cocoex.utilities.ProblemNonAnytime

class documentation

class ProblemNonAnytime:

Constructor: ProblemNonAnytime(suite, observer, index)

The non-anytime problem class.

Serves to benchmark a "budgeted" algorithm whose behavior decisively depends on a budget input parameter.

Usage:

# given: suite and observer instances, budget_list, fmin
for index in range(len(suite)):
    with ProblemNonAnytime(suite, observer, index) as problem:
        x0 = problem.initial_solution  # or whatever fmin needs as input
        for budget in sorted(budget_list):
            x = fmin(problem, x0, budget)  # minimize
            problem.delivered(x)  # prepares for next budget also
            if problem.final_target_hit:
                break

Details: This class maintains two problems - one observed and the other unobserved. It switches from unobserved to observed when p_unobserved.evaluations >= p_observed.evaluations.

Method	`__call__`	Undocumented
Method	`__enter__`	Undocumented
Method	`__exit__`	Undocumented
Method	`__init__`	Undocumented
Method	`delivered`	to be called with the solution returned by the solver.
Method	`free`	Undocumented
Method	`print_progress`	Undocumented
Method	`reset`	prepare the problem to be run on the next budget
Class Variable	`inherited_constant_attributes`	Undocumented
Instance Variable	`evaluations`	Undocumented
Instance Variable	`p_observed`	Undocumented
Instance Variable	`p_unobserved`	Undocumented
Instance Variable	`suite`	Undocumented
Property	`best_observed_fvalue1`	Undocumented
Property	`final_target_hit`	Undocumented
Property	`random_solution`	Undocumented
Instance Variable	`_number_of_delivery_evaluations`	Undocumented
Instance Variable	`_p`	Undocumented

def __call__(self, x, *args, **kwargs): ¶

Undocumented

def __enter__(self): ¶

Undocumented

def __exit__(self, *args): ¶

Undocumented

def __init__(self, suite, observer, index): ¶

Undocumented

def delivered(self, x, *args, **kwargs): ¶

to be called with the solution returned by the solver.

The method records the delivered solution (if necessary) and prepares the problem to be run on the next budget by calling reset.

Whether the evaluation is done is based on the comparison with the best so-far fvalue, which is a hack. When the behavior of the solver is known, it is better to use:

problem(x)  # depending on whether this is necessary or not
problem.reset()

def free(self): ¶

Undocumented

def print_progress(self): ¶

Undocumented

def reset(self): ¶

prepare the problem to be run on the next budget

inherited_constant_attributes: list[str] = ¶

Undocumented

evaluations: int = ¶

Undocumented

p_observed = ¶

Undocumented

p_unobserved = ¶

Undocumented

suite = ¶

Undocumented

@property
best_observed_fvalue1 = ¶

Undocumented

@property
final_target_hit = ¶

Undocumented

@property
random_solution = ¶

Undocumented

_number_of_delivery_evaluations: int = ¶

Undocumented

_p = ¶

Undocumented