|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import time |
|
6
|
|
|
import pytest |
|
7
|
|
|
import random |
|
8
|
|
|
import numpy as np |
|
9
|
|
|
|
|
10
|
|
|
from gradient_free_optimizers import LipschitzOptimizer |
|
11
|
|
|
from ._base_para_test import _base_para_test_func |
|
12
|
|
|
from gradient_free_optimizers import RandomSearchOptimizer |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
def objective_function_nan(para): |
|
16
|
|
|
rand = random.randint(0, 1) |
|
17
|
|
|
|
|
18
|
|
|
if rand == 0: |
|
19
|
|
|
return 1 |
|
20
|
|
|
else: |
|
21
|
|
|
return np.nan |
|
22
|
|
|
|
|
23
|
|
|
|
|
24
|
|
|
def objective_function_m_inf(para): |
|
25
|
|
|
rand = random.randint(0, 1) |
|
26
|
|
|
|
|
27
|
|
|
if rand == 0: |
|
28
|
|
|
return 1 |
|
29
|
|
|
else: |
|
30
|
|
|
return -np.inf |
|
31
|
|
|
|
|
32
|
|
|
|
|
33
|
|
|
def objective_function_inf(para): |
|
34
|
|
|
rand = random.randint(0, 1) |
|
35
|
|
|
|
|
36
|
|
|
if rand == 0: |
|
37
|
|
|
return 1 |
|
38
|
|
|
else: |
|
39
|
|
|
return np.inf |
|
40
|
|
|
|
|
41
|
|
|
|
|
42
|
|
|
search_space_ = {"x1": np.arange(0, 20, 1)} |
|
43
|
|
|
|
|
44
|
|
|
|
|
45
|
|
|
def objective_function(para): |
|
46
|
|
|
score = -para["x1"] * para["x1"] |
|
47
|
|
|
return score |
|
48
|
|
|
|
|
49
|
|
|
|
|
50
|
|
|
search_space = {"x1": np.arange(-10, 11, 1)} |
|
51
|
|
|
search_space2 = {"x1": np.arange(-10, 51, 1)} |
|
52
|
|
|
search_space3 = {"x1": np.arange(-50, 11, 1)} |
|
53
|
|
|
|
|
54
|
|
|
|
|
55
|
|
|
opt1 = RandomSearchOptimizer(search_space) |
|
56
|
|
|
opt2 = RandomSearchOptimizer(search_space2) |
|
57
|
|
|
opt3 = RandomSearchOptimizer(search_space3) |
|
58
|
|
|
opt4 = RandomSearchOptimizer(search_space_) |
|
59
|
|
|
opt5 = RandomSearchOptimizer(search_space_) |
|
60
|
|
|
opt6 = RandomSearchOptimizer(search_space_) |
|
61
|
|
|
|
|
62
|
|
|
opt1.search(objective_function, n_iter=30) |
|
63
|
|
|
opt2.search(objective_function, n_iter=30) |
|
64
|
|
|
opt3.search(objective_function, n_iter=30) |
|
65
|
|
|
opt4.search(objective_function_nan, n_iter=30) |
|
66
|
|
|
opt5.search(objective_function_m_inf, n_iter=30) |
|
67
|
|
|
opt6.search(objective_function_inf, n_iter=30) |
|
68
|
|
|
|
|
69
|
|
|
search_data1 = opt1.search_data |
|
70
|
|
|
search_data2 = opt2.search_data |
|
71
|
|
|
search_data3 = opt3.search_data |
|
72
|
|
|
search_data4 = opt4.search_data |
|
73
|
|
|
search_data5 = opt5.search_data |
|
74
|
|
|
search_data6 = opt6.search_data |
|
75
|
|
|
|
|
76
|
|
|
|
|
77
|
|
|
lipschitz_para = [ |
|
78
|
|
|
({"warm_start_smbo": None}), |
|
79
|
|
|
({"warm_start_smbo": search_data1}), |
|
80
|
|
|
({"warm_start_smbo": search_data2}), |
|
81
|
|
|
({"warm_start_smbo": search_data3}), |
|
82
|
|
|
({"warm_start_smbo": search_data4}), |
|
83
|
|
|
({"warm_start_smbo": search_data5}), |
|
84
|
|
|
({"warm_start_smbo": search_data6}), |
|
85
|
|
|
({"max_sample_size": 10000000}), |
|
86
|
|
|
({"max_sample_size": 10000}), |
|
87
|
|
|
({"max_sample_size": 1000000000}), |
|
88
|
|
|
({"sampling": False}), |
|
89
|
|
|
({"sampling": {"random": 1}}), |
|
90
|
|
|
({"sampling": {"random": 100000000}}), |
|
91
|
|
|
({"replacement": True}), |
|
92
|
|
|
({"replacement": False}), |
|
93
|
|
|
] |
|
94
|
|
|
|
|
95
|
|
|
|
|
96
|
|
|
pytest_wrapper = ("opt_para", lipschitz_para) |
|
97
|
|
|
|
|
98
|
|
|
|
|
99
|
|
|
@pytest.mark.parametrize(*pytest_wrapper) |
|
100
|
|
|
def test_lipschitz_para(opt_para): |
|
101
|
|
|
_base_para_test_func(opt_para, LipschitzOptimizer) |
|
102
|
|
|
|
|
103
|
|
|
|
|
104
|
|
|
def test_warm_start_0(): |
|
105
|
|
|
opt = LipschitzOptimizer(search_space, warm_start_smbo=search_data1) |
|
106
|
|
|
|
|
107
|
|
|
assert len(opt.X_sample) == 30 |
|
108
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
