), date))
                .asConstraint("Schedule training as early as possible");
    }

[triceo]

Hello @jgazegers!
What specifically is the issue here? Problem facts are automatically available in Constraint Streams already.

[jgazegers]

I am gonna close this discussion because I believe I have found an easy solution, which is just to pass the required objects into the ConstraintProvider constructor when starting a job.

Thank you for the fast response anyways :)

[triceo]

I am gonna close this discussion because I believe I have found an easy solution, which is just to pass the required objects into the ConstraintProvider constructor when starting a job.

Please do not do this. Constraint providers are meant to be stateless, and you're exposing yourself to all kinds of potential trouble this way.

Of course I don't want this value to be retrieved on every single iteration of the constraint, as the value does not actually change (definition of a problem fact).

It wouldn't be retrieved on every single iteration - that's the benefits of incrementality. Write a constraint which works with this data, and as long as the data doesn't change, the constraint will not be reevaluated.

[jgazegers]

Please do not do this. Constraint providers are meant to be stateless, and you're exposing yourself to all kinds of potential trouble this way.

Definitely not doing this then.

It wouldn't be retrieved on every single iteration - that's the benefits of incrementality. Write a constraint which works with this data, and as long as the data doesn't change, the constraint will not be reevaluated.

Oh that's actually beautiful. Did not know it worked this way. Any chance you perhaps could point me to the place in the docs where I could read more about this?

I am however still having trouble understanding how I can get the problem fact into the stream. Can I just join on LocalDate.class or something?

For example:

    public Constraint foo(ConstraintFactory constraintFactory) {
        return constraintFactory
                .forEach(TrainingAssignment.class)
                .join(LocalDate.class)
                .penalize(HardSoftScore.ONE_HARD)
                .asConstraint("foo");
    }

[Christopher-Chianelli]

You can join on LocalDate.class if it is exposed as a ProblemFactProperty or a ProblemFactCollectionProperty in your PlanningSolution. That being said, I would put this in a ScheduleParameterization class that contains various facts about the solution:

public record ScheduleParameterization(LocalDate firstAvaiableDate, ...) {}

@PlanningSolution
public class Schedule {
    @ProblemFactProperty
    ScheduleParameterization parameterization;
    // ...
}

public Constraint foo(ConstraintFactory constraintFactory) {
     return constraintFactory
         .forEach(TrainingAssignment.class)
         .join(ScheduleParameterization.class)
         .filter((ta, p) -> !ta.getDate().equals(p.firstAvailableDate()))
         .penalize(HardSoftScore.ONE_SOFT, (ta, p) ->  (int) ChronoUnit.DAYS.between(ta.getDate(), p.firstAvailableDate()))
         .asConstraint("Schedule as early as possible");
}

[triceo]

Oh that's actually beautiful. Did not know it worked this way. Any chance you perhaps could point me to the place in the docs where I could read more about this?

In general, here.

The actual implementation under the hood of Constraint Streams is a relatively complex algorithm inspired by RETE.