from numpy.core.numeric import indices
from numpy.lib.arraysetops import unique
from pandas.core.frame import DataFrame
from desdeo_tools.scalarization.ASF import SimpleASF, ReferencePointASF
from numba import njit
import numpy as np
import pandas as pd
import copy
from desdeo_problem.surrogatemodels.SurrogateModels import GaussianProcessRegressor
from sklearn.gaussian_process.kernels import DotProduct,\
WhiteKernel, RBF, Matern, ConstantKernel
from desdeo_problem import MOProblem
#TODO: add @njit function here
[docs]def remove_duplicate(
X: np.ndarray,
archive_x: np.ndarray
):
"""identifiesthe duplicate rows for decision variables
Args:
X (np.ndarray): the current decision variables.
archive_x (np.ndarray): The decision variables in the archive.
Returns:
indicies (np.ndarray): the indicies of solutions that are NOT already in the archive.
"""
all_variables = np.vstack((archive_x,X))
all_variables_indicies = np.arange(len(all_variables))
e,unique_variables_indicies = np.unique(all_variables, return_index=True, axis=0)
repeated_all_variables = np.delete(all_variables_indicies, unique_variables_indicies)
repeated_in_X = repeated_all_variables - len(archive_x)
X_indicies = np.arange(len(X))
X_uniqe_indicies = np.delete(X_indicies, repeated_in_X)
return X_uniqe_indicies
[docs]def ikrvea_mm(
reference_point: np.ndarray,
individuals: np.ndarray,
objectives: np.ndarray,
uncertainity: np.ndarray,
problem: MOProblem,
u: int) -> float:
""" Selects the solutions that need to be reevaluated with the original functions.
This model management is based on the following papaer:
'P. Aghaei Pour, T. Rodemann, J. Hakanen, and K. Miettinen, “Surrogate assisted interactive
multiobjective optimization in energy system design of buildings,”
Optimization and Engineering, 2021.'
Args:
reference_front (np.ndarray): The reference front that the current front is being compared to.
Should be an one-dimensional array.
individuals (np.ndarray): Current individuals generated by using surrogate models
objectives (np.ndarray): Current objectives generated by using surrogate models
uncertainity (np.ndarray): Current Uncertainty values generated by using surrogate models
problem : the problem class
Returns:
float: the new problem object that has an updated archive.
"""
nd = remove_duplicate(individuals, problem.archive.drop(
problem.objective_names, axis=1).to_numpy()) #removing duplicate solutions
if len(nd) == 0:
return problem
else:
non_duplicate_dv = individuals[nd]
non_duplicate_obj = objectives[nd]
non_duplicate_unc = uncertainity[nd]
# Selecting solutions with lowest ASF values
asf_solutions = SimpleASF([1]*problem.n_of_objectives).__call__(non_duplicate_obj, reference_point)
idx = np.argpartition(asf_solutions, 2*u)
asf_unc = np.max(non_duplicate_unc [idx[0:2*u]], axis= 1)
# index of solutions with lowest Uncertainty
lowest_unc_index = np.argpartition(asf_unc, u)[0:u]
# evaluating the solutions in asf_unc with lowest uncertainty. The archive will get update in problem.evaluate()
problem.evaluate(non_duplicate_dv[lowest_unc_index], use_surrogate=False)[0]
problem.train(models=GaussianProcessRegressor,\
model_parameters={'kernel': Matern(nu=1.5)})
return problem
