Experimenter using own problem

[sabrinadraude]

Hi there,

I was just wondering how to use the experimenter using ones own problem rather than a predefined problem?

Run the algorithm

N = 2000
problem = Problem(multiple_days_assigned_size, 2, 2)
problem.directions[1] = Problem.MINIMIZE
problem.directions[0] = Problem.MAXIMIZE
problem.constraints[:] = ">=0"
problem.types[:] = Integer(0, my_counter_again - 1)
problem.function = lambda x: evaluation_function2(x, jobs, c_time_added_up, total_allowed_time_all_jobs,
extra_time)
algorithm = NSGAII(problem)
algorithm.run(N)
feasible_solutions_total = [s for s in algorithm.result if s.feasible]
Nondominated_solutions_total = nondominated(algorithm.result)

calculate the difference in models nsga2 and nsga3

if name == "main":
algorithms = [NSGAII, (NSGAIII, {"divisions_outer":12})]
problems = [Problem(multiple_days_assigned_size, 2, 2)] #what would I put in here?

# run the experiment
results = experiment(algorithms, problems, nfe=10000, seeds=10)

Any tips would be greatly appreciated!

Thanks

[grafkevin]

I am having the same problem.

[sabrinadraude]

Sorry I have not yet received a reply.

[jetuk]

What error message are you getting?

[sabrinadraude]

solution.variables = [x.rand() for x in problem.types]
AttributeError: 'NoneType' object has no attribute 'rand'

My question is how to set out the code for using your own problem in the experimenter generator rather than a predefined problem such as DTLZ2 or Belegundu.

N = 2 problem = Problem(multiple_days_assigned_size, 2, 2) problem.directions[1] = Problem.MINIMIZE problem.directions[0] = Problem.MAXIMIZE problem.constraints[:] = ">=0" problem.types[:] = Integer(0, my_counter_again - 1) problem.function = lambda x: evaluation_function2(x, jobs, c_time_added_up, total_allowed_time_all_jobs, extra_time)

# calculate the difference in models nsga2 and nsga3 if __name__ == "__main__": algorithms = [NSGAII, (NSGAIII, {"divisions_outer":12})] problems = [Problem(multiple_days_assigned_size, 2, 2)]

# run the experiment
results = experiment(algorithms, problems, nfe=10000, seeds=10) `

# calculate the hypervolume indicator between two algorithms hyp = Hypervolume(minimum=[min_pri, min_cost], maximum=[max_pri, max_cost]) hyp_result = calculate(results, hyp) display(hyp_result, ndigits=3)

[jetuk]

I think the issue is that you are redefining the problem on this line:

problems = [Problem(multiple_days_assigned_size, 2, 2)]

I.e. the problem here is not the same Problem instances as is assigned to the problem variable at the start of the script. Consequently this new instance has no types, etc. defined, hence the error about None.

I think you need to use the reference/variable to the problem you have setup at the start of the script, and pass that to experiment. Something like this:

N = 2
problem = Problem(multiple_days_assigned_size, 2, 2)
problem.directions[1] = Problem.MINIMIZE
problem.directions[0] = Problem.MAXIMIZE
problem.constraints[:] = ">=0"
problem.types[:] = Integer(0, my_counter_again - 1)
problem.function = lambda x: evaluation_function2(x, jobs, c_time_added_up, total_allowed_time_all_jobs, extra_time)

# calculate the difference in models nsga2 and nsga3
if __name__ == "__main__": 
    algorithms = [NSGAII, (NSGAIII, {"divisions_outer":12})]

    # run the experiment
    results = experiment(algorithms, [problem], nfe=10000, seeds=10) `

    # calculate the hypervolume indicator between two algorithms
    hyp = Hypervolume(minimum=[min_pri, min_cost], maximum=[max_pri, max_cost]) 
    hyp_result = calculate(results, hyp) display(hyp_result, ndigits=3)

PS if you put your code between three back-ticks it'll format correctly and be easier to read. See: https://guides.github.com/features/mastering-markdown/

[grafkevin]

Hi Jetuk, thanks for your help!

I achieved to get passed the AttributeError: 'NoneType' object has no attribute 'rand' error. But now I get this error: AttributeError: 'NSGAII' object has no attribute '__name__'

Here's my code:

# Custom experimenter
from platypus import *

problem = Problem(2, 2)
problem.types[0:] = Permutation(range(len(carbon_bus)))
problem.types[1:] = Binary(len(carbon_bus))
problem.directions[:] = Problem.MINIMIZE
problem.function = two_opt

if __name__ == "__main__":
    algorithms = [NSGAII(problem,variator=PMX()), (NSGAIII, {"divisions_outer":12})]
    
    # run the experiment
    results = experiment(algorithms, problem, seeds=5, nfe=1000)

    # calculate the hypervolume indicator 
    #hyp = Hypervolume(minimum=[0, 0, 0], maximum=[1, 1, 1])
    #hyp_result = calculate(results, hyp)
    #display(hyp_result, ndigits=3)

I am working on a bi-objective TSP.

[jetuk]

Can you paste the full traceback?

[grafkevin]

Sure, sorry for the confusion!

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-91-4e87550fe452> in <module>
     12 
     13     # run the experiment
---> 14     results = experiment(algorithms, problem, seeds=5, nfe=1000)
     15 
     16     # calculate the hypervolume indicator

c:\users\kevin\appdata\local\programs\python\python37\lib\site-packages\platypus\experimenter.py in experiment(algorithms, problems, seeds, nfe, evaluator, display_stats)
    179         evaluator = PlatypusConfig.default_evaluator
    180 
--> 181     job_results = evaluator.evaluate_all(generator)
    182 
    183     # convert results to structured format

c:\users\kevin\appdata\local\programs\python\python37\lib\site-packages\platypus\evaluator.py in evaluate_all(self, jobs, **kwargs)
     86 
     87         if log_frequency is None:
---> 88             return list(self.map_func(run_job, jobs))
     89         else:
     90             result = []

c:\users\kevin\appdata\local\programs\python\python37\lib\site-packages\platypus\experimenter.py in evaluate_job_generator(algorithms, problems, seeds, nfe, display_stats)
     89         else:
     90             algorithm = algorithms[i]
---> 91             algorithm_name = algorithm.__name__
     92             kwargs = {}
     93 

AttributeError: 'NSGAII' object has no attribute '__name__'

[jetuk]

I think the experimenter is expecting algorithm classes, not instances. Try this:

# Custom experimenter
from platypus import *

problem = Problem(2, 2)
problem.types[0:] = Permutation(range(len(carbon_bus)))
problem.types[1:] = Binary(len(carbon_bus))
problem.directions[:] = Problem.MINIMIZE
problem.function = two_opt

if __name__ == "__main__":
    algorithms = [(NSGAII, {"variator": PMX()}), (NSGAIII, {"divisions_outer":12})]
    
    # run the experiment
    results = experiment(algorithms, problem, seeds=5, nfe=1000)

    # calculate the hypervolume indicator 
    #hyp = Hypervolume(minimum=[0, 0, 0], maximum=[1, 1, 1])
    #hyp_result = calculate(results, hyp)
    #display(hyp_result, ndigits=3)

[sabrinadraude]

Hi there,

Thank you for your help.

I think the new code worked however I got the issue:

raise PlatypusError("NSGAIII currently only works with minimization problems")
platypus.core.PlatypusError: NSGAIII currently only works with minimization problems

Do any of the other GAs work with maximization problems or is it just NSGAII?

Cheers,

Sabrina

[jetuk]

I am not sure, you'll have to check the code. A work around is to use minimisation and return the negative of the value for objectives you wish to maximise.

[sabrinadraude]

Thanks the code works for minimisation.

However when I try and run the code using the evaluator it hasn't finished, even though the code has been running for three hours now:

N = 10
problem = Problem(multiple_days_assigned_size, 2, 2)
problem.directions[1] = Problem.MINIMIZE
problem.directions[0] = Problem.MINIMIZE
problem.constraints[:] = ">=0"
problem.types[:] = Integer(0, my_counter_again - 1)
problem.function = lambda x: evaluation_function2(x, jobs, c_time_added_up, total_allowed_time_all_jobs,
                                                  extra_time)
algorithm = NSGAII(problem)
algorithm.run(N)

# Calculate the hypervolume using assigned objective max and mins for comparing nsga2
min_pri = 0
min_cost = multiple_days_assigned_size * 30
max_pri = multiple_days_assigned_size * 6.53
max_cost = multiple_days_assigned_size * 130
hyp = Hypervolume(minimum=[min_pri, min_cost], maximum=[max_pri, max_cost])

if __name__ == "__main__":
    algorithms = [NSGAII, (NSGAIII, {"divisions_outer": 12})]

    with ProcessPoolEvaluator(2) as evaluator:
        results = experiment(algorithms, [problem], nfe=100, evaluator=evaluator)
        hyp = Hypervolume(minimum=[min_pri, min_cost], maximum=[max_pri, max_cost])
        hyp_result = calculate(results, hyp, evaluator=evaluator)
        display(hyp_result, ndigits=2)