Add structure to solve optimization problems

[S-Dafarra]

An optimization problem can be set as follows

    @dataclasses.dataclass
    class SwitchVar(OptimizationObject):
        x: StorageType = default_storage_field(ContinuousVariable)
        y: StorageType = default_storage_field(ContinuousVariable)

        def __post_init__(self):
            self.x = np.zeros(1)
            self.y = np.zeros(1)

    a = 10
    new_problem = OptimizationProblem()
    new_variables = new_problem.generate_optimization_objects(SwitchVar())
    new_problem.add_expression(
        ExpressionType.minimize, a * new_variables.y * new_variables.y
    )
    new_problem.add_expression(
        ExpressionType.subject_to, new_variables.x + new_variables.y == a - 1
    ) 
    new_problem.add_expression(ExpressionType.minimize, new_variables.x == 5)
    output, cost_value = new_problem.solver().solve()

Note the line

new_problem.add_expression(ExpressionType.minimize, new_variables.x == 5)

Equalities can be cast into a cost. This might simplify the setting of tracking tasks.

[traversaro]

Ok, some suggestions on the API but non blocking

[S-Dafarra]

Another suggestion would be to add named arguments to have a scaling when passing a costraint and cast it into a cost.

[S-Dafarra]

Another suggestion would be to add named arguments to have a scaling when passing a costraint and cast it into a cost.

Done in 109859c

[S-Dafarra]

Friendly ping @traversaro @diegoferigo

[traversaro]

Another suggestion would be to add named arguments to have a scaling when passing a costraint and cast it into a cost.

Done in 109859c

Great, thanks!

[diegoferigo]

Nice @S-Dafarra 🚀 I like the direction. I left (too many) comments on this PR, feel free to include here what you think useful and -hopefully- address the nits/suggestions/python syntax enhancements in next ones. Approving regardless so you can move forward.

[diegoferigo]

Nit

Suggested change

	if isinstance(input_structure, OptimizationObject):
	return self._generate_objects_from_instance(input_structure=input_structure)
	return self._generate_objects_from_list(input_structure=input_structure)
	return (
	self._generate_objects_from_instance if isinstance(input_structure, OptimizationObject)
	else self._generate_objects_from_instance
	)(input_structure=input_structure)

[S-Dafarra]

I tried, but it does not work because it returns a function, not an object

[diegoferigo]

The first () return a function selecting one among the two options, the second () passes to it the argument, this should work fine

[S-Dafarra]

If I apply the suggestion, this test fails with the error. Any suggestion?

test/test_optimization_problem.py:129 (test_opti_solver_with_parameters_and_lists)
class_or_instance = [MyTestVarAndPar(composite=MyTestVar(variable=array([0., 0., 0.])), parameter=array([0., 0., 0.])), MyTestVarAndPar(co...rray([0., 0., 0.])), MyTestVarAndPar(composite=MyTestVar(variable=array([0., 0., 0.])), parameter=array([0., 0., 0.]))]

    def fields(class_or_instance):
        """Return a tuple describing the fields of this dataclass.
    
        Accepts a dataclass or an instance of one. Tuple elements are of
        type Field.
        """
    
        # Might it be worth caching this, per class?
        try:
>           fields = getattr(class_or_instance, _FIELDS)
E           AttributeError: 'list' object has no attribute '__dataclass_fields__'

../../mambaforge/envs/hippopt/lib/python3.10/dataclasses.py:1196: AttributeError

During handling of the above exception, another exception occurred:

    def test_opti_solver_with_parameters_and_lists():
        problem = OptimizationProblem()
        initial_guess = []
        for _ in range(3):
            initial_guess.append(MyTestVarAndPar())
    
>       var = problem.generate_optimization_objects(input_structure=initial_guess)

test_optimization_problem.py:136: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../src/hippopt/base/optimization_problem.py:37: in generate_optimization_objects
    return self._solver.generate_optimization_objects(
../src/hippopt/base/opti_solver.py:346: in generate_optimization_objects
    return (
../src/hippopt/base/opti_solver.py:67: in _generate_objects_from_instance
    for field in dataclasses.fields(output):
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

class_or_instance = [MyTestVarAndPar(composite=MyTestVar(variable=array([0., 0., 0.])), parameter=array([0., 0., 0.])), MyTestVarAndPar(co...rray([0., 0., 0.])), MyTestVarAndPar(composite=MyTestVar(variable=array([0., 0., 0.])), parameter=array([0., 0., 0.]))]

    def fields(class_or_instance):
        """Return a tuple describing the fields of this dataclass.
    
        Accepts a dataclass or an instance of one. Tuple elements are of
        type Field.
        """
    
        # Might it be worth caching this, per class?
        try:
            fields = getattr(class_or_instance, _FIELDS)
        except AttributeError:
>           raise TypeError('must be called with a dataclass type or instance')
E           TypeError: must be called with a dataclass type or instance

../../mambaforge/envs/hippopt/lib/python3.10/dataclasses.py:1198: TypeError

[S-Dafarra]

@diegoferigo I should have addressed your comments (except #4 (comment) where I am not sure how to continue).

I set the comments to resolved quite arbitrarily, so feel free to "Unresolve" the points that need some further discussion.

Let me know how you would like to proceed 😉

[S-Dafarra]

I will proceed in merging this. Other open points will be addressed in future PRs.