from functools import partial
import numpy as np
from smt_optim.benchmarks.base import BenchmarkProblem
from smt_optim.benchmarks.avt311.avt311 import MFRosenbrock
# class MFBarnes(BenchmarkProblem):
# """
# Fischer, C. C. (2021). Bayesian Inspired Multi-Fidelity Optimization with Aerodynamic Design.
# """
# def __init__(self):
# super().__init__()
#
# self.num_dim = 2
# self.num_cstr = 3
# self.num_fidelity = 2
# self.num_obj = 1
# self.bounds = np.array([
# [0, 80],
# ] * self.num_dim)
#
# self.objective = [
# # Taylor expansion n=3 around x=(30, 40), -> to add when SMT-optim dependancies include a package with
# # autodiff functionalities.
# self.func
# ]
#
# self.constraints = [
# [self.g1_lf, self.g1],
# [self.g2_lf, self.g2],
# [self.g3_lf, self.g3],
# ]
#
# def func(self, x):
# return (
# 75.196
# - 3.8112 * x[0]
# + 0.12694 * x[0] ** 2
# - 2.0567e-3 * x[0] ** 3
# + 1.0345e-5 * x[0] ** 4
# - 6.8306 * x[1]
# + 0.25645 * x[1] ** 2
# - 3.4604e-3 * x[1] ** 3
# + 1.3514e-5 * x[1] ** 4
# + 0.030234 * x[0] * x[1]
# - 1.28134e-3 * x[0] * x[1] ** 2
# + 3.5256e-5 * x[0] * x[1] ** 3
# - 2.266e-7 * x[0] * x[1] ** 4
# + 3.4054e-4 * x[0] ** 2 * x[1]
# - 5.2375e-6 * x[0] ** 2 * x[1] ** 2
# - 6.3e-8 * x[0] ** 2 * x[1] ** 3
# - 1.6638e-6 * x[0] ** 3 * x[1]
# + 7.0e-10 * x[0] ** 3 * x[1] ** 3
# - 28.106 / (x[1] + 1.0)
# - 2.8673 * np.exp(5.0e-4 * x[0] * x[1])
# )
#
# def g1(self, x):
# return 1 - x[0]*x[1]/700
#
# def g2(self, x):
# return x[0]**2/625 - x[1]/5
#
# def g3(self, x):
# return (x[0]/500 - 0.11) - (x[1]/50 - 1)**2
#
# def g1_lf(self, x):
# return (50 - x[0] - x[1])/10
#
# def g2_lf(self, x):
# return (0.64*x[0] - x[1])/6
#
# def g3_lf(self, x):
# if x[1] > 50:
# return 0.34 + 0.006*x[0] - 0.0134*x[1]
# else:
# return 0.0134*x[1] + 0.006*x[0] - 1
[docs]
class MFConstraintRosenbrock(BenchmarkProblem):
"""
Fischer, C. C. (2021). Bayesian Inspired Multi-Fidelity Optimization with Aerodynamic Design.
"""
def __init__(self):
super().__init__()
self.num_dim = 2
self.num_cstr = 1
self.num_fidelity = 2
self.num_obj = 1
self.bounds = np.array(
[
[-2, 2],
]
* self.num_dim
)
self.ref_problem = MFRosenbrock()
self.objective = [
partial(self.ref_problem.f, fid=0),
partial(self.ref_problem.f, fid=2),
]
self.constraints = [[self.g1_lf, self.g1]]
self.tags = ["n_variable"]
[docs]
def set_dim(self, dim):
self.ref_problem.set_dim(dim)
self.num_dim = dim
self.bounds = np.array([[-2, 2]] * dim)
[docs]
def g1(self, x):
return x[0] ** 2 + np.sum(x[1:] - 1) / 2
[docs]
def g1_lf(self, x):
return self.g1(x) + 0.1 * np.sin(10 * x[0] + np.sum(5 * x[1:]))
if __name__ == "__main__":
from smt_optim.subsolvers.multistart import multistart_minimize
prob = MFConstraintRosenbrock()
prob.set_dim(2)
constraints = [
{
"fun": lambda x, f=prob.g1: -f(x),
"type": "ineq",
},
# {
# "fun": lambda x, f=prob.g2: -f(x),
# "type": "ineq",
# },
# {
# "fun": lambda x, f=prob.g3: -f(x),
# "type": "ineq",
# },
]
# res = so.minimize(prob.objective[-1], x0=np.array([0., 0.]), bounds=prob.bounds, constraints=constraints)
res = multistart_minimize(
prob.objective[-1], bounds=prob.bounds, constraints=constraints
)
print(res.x)
print(res.fun)
print(prob.g1(res.x))
# print(res)
# import matplotlib.pyplot as plt
# from smt_optim.utils.plot_2d import get_plot2d_data
#
# prob.set_dim(2)
# XX, YY, ZZ = get_plot2d_data(prob.constraints[0][0], prob.bounds)
#
# fig, ax = plt.subplots()
# ax.contourf(XX, YY, np.where(ZZ<=0, np.nan, ZZ))
# plt.show()
