Reproducibility for the solutions in nsga

[muhammadafdal09]

Hello, does anyone know how to make nsga 3 reproductible? I have tried to add seed in minimize but the solutions were still different everytime i execute the cell. I have tried to restart session and run all more than once and it will always give different solutions each time

This is the code

from scipy.stats import skew, kurtosis
from pymoo.core.problem import Problem
from pymoo.algorithms.moo.nsga3 import NSGA3
from pymoo.operators.crossover.sbx import SBX
from pymoo.operators.mutation.pm import PM
from pymoo.optimize import minimize
from pymoo.util.ref_dirs import get_reference_directions
from pymoo.termination.default import DefaultMultiObjectiveTermination

# Define Reference Directions
ref_dirs = get_reference_directions("das-dennis", 5, n_partitions=6)

class PortofolioProblem(Problem):
    def __init__(self, data_return_df):
        self.data_m = np.ascontiguousarray(data_return_df.sort_index(axis=1).values, dtype=np.float64)
        n_assets = self.data_m.shape[1]
        self.mu = self.data_m.mean(axis=0)
        self.cov_mat = np.cov(self.data_m, rowvar=False)
        xl, xu = np.full(n_assets, 0.01), np.full(n_assets, 0.15)
        super().__init__(n_var=n_assets, n_obj=5, n_ieq_constr=3, xl=xl, xu=xu)

    def _evaluate(self, x, out, *args, **kwargs):
        w = x / np.sum(x, axis=1)[:, None]
        rets = np.dot(w, self.mu)
        v = np.array([np.dot(wi.T, np.dot(self.cov_mat, wi)) for wi in w])
        stds = np.sqrt(np.maximum(v, 0))

        port_rets_series = np.dot(w, self.data_m.T)
        skews = skew(port_rets_series, axis=1)
        kurts = kurtosis(port_rets_series, axis=1)

        idx = max(1, int(0.05 * self.data_m.shape[0]))
        cvars = -np.mean(np.sort(port_rets_series, axis=1, kind='mergesort')[:, :idx], axis=1)

        out["F"] = np.column_stack([-rets, stds, -skews, kurts, cvars])

        epsilon = 1e-6
        n_sel = np.sum(w > (0.01 - epsilon), axis=1)

        out["G"] = np.column_stack([
            np.abs(np.sum(w, axis=1) - 1.0) - 1e-4,
            5 - n_sel,
            n_sel - 15
        ])

algorithm = NSGA3(
    ref_dirs=ref_dirs,
    crossover=SBX(prob=0.9, eta=20),
    mutation=PM(prob=0.05, eta=20),
    eliminate_duplicates=True
)

problem_instance = PortofolioProblem(data_training_terpilih)
terminasi_otomatis = DefaultMultiObjectiveTermination(
    xtol=1e-4, cvtol=1e-6, ftol=0.0025, period=50, n_max_gen=2500, n_max_evals=577500
)

optimal_portofolios = minimize(
    problem_instance,
    algorithm,
    termination=terminasi_otomatis,
    seed=42,
    verbose=True,
    save_history=True
)