|
1
|
|
|
"""Test module for constraint optimization strategy.""" |
|
2
|
|
|
|
|
3
|
|
|
import numpy as np |
|
4
|
|
|
|
|
5
|
|
|
from hyperactive import Hyperactive |
|
6
|
|
|
from hyperactive.optimizers import HillClimbingOptimizer, RandomSearchOptimizer |
|
7
|
|
|
from hyperactive.optimizers.strategies import CustomOptimizationStrategy |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
def test_constr_opt_0(): |
|
11
|
|
|
"""Test constrained optimization with single constraint.""" |
|
12
|
|
|
|
|
13
|
|
|
def objective_function(para): |
|
14
|
|
|
score = -para["x1"] * para["x1"] |
|
15
|
|
|
return score |
|
16
|
|
|
|
|
17
|
|
|
search_space = { |
|
18
|
|
|
"x1": list(np.arange(-15, 15, 1)), |
|
19
|
|
|
} |
|
20
|
|
|
|
|
21
|
|
|
def constraint_1(para): |
|
22
|
|
|
print(" para", para) |
|
23
|
|
|
|
|
24
|
|
|
return para["x1"] > -5 |
|
25
|
|
|
|
|
26
|
|
|
constraints_list = [constraint_1] |
|
27
|
|
|
|
|
28
|
|
|
optimizer1 = RandomSearchOptimizer() |
|
29
|
|
|
optimizer2 = HillClimbingOptimizer() |
|
30
|
|
|
|
|
31
|
|
|
opt_strat = CustomOptimizationStrategy() |
|
32
|
|
|
opt_strat.add_optimizer(optimizer1, duration=0.7) |
|
33
|
|
|
opt_strat.add_optimizer(optimizer2, duration=0.3) |
|
34
|
|
|
|
|
35
|
|
|
hyper = Hyperactive() |
|
36
|
|
|
hyper.add_search( |
|
37
|
|
|
objective_function, |
|
38
|
|
|
search_space, |
|
39
|
|
|
n_iter=50, |
|
40
|
|
|
constraints=constraints_list, |
|
41
|
|
|
optimizer=opt_strat, |
|
42
|
|
|
) |
|
43
|
|
|
hyper.run() |
|
44
|
|
|
|
|
45
|
|
|
search_data = hyper.search_data(objective_function) |
|
46
|
|
|
x0_values = search_data["x1"].values |
|
47
|
|
|
|
|
48
|
|
|
print("\n search_data \n", search_data, "\n") |
|
49
|
|
|
|
|
50
|
|
|
assert np.all(x0_values > -5) |
|
51
|
|
|
|
|
52
|
|
|
|
|
53
|
|
|
def test_constr_opt_1(): |
|
54
|
|
|
"""Test constrained optimization with multi-dimensional search space.""" |
|
55
|
|
|
|
|
56
|
|
|
def objective_function(para): |
|
57
|
|
|
score = -(para["x1"] * para["x1"] + para["x2"] * para["x2"]) |
|
58
|
|
|
return score |
|
59
|
|
|
|
|
60
|
|
|
search_space = { |
|
61
|
|
|
"x1": list(np.arange(-10, 10, 1)), |
|
62
|
|
|
"x2": list(np.arange(-10, 10, 1)), |
|
63
|
|
|
} |
|
64
|
|
|
|
|
65
|
|
|
def constraint_1(para): |
|
66
|
|
|
return para["x1"] > -5 |
|
67
|
|
|
|
|
68
|
|
|
constraints_list = [constraint_1] |
|
69
|
|
|
|
|
70
|
|
|
optimizer1 = RandomSearchOptimizer() |
|
71
|
|
|
optimizer2 = HillClimbingOptimizer() |
|
72
|
|
|
|
|
73
|
|
|
opt_strat = CustomOptimizationStrategy() |
|
74
|
|
|
opt_strat.add_optimizer(optimizer1, duration=0.7) |
|
75
|
|
|
opt_strat.add_optimizer(optimizer2, duration=0.3) |
|
76
|
|
|
|
|
77
|
|
|
hyper = Hyperactive() |
|
78
|
|
|
hyper.add_search( |
|
79
|
|
|
objective_function, |
|
80
|
|
|
search_space, |
|
81
|
|
|
n_iter=50, |
|
82
|
|
|
constraints=constraints_list, |
|
83
|
|
|
optimizer=opt_strat, |
|
84
|
|
|
) |
|
85
|
|
|
hyper.run() |
|
86
|
|
|
|
|
87
|
|
|
search_data = hyper.search_data(objective_function) |
|
88
|
|
|
x0_values = search_data["x1"].values |
|
89
|
|
|
|
|
90
|
|
|
print("\n search_data \n", search_data, "\n") |
|
91
|
|
|
|
|
92
|
|
|
assert np.all(x0_values > -5) |
|
93
|
|
|
|
|
94
|
|
|
|
|
95
|
|
|
def test_constr_opt_2(): |
|
96
|
|
|
"""Test constrained optimization with multiple constraints.""" |
|
97
|
|
|
n_iter = 50 |
|
98
|
|
|
|
|
99
|
|
|
def objective_function(para): |
|
100
|
|
|
score = -para["x1"] * para["x1"] |
|
101
|
|
|
return score |
|
102
|
|
|
|
|
103
|
|
|
search_space = { |
|
104
|
|
|
"x1": list(np.arange(-10, 10, 0.1)), |
|
105
|
|
|
} |
|
106
|
|
|
|
|
107
|
|
|
def constraint_1(para): |
|
108
|
|
|
return para["x1"] > -5 |
|
109
|
|
|
|
|
110
|
|
|
def constraint_2(para): |
|
111
|
|
|
return para["x1"] < 5 |
|
112
|
|
|
|
|
113
|
|
|
constraints_list = [constraint_1, constraint_2] |
|
114
|
|
|
|
|
115
|
|
|
optimizer1 = RandomSearchOptimizer() |
|
116
|
|
|
optimizer2 = HillClimbingOptimizer() |
|
117
|
|
|
|
|
118
|
|
|
opt_strat = CustomOptimizationStrategy() |
|
119
|
|
|
opt_strat.add_optimizer(optimizer1, duration=0.7) |
|
120
|
|
|
opt_strat.add_optimizer(optimizer2, duration=0.3) |
|
121
|
|
|
|
|
122
|
|
|
hyper = Hyperactive() |
|
123
|
|
|
hyper.add_search( |
|
124
|
|
|
objective_function, |
|
125
|
|
|
search_space, |
|
126
|
|
|
n_iter=50, |
|
127
|
|
|
constraints=constraints_list, |
|
128
|
|
|
optimizer=opt_strat, |
|
129
|
|
|
) |
|
130
|
|
|
hyper.run() |
|
131
|
|
|
|
|
132
|
|
|
search_data = hyper.search_data(objective_function) |
|
133
|
|
|
x0_values = search_data["x1"].values |
|
134
|
|
|
|
|
135
|
|
|
print("\n search_data \n", search_data, "\n") |
|
136
|
|
|
|
|
137
|
|
|
assert np.all(x0_values > -5) |
|
138
|
|
|
assert np.all(x0_values < 5) |
|
139
|
|
|
|
|
140
|
|
|
n_current_positions = 0 |
|
141
|
|
|
n_current_scores = 0 |
|
142
|
|
|
|
|
143
|
|
|
n_best_positions = 0 |
|
144
|
|
|
n_best_scores = 0 |
|
145
|
|
|
|
|
146
|
|
|
for optimizer_setup in list(hyper.opt_pros.values())[0].optimizer_setup_l: |
|
147
|
|
|
optimizer = optimizer_setup["optimizer"].gfo_optimizer |
|
148
|
|
|
duration = optimizer_setup["duration"] |
|
149
|
|
|
|
|
150
|
|
|
duration_sum = 1 |
|
151
|
|
|
n_iter_expected = round(n_iter * duration / duration_sum) |
|
152
|
|
|
|
|
153
|
|
|
n_current_positions = n_current_positions + len(optimizer.pos_current_list) |
|
154
|
|
|
n_current_scores = n_current_scores + len(optimizer.score_current_list) |
|
155
|
|
|
|
|
156
|
|
|
n_best_positions = n_best_positions + len(optimizer.pos_best_list) |
|
157
|
|
|
n_best_scores = n_best_scores + len(optimizer.score_best_list) |
|
158
|
|
|
|
|
159
|
|
|
print("\n optimizer", optimizer) |
|
160
|
|
|
print(" n_new_positions", optimizer.pos_new_list, len(optimizer.pos_new_list)) |
|
161
|
|
|
print(" n_new_scores", optimizer.score_new_list, len(optimizer.score_new_list)) |
|
162
|
|
|
print(" n_iter_expected", n_iter_expected) |
|
163
|
|
|
|
|
164
|
|
|
assert len(optimizer.pos_new_list) == n_iter_expected |
|
165
|
|
|
assert len(optimizer.score_new_list) == n_iter_expected |
|
166
|
|
|
|
SimonBlanke /
Hyperactive
