|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import math |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
|
|
8
|
|
|
from ..core_optimizer import CoreOptimizer |
|
9
|
|
|
|
|
10
|
|
|
|
|
11
|
|
|
def split(positions_l, population): |
|
12
|
|
|
div_int = math.ceil(len(positions_l) / population) |
|
13
|
|
|
dist_init_positions = [] |
|
14
|
|
|
|
|
15
|
|
|
for nth_indiv in range(population): |
|
16
|
|
|
indiv_pos = [] |
|
17
|
|
|
for nth_indiv_pos in range(div_int): |
|
18
|
|
|
idx = nth_indiv + nth_indiv_pos * population |
|
19
|
|
|
if idx < len(positions_l): |
|
20
|
|
|
indiv_pos.append(positions_l[idx]) |
|
21
|
|
|
|
|
22
|
|
|
dist_init_positions.append(indiv_pos) |
|
23
|
|
|
|
|
24
|
|
|
return dist_init_positions |
|
25
|
|
|
|
|
26
|
|
|
|
|
27
|
|
|
class BasePopulationOptimizer(CoreOptimizer): |
|
28
|
|
|
def __init__(self, *args, **kwargs): |
|
29
|
|
|
super().__init__(*args, **kwargs) |
|
30
|
|
|
|
|
31
|
|
|
self.eval_times = [] |
|
32
|
|
|
self.iter_times = [] |
|
33
|
|
|
|
|
34
|
|
|
self.init_done = False |
|
35
|
|
|
|
|
36
|
|
|
def _iterations(self, positioners): |
|
37
|
|
|
nth_iter = 0 |
|
38
|
|
|
for p in positioners: |
|
39
|
|
|
nth_iter = nth_iter + len(p.pos_new_list) |
|
40
|
|
|
|
|
41
|
|
|
return nth_iter |
|
42
|
|
|
|
|
43
|
|
|
def sort_pop_best_score(self): |
|
44
|
|
|
scores_list = [] |
|
45
|
|
|
for _p_ in self.optimizers: |
|
46
|
|
|
scores_list.append(_p_.score_current) |
|
47
|
|
|
|
|
48
|
|
|
scores_np = np.array(scores_list) |
|
49
|
|
|
idx_sorted_ind = list(scores_np.argsort()[::-1]) |
|
50
|
|
|
|
|
51
|
|
|
self.pop_sorted = [self.optimizers[i] for i in idx_sorted_ind] |
|
52
|
|
|
|
|
53
|
|
|
def _create_population(self, Optimizer): |
|
54
|
|
|
if isinstance(self.population, int): |
|
55
|
|
|
pop_size = self.population |
|
56
|
|
|
else: |
|
57
|
|
|
pop_size = len(self.population) |
|
58
|
|
|
diff_init = pop_size - self.init.n_inits |
|
59
|
|
|
|
|
60
|
|
|
if diff_init > 0: |
|
61
|
|
|
self.init.add_n_random_init_pos(diff_init) |
|
62
|
|
|
|
|
63
|
|
|
if isinstance(self.population, int): |
|
64
|
|
|
distributed_init_positions = split( |
|
65
|
|
|
self.init.init_positions_l, self.population |
|
66
|
|
|
) |
|
67
|
|
|
|
|
68
|
|
|
population = [] |
|
69
|
|
|
for init_positions in distributed_init_positions: |
|
70
|
|
|
init_values = self.conv.positions2values(init_positions) |
|
71
|
|
|
init_paras = self.conv.values2paras(init_values) |
|
72
|
|
|
|
|
73
|
|
|
population.append( |
|
74
|
|
|
Optimizer( |
|
75
|
|
|
self.conv.search_space, |
|
76
|
|
|
rand_rest_p=self.rand_rest_p, |
|
77
|
|
|
initialize={"warm_start": init_paras}, |
|
78
|
|
|
) |
|
79
|
|
|
) |
|
80
|
|
|
else: |
|
81
|
|
|
population = self.population |
|
82
|
|
|
|
|
83
|
|
|
return population |
|
84
|
|
|
|
|
85
|
|
|
@CoreOptimizer.track_new_score |
|
86
|
|
|
def evaluate_init(self, score_new): |
|
87
|
|
|
self.p_current.evaluate_init(score_new) |
|
88
|
|
|
|
|
89
|
|
|
def finish_initialization(self): |
|
90
|
|
|
self.search_state = "iter" |
|
91
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
