CODEWITHSUNDEEP
pythonoptimizationpymoo
PeterBe
PeterBe

Reputation: 830

How to set a starting solution in Pymoo

i want to use a starting solution for pymoo in the algorithms NSGA2. For that I have the following code

intial_solution = ICSimulation.simulateDays_ConventionalControl()
algorithm = NSGA2(
    pop_size=5,
    n_offsprings=2,
    sampling=FloatRandomSampling(),
    crossover=SBX(prob=0.7, eta=20),
    mutation=PM(eta=40),
    eliminate_duplicates=True
)



algorithm.setup(problem, x0=intial_solution )

So i just create an initial_solution by using an external file that returns an x vector that is the vector of decision variables in pymoo for the evalutation. Then I add algorithm.setup(problem, x0=intial_solution ). When adding this line, the algorithm does not seem to terminate or to generally show some outputs. Further, in the evaluation method I can clearly see, that the algorithm, even in the first iteration, does not use the inital_solution as all solutions in the first iteration are just randomly generated as if there is no inital_solution.

So I want to know, how can I tell pymoo to use the inital_solution as a good starting point for the optimization instead of just radomly initiallyzing the inital solutions?

Reminder: Does anyone have an idea how to do this?

Upvotes: 1

Views: 494

Answers (1)

Pieter-Jan
Pieter-Jan

Reputation: 540

The first iteration is randomly generated because you're setting sampling=FloatRandomSampling(). You should give your array of the initial design space instead.

For more info see here: https://pymoo.org/customization/initialization.html

Try this:

intial_solution = ICSimulation.simulateDays_ConventionalControl()
algorithm = NSGA2(
    pop_size=5,
    n_offsprings=2,
    sampling=intial_solution,
    crossover=SBX(prob=0.7, eta=20),
    mutation=PM(eta=40),
    eliminate_duplicates=True
)

Upvotes: 0

Related Questions