|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import random |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
|
|
8
|
|
|
from .base_population_optimizer import BasePopulationOptimizer |
|
9
|
|
|
from ...search import Search |
|
10
|
|
|
from ._individual import Individual |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
class EvolutionStrategyOptimizer(BasePopulationOptimizer, Search): |
|
14
|
|
|
name = "Evolution Strategý" |
|
15
|
|
|
|
|
16
|
|
|
def __init__( |
|
17
|
|
|
self, |
|
18
|
|
|
*args, |
|
19
|
|
|
population=10, |
|
20
|
|
|
mutation_rate=0.7, |
|
21
|
|
|
crossover_rate=0.3, |
|
22
|
|
|
**kwargs, |
|
23
|
|
|
): |
|
24
|
|
|
super().__init__(*args, **kwargs) |
|
25
|
|
|
|
|
26
|
|
|
self.population = population |
|
27
|
|
|
self.mutation_rate = mutation_rate |
|
28
|
|
|
self.crossover_rate = crossover_rate |
|
29
|
|
|
|
|
30
|
|
|
self.individuals = self._create_population(Individual) |
|
31
|
|
|
self.optimizers = self.individuals |
|
32
|
|
|
|
|
33
|
|
|
def _random_cross(self, array_list): |
|
34
|
|
|
n_arrays = len(array_list) |
|
35
|
|
|
size = array_list[0].size |
|
36
|
|
|
|
|
37
|
|
|
choice = [True, False] |
|
38
|
|
|
if size > 2: |
|
39
|
|
|
add_choice = np.random.randint(n_arrays, size=size - 2).astype(bool) |
|
40
|
|
|
choice += list(add_choice) |
|
41
|
|
|
|
|
42
|
|
|
cross_array = np.choose(choice, array_list) |
|
43
|
|
|
return cross_array |
|
44
|
|
|
|
|
45
|
|
|
def _sort_best(self): |
|
46
|
|
|
scores_list = [] |
|
47
|
|
|
for ind in self.individuals: |
|
48
|
|
|
scores_list.append(ind.score_current) |
|
49
|
|
|
|
|
50
|
|
|
scores_np = np.array(scores_list) |
|
51
|
|
|
idx_sorted_ind = list(scores_np.argsort()[::-1]) |
|
52
|
|
|
|
|
53
|
|
|
return [self.individuals[idx] for idx in idx_sorted_ind] |
|
54
|
|
|
|
|
55
|
|
|
def _cross(self): |
|
56
|
|
|
if len(self.individuals) > 2: |
|
57
|
|
|
rnd_int2 = random.choice( |
|
58
|
|
|
[i for i in range(0, self.n_ind - 1) if i not in [self.rnd_int]] |
|
59
|
|
|
) |
|
60
|
|
|
else: |
|
61
|
|
|
rnd_int2 = random.choice( |
|
62
|
|
|
[i for i in range(0, self.n_ind) if i not in [self.rnd_int]] |
|
63
|
|
|
) |
|
64
|
|
|
|
|
65
|
|
|
p_sec = self.ind_sorted[rnd_int2] |
|
66
|
|
|
p_worst = self.ind_sorted[-1] |
|
67
|
|
|
|
|
68
|
|
|
two_best_pos = [self.p_current.pos_current, p_sec.pos_current] |
|
69
|
|
|
pos_new = self._random_cross(two_best_pos) |
|
70
|
|
|
|
|
71
|
|
|
self.p_current = p_worst |
|
72
|
|
|
p_worst.pos_new = pos_new |
|
73
|
|
|
|
|
74
|
|
|
return pos_new |
|
75
|
|
|
|
|
76
|
|
|
def init_pos(self, pos): |
|
77
|
|
|
nth_pop = self.nth_iter % len(self.individuals) |
|
78
|
|
|
|
|
79
|
|
|
self.p_current = self.individuals[nth_pop] |
|
80
|
|
|
self.p_current.init_pos(pos) |
|
81
|
|
|
|
|
82
|
|
|
def iterate(self): |
|
83
|
|
|
self.n_ind = len(self.individuals) |
|
84
|
|
|
|
|
85
|
|
|
if self.n_ind == 1: |
|
86
|
|
|
self.p_current = self.individuals[0] |
|
87
|
|
|
return self.p_current.iterate() |
|
88
|
|
|
|
|
89
|
|
|
self.ind_sorted = self._sort_best() |
|
90
|
|
|
self.rnd_int = random.randint(0, len(self.ind_sorted) - 1) |
|
91
|
|
|
self.p_current = self.ind_sorted[self.rnd_int] |
|
92
|
|
|
|
|
93
|
|
|
total_rate = self.mutation_rate + self.crossover_rate |
|
94
|
|
|
rand = np.random.uniform(low=0, high=total_rate) |
|
95
|
|
|
|
|
96
|
|
|
if rand <= self.mutation_rate: |
|
97
|
|
|
return self.p_current.iterate() |
|
98
|
|
|
else: |
|
99
|
|
|
return self._cross() |
|
100
|
|
|
|
|
101
|
|
|
def evaluate(self, score_new): |
|
102
|
|
|
self.p_current.evaluate(score_new) |
|
103
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
