Python solver: support JSON of LoadBalance justification info

[ge0ffrey]

To reproduce:

• run score analysis on a load balancing use case and serialize to JSON (through a rest endpoint)
• with python

Error:

Unable to serialize unknown type: <java class 'ai.timefold.solver.core.impl.score.stream.collector.LoadBalanceImpl'>

[PatrickDiallo23]

Hello @ge0ffrey,

I have reproduced this issue here where I tried to port the tournament example from Java to Python. It can be tested by pressing the "analyze" button after the Solver finishes its' job.

For a larger context, check the StackOverflow question that I submitted.

Also, I saw in the Java example that the expected JSON response should look like this:

{
    "score": "0hard/-1.19523medium/-2.56348soft",
    "constraints": [
        {
            "package": "org.acme.tournamentschedule.domain",
            "name": "evenlyConfrontationCount",
            "weight": "0hard/0medium/-1soft",
            "score": "0hard/0medium/-2.56348soft",
            "matches": [
                {
                    "score": "0hard/0medium/-2.56348soft",
                    "justification": [
                        {
                            "unfairness": 2.56348
                        }
                    ]
                }
            ],
            "matchCount": 1
        },
        {
            "package": "org.acme.tournamentschedule.domain",
            "name": "fairAssignmentCountPerTeam",
            "weight": "0hard/-1medium/0soft",
            "score": "0hard/-1.19523medium/0soft",
            "matches": [
                {
                    "score": "0hard/-1.19523medium/0soft",
                    "justification": [
                        {
                            "unfairness": 1.19523
                        }
                    ]
                }
            ],
            "matchCount": 1
        },
        {
            "package": "org.acme.tournamentschedule.domain",
            "name": "oneAssignmentPerDatePerTeam",
            "weight": "-1hard/0medium/0soft",
            "score": "0hard/0medium/0soft",
            "matches": [],
            "matchCount": 0
        },
        {
            "package": "org.acme.tournamentschedule.domain",
            "name": "unavailabilityPenalty",
            "weight": "-1hard/0medium/0soft",
            "score": "0hard/0medium/0soft",
            "matches": [],
            "matchCount": 0
        }
    ]
}

I hope these details are useful to you.

Thank you,
Patrick Diallo

[Christopher-Chianelli]

@PatrickDiallo23 there is a workaround; use a custom justification function.

@dataclass
class LoadBalanceJustification(ConstraintJustification):
    unfairness: Decimal

 def fair_assignment_count_per_team(constraint_factory: ConstraintFactory) -> Constraint:
     return (constraint_factory
             .for_each(TeamAssignment)
             .group_by(ConstraintCollectors.load_balance(lambda team_assignment: team_assignment.team))
             .penalize_decimal(HardMediumSoftDecimalScore.ONE_MEDIUM, lambda balance: balance.unfairness())
             .justify_with(lambda balance, score: LoadBalanceJustification(balance.unfairness()))
             .as_constraint("fairAssignmentCountPerTeam"))

This causes (the Java LoadBalance class) to not be in the justification list; instead it is replaced by a double.
Unfortunately, there is a bug currently when custom justifications are used with Decimal scores, which I created a fix for: #1344 .

You can try it out locally by checking out the branch and do pip install path/to/timefold-solver when your virtual environment is activated. Alternatively, you can install it directly with pip install git+https://github.com/Christopher-Chianelli/timefold-solver.git@fix/python-justifications.

[PatrickDiallo23]

Hello @Christopher-Chianelli,

It works. Thank you very much!