|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
import numpy as np |
|
7
|
|
|
import random |
|
8
|
|
|
|
|
9
|
|
|
from .particle_swarm_optimization import ParticleSwarmOptimizer |
|
10
|
|
|
from ...base_positioner import BasePositioner |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
class EvolutionStrategyOptimizer(ParticleSwarmOptimizer): |
|
14
|
|
|
def __init__(self, *args, **kwargs): |
|
15
|
|
|
super().__init__(*args, **kwargs) |
|
16
|
|
|
self.n_mutations = int(round(self._arg_.individuals * self._arg_.mutation_rate)) |
|
17
|
|
|
self.n_crossovers = int( |
|
18
|
|
|
round(self._arg_.individuals * self._arg_.crossover_rate) |
|
19
|
|
|
) |
|
20
|
|
|
|
|
21
|
|
|
def _init_individuals(self, _cand_): |
|
22
|
|
|
_p_list_ = [Individual() for _ in range(self._arg_.individuals)] |
|
23
|
|
|
for _p_ in _p_list_: |
|
24
|
|
|
_p_.pos_current = _p_.move_random(_cand_) |
|
25
|
|
|
_p_.pos_best = _p_.pos_current |
|
26
|
|
|
|
|
27
|
|
|
return _p_list_ |
|
28
|
|
|
|
|
29
|
|
|
def _mutate_individuals(self, _cand_, _p_list_, mutate_idx): |
|
30
|
|
|
_p_list_ = np.array(_p_list_) |
|
31
|
|
|
for _p_ in _p_list_[mutate_idx]: |
|
32
|
|
|
_p_.pos_new = _p_.move_climb(_cand_, _p_.pos_current) |
|
33
|
|
|
|
|
34
|
|
|
def _crossover(self, _cand_, _p_list_, cross_idx, replace_idx): |
|
35
|
|
|
_p_list_ = np.array(_p_list_) |
|
36
|
|
|
for i, _p_ in enumerate(_p_list_[replace_idx]): |
|
37
|
|
|
j = i + 1 |
|
38
|
|
|
if j == len(cross_idx): |
|
39
|
|
|
j = 0 |
|
40
|
|
|
|
|
41
|
|
|
pos_new = self._cross_two_ind( |
|
42
|
|
|
[_p_list_[cross_idx][i], _p_list_[cross_idx][j]] |
|
43
|
|
|
) |
|
44
|
|
|
|
|
45
|
|
|
_p_.pos_new = pos_new |
|
46
|
|
|
|
|
47
|
|
|
def _cross_two_ind(self, _p_list_): |
|
48
|
|
|
pos_new = [] |
|
49
|
|
|
|
|
50
|
|
|
for pos1, pos2 in zip(_p_list_[0].pos_current, _p_list_[1].pos_current): |
|
51
|
|
|
rand = random.randint(0, 1) |
|
52
|
|
|
if rand == 0: |
|
53
|
|
|
pos_new.append(pos1) |
|
54
|
|
|
else: |
|
55
|
|
|
pos_new.append(pos2) |
|
56
|
|
|
|
|
57
|
|
|
return np.array(pos_new) |
|
58
|
|
|
|
|
59
|
|
|
def _move_positioners(self, _cand_, _p_list_): |
|
60
|
|
|
idx_sorted_ind = self._rank_individuals(_p_list_) |
|
61
|
|
|
mutate_idx, cross_idx, replace_idx = self._select_individuals(idx_sorted_ind) |
|
62
|
|
|
|
|
63
|
|
|
self._mutate_individuals(_cand_, _p_list_, mutate_idx) |
|
64
|
|
|
self._crossover(_cand_, _p_list_, cross_idx, replace_idx) |
|
65
|
|
|
|
|
66
|
|
|
def _rank_individuals(self, _p_list_): |
|
67
|
|
|
scores_list = [] |
|
68
|
|
|
for _p_ in _p_list_: |
|
69
|
|
|
scores_list.append(_p_.score_current) |
|
70
|
|
|
|
|
71
|
|
|
scores_np = np.array(scores_list) |
|
72
|
|
|
idx_sorted_ind = list(scores_np.argsort()[::-1]) |
|
73
|
|
|
|
|
74
|
|
|
return idx_sorted_ind |
|
75
|
|
|
|
|
76
|
|
|
def _select_individuals(self, index_best): |
|
77
|
|
|
mutate_idx = index_best[: self.n_mutations] |
|
78
|
|
|
cross_idx = index_best[: self.n_crossovers] |
|
79
|
|
|
|
|
80
|
|
|
n = self._arg_.individuals - max(self.n_mutations, self.n_crossovers) |
|
81
|
|
|
replace_idx = index_best[-n:] |
|
82
|
|
|
|
|
83
|
|
|
return mutate_idx, cross_idx, replace_idx |
|
84
|
|
|
|
|
85
|
|
|
# use _iterate from ParticleSwarmOptimizer |
|
86
|
|
|
|
|
87
|
|
|
def _init_opt_positioner(self, _cand_, X, y): |
|
88
|
|
|
_p_list_ = self._init_individuals(_cand_) |
|
89
|
|
|
|
|
90
|
|
|
for _p_ in _p_list_: |
|
91
|
|
|
_p_.score_current = _cand_.eval_pos(_p_.pos_current, X, y) |
|
92
|
|
|
_p_.score_best = _p_.score_current |
|
93
|
|
|
|
|
94
|
|
|
return _p_list_ |
|
95
|
|
|
|
|
96
|
|
|
|
|
97
|
|
|
class Individual(BasePositioner): |
|
98
|
|
|
def __init__(self, eps=1): |
|
99
|
|
|
super().__init__(eps) |
|
100
|
|
|
|
SimonBlanke /
Hyperactive
