Allocation should return same strategy if a model is run twice

[Jingru923]

What
When building up benchmark for Ribasim, it has come to attention that if I run a model with allocation twice, I will get different water balance. This is because some optimization problems have unlimited solutions and the solver returns a random one. Therefore the output of allocation model is always different. This should be investigated.

How

• Investigate JuMP and its solvers. Given an optimization problem with non-unique solution, the solver should return the same solution every time.
• Run model with allocation twice, and see if the output is the same
• Update the benchmark for allocation models.

[Jingru923]

In the regression_test.jl, these tests should be added to the testitem for regression_ode_solvers_allocation.

                # TODO @test all(q -> abs(q) < 12.0, basin.storage - basin_bench.storage)
                # TODO @test all(
                #     q -> abs(q) < 100.0,
                #     basin.storage[(end - 16):end] - basin_bench.storage[(end - 16):end],
                # )
                # TODO @test all(
                #     q -> abs(q) < 2.0,
                #     basin.level[(end - 16):end] - basin_bench.level[(end - 16):end],
                # )
                # TODO @test all(q -> abs(q) < 1e-1, basin.balance_error)
                # TODO @test all(q -> abs(q) < 30.0, basin.relative_error)

[Huite]

Given an optimization problem with non-unique solution, the solver should return the same solution every time

Wouldn't it be better to make sure you have a unique solution? I.e. we should make it so that the optimization problem is well defined?

[SouthEndMusic]

Yes, that's what upcoming issues for allocation are for, especially optimizing per source individually. Still, it's weird that the optimizer is not deterministic.