|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
import random |
|
7
|
|
|
import numpy as np |
|
8
|
|
|
|
|
9
|
|
|
from ..search_tracker import SearchTracker |
|
10
|
|
|
from ...converter import Converter |
|
11
|
|
|
from ...results_manager import ResultsManager |
|
12
|
|
|
from ...optimizers.base_optimizer import get_n_inits |
|
13
|
|
|
from ...init_positions import Initializer |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
View Code Duplication |
def set_random_seed(nth_process, random_state): |
|
|
|
|
|
|
17
|
|
|
""" |
|
18
|
|
|
Sets the random seed separately for each thread |
|
19
|
|
|
(to avoid getting the same results in each thread) |
|
20
|
|
|
""" |
|
21
|
|
|
if nth_process is None: |
|
22
|
|
|
nth_process = 0 |
|
23
|
|
|
|
|
24
|
|
|
if random_state is None: |
|
25
|
|
|
random_state = np.random.randint(0, high=2 ** 31 - 2, dtype=np.int64) |
|
26
|
|
|
|
|
27
|
|
|
random.seed(random_state + nth_process) |
|
28
|
|
|
np.random.seed(random_state + nth_process) |
|
29
|
|
|
|
|
30
|
|
|
|
|
31
|
|
|
class BasePopulationOptimizer: |
|
32
|
|
|
def __init__( |
|
33
|
|
|
self, |
|
34
|
|
|
search_space, |
|
35
|
|
|
initialize={"grid": 4, "random": 2, "vertices": 4}, |
|
36
|
|
|
random_state=None, |
|
37
|
|
|
rand_rest_p=0, |
|
38
|
|
|
nth_process=None, |
|
39
|
|
|
): |
|
40
|
|
|
super().__init__() |
|
41
|
|
|
self.conv = Converter(search_space) |
|
42
|
|
|
self.results_mang = ResultsManager(self.conv) |
|
43
|
|
|
self.initialize = initialize |
|
44
|
|
|
self.random_state = random_state |
|
45
|
|
|
self.rand_rest_p = rand_rest_p |
|
46
|
|
|
self.nth_process = nth_process |
|
47
|
|
|
|
|
48
|
|
|
self.eval_times = [] |
|
49
|
|
|
self.iter_times = [] |
|
50
|
|
|
|
|
51
|
|
|
set_random_seed(nth_process, random_state) |
|
52
|
|
|
|
|
53
|
|
|
# get init positions |
|
54
|
|
|
init = Initializer(self.conv) |
|
55
|
|
|
self.init_positions = init.set_pos(self.initialize) |
|
56
|
|
|
|
|
57
|
|
|
def _iterations(self, positioners): |
|
58
|
|
|
nth_iter = 0 |
|
59
|
|
|
for p in positioners: |
|
60
|
|
|
nth_iter = nth_iter + len(p.pos_new_list) |
|
61
|
|
|
|
|
62
|
|
|
return nth_iter |
|
63
|
|
|
|
|
64
|
|
|
def _create_population(self, Optimizer): |
|
65
|
|
|
if isinstance(self.population, int): |
|
66
|
|
|
population = [] |
|
67
|
|
|
for pop_ in range(self.population): |
|
68
|
|
|
population.append( |
|
69
|
|
|
Optimizer(self.conv.search_space, rand_rest_p=self.rand_rest_p) |
|
70
|
|
|
) |
|
71
|
|
|
else: |
|
72
|
|
|
population = self.population |
|
73
|
|
|
|
|
74
|
|
|
n_inits = get_n_inits(self.initialize) |
|
75
|
|
|
|
|
76
|
|
|
if n_inits < len(population): |
|
77
|
|
|
print("\n Warning: Not enough initial positions for population size") |
|
78
|
|
|
print(" Population size is reduced to", n_inits) |
|
79
|
|
|
population = population[:n_inits] |
|
80
|
|
|
|
|
81
|
|
|
return population |
|
82
|
|
|
|
|
83
|
|
|
def finish_initialization(self): |
|
84
|
|
|
pass |
|
85
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
Loading history...