Store extra solution information along with optimization results?

[khoopes]

Hello,

I am just getting started with platypus, switching to python from MATLAB, and everything is working well so far. I am wondering if there is a way to store additional information for each solution that I could use later for post processing? Right now the function to be optimized returns the result along with any constraints, what would really help is if I could also pass along another variable, a dictionary say, that would live with each solution. Then when I am post processing the Pareto front I would not have to reevaluate all the non-dominated solutions to get this extra information.

Thanks,

[kimbrerfesten]

Hi

Would be easy to have this in you objective function. Just write/append to a file all the data and variables.

I use this so I can use Xdat to make parallel plots.

Not that it couldn't be part of Platypus, just saying until then you can do it yourself.

[Wbec]

I have been using this archive class for that purpose. Switch platypus.Archive for any other Archive that you need, and then call the algorithm with archive = LoggingArchive(any_args_here)

class LoggingArchive(platypus.Archive):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, *kwargs)
        self.log = []
    
    def add(self, solution):
        super().add(solution)
        self.log.append(solution)

Then use algorithm.archive.log to retrieve the list of solutions.

[morecfd]

Can you please show snippets how to use the procedure in an optimization model (for a python and platypus newbie)?

[Wbec]

from platypus import NSGAII, Problem, Real, Archive

# We can subclass other kinds of Archive instead if needed.
class LoggingArchive(Archive):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, *kwargs)
        self.log = []
    
    def add(self, solution):
        super().add(solution)
        self.log.append(solution)
        
log_archive = LoggingArchive()

# copy the example from the platypus README
def schaffer(x):
    return [x[0]**2, (x[0]-2)**2]
problem = Problem(1, 2)
problem.types[:] = Real(-10, 10)
problem.function = schaffer

# here we tell the optimizer to use the archive we chose.
algorithm = NSGAII(problem, archive = log_archive)
algorithm.run(10)

print(log_archive.log[:5]) # print the first 5 solutions in the log

[tianjingyao]

from platypus import NSGAII, Problem, Real, Archive

# We can subclass other kinds of Archive instead if needed.
class LoggingArchive(Archive):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, *kwargs)
        self.log = []
    
    def add(self, solution):
        super().add(solution)
        self.log.append(solution)
        
log_archive = LoggingArchive()

# copy the example from the platypus README
def schaffer(x):
    return [x[0]**2, (x[0]-2)**2]
problem = Problem(1, 2)
problem.types[:] = Real(-10, 10)
problem.function = schaffer

# here we tell the optimizer to use the archive we chose.
algorithm = NSGAII(problem, archive = log_archive)
algorithm.run(10)

print(log_archive.log[:5]) # print the first 5 solutions in the log

Thanks, but there is another question. I want to save the records from EpsNSGAII. Could you show me how to do that?

[sabrinadraude]

Has anyone found a way to use this data to create an unbounded archive? So that it is guaranteed that platypus keeps the best results from the previous iteration into the next iteration.

I am doing a 3 objective optimisation at the moment, but currently the hypervolume does not improve greatly over 50,000 iterations. I thought maybe an unbound archive might help with the huge search space? Please let me know if anyone has any tips.

[github-actions[bot]]

This issue is stale and will be closed soon. If you feel this issue is still relevant, please comment to keep it active. Please also consider working on a fix and submitting a PR.