How to change a value in a constraint in each generation using NSGA-II in Pymoo?

[jadiazgu]

Hi Julian,

Please, I need your help to fix my code using Pymoo in Python. Next, I will explain what I need to do:

My code starts finding a "value simulated" as follows:

import numpy
value_simulated = numpy.min(numpy.random.uniform(0.1, 1, 1000))

This "value simulated" is used in class "MyProblem" in order to affect the behavior of the constraint "g1" as follows:

from pymoo.core.problem import ElementwiseProblem
class MyProblem(ElementwiseProblem):

    def __init__(self):
        super().__init__(n_var=2,
                         n_obj=2,
                         n_ieq_constr=1,
                         xl=numpy.array([0.1, 0]),
                         xu=numpy.array([1, 5]))

    def _evaluate(self, x, out, *args, **kwargs):
        f1 = x[0]
        f2 = (1 + x[1]) / x[0]

        if value_simulated <= 1:
            a = 0
        else:
            a = 1

        g1 = a * x[0] + a * x[1]

        out["F"] = [f1, f2]
        out["G"] = [g1]

problem = MyProblem()

I need this "value_simulated" affect constraint "g1" only during generation #1. During next generations "value_simulated" must be replaced by the min value that will be found in the population. By that reason I coded the class "Callback" to detect this min value, but I don't know how to call this value in the next generations in order to replace the "value_simulated" in the problem, callback is coded as follows:

from pymoo.core.callback import Callback

class CustomCallback(Callback):
    def __init__(self):
        super().__init__()

    def notify(self, algorithm):
        solutions = algorithm.pop.get("F")
        value_f1min = min([row[0] for row in solutions])

        return value_f1min

custom_callback = CustomCallback()

I used the function "notify", but I suspect that the right function is "update", but I don't know how to use the function "update". The last part of my code is the following:

from pymoo.algorithms.moo.nsga2 import NSGA2
from pymoo.optimize import minimize

algorithm = NSGA2(pop_size=100, callback=custom_callback)

res = minimize(problem, algorithm, ('n_gen', 10), seed=1, verbose=True)

F = res.F
Pop = res.pop
final_solutions = res.X

Could you help me please to fix my code in order to affect constaint "g1" correctly in each generation? I will appreciate your help a lot! Thank you.

[blankjul]

What about passing the evaluation population during initialization?
https://pymoo.org/customization/initialization.html

Does this resolve this issue?

[jadiazgu]

Thank you Julian for your answer!

I've tried your suggestion but I'm afraid it doesn't solve this issue because I'm looking for a way to modify, in each generation, a parameter included in constraint "g1". This parameter must be taken from the min value of f1 that belongs to previous generation.

[blankjul]

For this I would probably overwrite an existing algorithm to achieve this. It sounds more like an algorithm customization.

[jadiazgu]

Thank you Julian for your advice.
I tried by declaring a global variable inside the Callback Class and with this I was able to use this variable inside the constraint in every generation.

from pymoo.core.callback import Callback

class CustomCallback(Callback):
    def __init__(self):
        super().__init__()

    def notify(self, algorithm):
	global value_f1min
        solutions = algorithm.pop.get("F")
        value_f1min = min([row[0] for row in solutions])

        return