allow constant in objective cost function

[RobbieKiwi]

Closes #236

Changes proposed in this Pull Request

Allow the objective cost function to contain a constant. While the constant is not needed for optimization, it can make life easier in some cases (see linked issue for discussion).

It was previously forbidden to have a constant in the objective cost function, I assume this is because most solvers expect the ocf to only contain variables.

Now the objective can contain a constant if it is explicitly allowed with

model.add_objective(..., allow_constant=True)

Otherwise adding an objective with a constant will result in an error.

A bit of refactoring was required. To avoid having to repeat the same logic on multiple code paths and touch too much of the core solve logic, I made a decorator which strips the constant term before calling the Model.solve method and then adds it back again after.

@strip_and_replace_constant_objective
def solve(
    self,
    ...

Checklist

• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in doc.
• Unit tests for new features were added (if applicable).
• A note for the release notes doc/release_notes.rst of the upcoming release is included.
• I consent to the release of this PR's code under the MIT license.

[FabianHofmann]

@RobbieKiwi thank you for the pr! I need a bit more time to think about the approach you've taken. I would rather like to have the constant integrated in the full pipeline, such that it appears at the solver side as well (and skip those which are not supporting it). because now you would always see a discrepancy between the linopy objective and the solver objective which on the one hand is confusing, on the other make debugging other things harder.

[RobbieKiwi]

@FabianHofmann what if we added a dummy variable and constrained it to be a certain value and included it in the objective cost function seen by the solver?
Then I guess it would be consistent from end to end apart from the fact we would add/remove the dummy variable before and after solving

[FabianHofmann]

Hey @RobbieKiwi this does unfortunately lead to a wide range in the objective function as soon as the constant is larger then the variables which can easily happen. So that would be also suboptimal.
I guess most solvers support constants in the objective function. I know that
Pyomo has that feature too. Did you do some research lp/mps files support a constant?

[RobbieKiwi]

@FabianHofmann
I think that trying to get the constant all the way through the solver may imply making some changes for each combination of solver and io_api which could be a lot of work for fairly limited gain. Maybe there is no nice way to do this and it is better to park it