|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
import random |
|
7
|
|
|
|
|
8
|
|
|
import numpy as np |
|
9
|
|
|
|
|
10
|
|
|
from .base_population_optimizer import BasePopulationOptimizer |
|
11
|
|
|
from ...search import Search |
|
12
|
|
|
from ..base_optimizer import BaseOptimizer |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
class ParticleSwarmOptimizer(BasePopulationOptimizer, Search): |
|
16
|
|
|
def __init__( |
|
17
|
|
|
self, search_space, inertia=0.5, cognitive_weight=0.5, social_weight=0.5, |
|
18
|
|
|
): |
|
19
|
|
|
super().__init__(search_space) |
|
20
|
|
|
|
|
21
|
|
|
self.inertia = inertia |
|
22
|
|
|
self.cognitive_weight = cognitive_weight |
|
23
|
|
|
self.social_weight = social_weight |
|
24
|
|
|
|
|
25
|
|
|
self.particles = [] |
|
26
|
|
|
|
|
27
|
|
|
def _move_part(self, pos, velo): |
|
28
|
|
|
pos_new = (pos + velo).astype(int) |
|
29
|
|
|
# limit movement |
|
30
|
|
|
n_zeros = [0] * len(self.space_dim) |
|
31
|
|
|
self.p_current.pos_new = np.clip(pos_new, n_zeros, self.space_dim) |
|
32
|
|
|
|
|
33
|
|
|
return self.p_current.pos_new |
|
34
|
|
|
|
|
35
|
|
|
def _move_positioner(self): |
|
36
|
|
|
r1, r2 = random.random(), random.random() |
|
37
|
|
|
|
|
38
|
|
|
A = self.inertia * self.p_current.velo |
|
39
|
|
|
B = ( |
|
40
|
|
|
self.cognitive_weight |
|
41
|
|
|
* r1 |
|
42
|
|
|
* np.subtract(self.p_current.pos_best, self.p_current.pos_current) |
|
43
|
|
|
) |
|
44
|
|
|
C = ( |
|
45
|
|
|
self.social_weight |
|
46
|
|
|
* r2 |
|
47
|
|
|
* np.subtract(self.global_pos_best, self.p_current.pos_current) |
|
48
|
|
|
) |
|
49
|
|
|
|
|
50
|
|
|
new_velocity = A + B + C |
|
51
|
|
|
|
|
52
|
|
|
return self._move_part(self.p_current.pos_current, new_velocity) |
|
53
|
|
|
|
|
54
|
|
|
def _sort_best(self): |
|
55
|
|
|
scores_list = [] |
|
56
|
|
|
for _p_ in self.particles: |
|
57
|
|
|
scores_list.append(_p_.score_current) |
|
58
|
|
|
|
|
59
|
|
|
scores_np = np.array(scores_list) |
|
60
|
|
|
idx_sorted_ind = list(scores_np.argsort()[::-1]) |
|
61
|
|
|
|
|
62
|
|
|
self.p_sorted = [self.particles[i] for i in idx_sorted_ind] |
|
63
|
|
|
|
|
64
|
|
|
def init_pos(self, pos): |
|
65
|
|
|
particle = BaseOptimizer(self.space_dim) |
|
66
|
|
|
self.particles.append(particle) |
|
67
|
|
|
particle.init_pos(pos) |
|
68
|
|
|
|
|
69
|
|
|
self.p_current = particle |
|
70
|
|
|
self.p_current.velo = np.zeros(len(self.space_dim)) |
|
71
|
|
|
|
|
72
|
|
|
def iterate(self): |
|
73
|
|
|
n_iter = self._iterations(self.particles) |
|
74
|
|
|
self.p_current = self.particles[n_iter % len(self.particles)] |
|
75
|
|
|
|
|
76
|
|
|
self._sort_best() |
|
77
|
|
|
self.global_pos_best = self.p_sorted[0].pos_best |
|
78
|
|
|
pos = self._move_positioner() |
|
79
|
|
|
|
|
80
|
|
|
return pos |
|
81
|
|
|
|
|
82
|
|
|
def evaluate(self, score_new): |
|
83
|
|
|
self.p_current.score_new = score_new |
|
84
|
|
|
|
|
85
|
|
|
self.p_current._evaluate_new2current(score_new) |
|
86
|
|
|
self.p_current._evaluate_current2best() |
|
87
|
|
|
|
|
88
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
