|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import pytest |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
|
|
8
|
|
|
from surfaces.test_functions.mathematical import SphereFunction |
|
9
|
|
|
|
|
10
|
|
|
from gradient_free_optimizers import StochasticHillClimbingOptimizer |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
sphere_function = SphereFunction(n_dim=2) |
|
14
|
|
|
objective_function = sphere_function.objective_function |
|
15
|
|
|
search_space = sphere_function.search_space() |
|
16
|
|
|
|
|
17
|
|
|
|
|
18
|
|
|
n_iter_para = ( |
|
19
|
|
|
"n_iter", |
|
20
|
|
|
[ |
|
21
|
|
|
(300), |
|
22
|
|
|
(500), |
|
23
|
|
|
(1000), |
|
24
|
|
|
], |
|
25
|
|
|
) |
|
26
|
|
|
|
|
27
|
|
|
|
|
28
|
|
|
@pytest.mark.parametrize(*n_iter_para) |
|
29
|
|
|
def test_p_accept(n_iter): |
|
30
|
|
|
p_accept_low = 0.5 |
|
31
|
|
|
p_accept_mid = 0.75 |
|
32
|
|
|
p_accept_high = 1 |
|
33
|
|
|
|
|
34
|
|
|
epsilon = 1 / np.inf |
|
35
|
|
|
|
|
36
|
|
|
opt = StochasticHillClimbingOptimizer( |
|
37
|
|
|
search_space, |
|
38
|
|
|
p_accept=p_accept_low, |
|
39
|
|
|
epsilon=epsilon, |
|
40
|
|
|
initialize={"random": 1}, |
|
41
|
|
|
) |
|
42
|
|
|
opt.search(objective_function, n_iter=n_iter) |
|
43
|
|
|
n_transitions_low = opt.n_transitions |
|
44
|
|
|
|
|
45
|
|
|
opt = StochasticHillClimbingOptimizer( |
|
46
|
|
|
search_space, |
|
47
|
|
|
p_accept=p_accept_mid, |
|
48
|
|
|
epsilon=epsilon, |
|
49
|
|
|
initialize={"random": 1}, |
|
50
|
|
|
) |
|
51
|
|
|
opt.search(objective_function, n_iter=n_iter) |
|
52
|
|
|
n_transitions_mid = opt.n_transitions |
|
53
|
|
|
|
|
54
|
|
|
opt = StochasticHillClimbingOptimizer( |
|
55
|
|
|
search_space, |
|
56
|
|
|
p_accept=p_accept_high, |
|
57
|
|
|
epsilon=epsilon, |
|
58
|
|
|
initialize={"random": 1}, |
|
59
|
|
|
) |
|
60
|
|
|
opt.search(objective_function, n_iter=n_iter) |
|
61
|
|
|
n_transitions_high = opt.n_transitions |
|
62
|
|
|
|
|
63
|
|
|
print("\n n_transitions_low", n_transitions_low) |
|
64
|
|
|
print("\n n_transitions_mid", n_transitions_mid) |
|
65
|
|
|
print("\n n_transitions_high", n_transitions_high) |
|
66
|
|
|
|
|
67
|
|
|
lower_bound = int(n_iter * p_accept_low) |
|
68
|
|
|
lower_bound -= lower_bound * 0.1 |
|
69
|
|
|
higher_bound = n_iter |
|
70
|
|
|
|
|
71
|
|
|
assert ( |
|
72
|
|
|
lower_bound |
|
73
|
|
|
< n_transitions_low |
|
74
|
|
|
< n_transitions_mid |
|
75
|
|
|
< n_transitions_high |
|
76
|
|
|
< higher_bound |
|
77
|
|
|
) |
|
78
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
