1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
| import numpy as np
import matplotlib.pyplot as plt
from tqdm import tqdm
class GA(object):
def __init__(self, gap):
self.pop = None
self.max_iter = gap['max_iter']
self.pc = 0.8
self.pm = 0.005
self.pop_n = gap['pop_n']
self.pop_d = gap['pop_d']
self.len = gap['len']
self.lb = gap['lb']
self.ub = gap['ub']
self.mm = gap['mm']
self.fan = gap['gf']
self.sf = gap['sf']
self.best_i = []
self.best_if = []
def init(self):
n = self.pop_n
d = self.pop_d
l = self.len
pop = np.ones((n, d, l))
for n_ in range(n):
for d_ in range(d):
pop[n_, d_] = np.random.randint(0, 2, l, dtype=int)
self.pop = pop
return self.pop
@staticmethod
def b2d(pop_b, lb, ub):
if np.ndim(pop_b) == 2:
pop_b = pop_b[np.newaxis, :]
elif np.ndim(pop_b) == 1:
pop_b = pop_b[np.newaxis, :]
pop_b = pop_b[np.newaxis, :]
n = pop_b.shape[0]
d = pop_b.shape[1]
l = pop_b.shape[2]
temp_1 = np.zeros((n, d))
for n_ in range(n):
for d_ in range(d):
for l_ in range(l):
temp_1[n_, d_] += pop_b[n_, d_, l_]*np.power(2, l_)
temp_2 = (ub - lb)/(np.power(2, l) - 1)
temp_1 = lb + temp_1*temp_2
return temp_1
def roulette(self, pop, fitness):
sort_idx = None
sort_pop = None
sort_fit = None
n = self.pop_n
if self.mm == 'max':
sort_idx = np.argsort(fitness)
sort_pop = pop[sort_idx]
sort_fit = fitness[sort_idx]
elif self.mm == 'min':
sort_idx = np.argsort(-fitness)
sort_pop = pop[sort_idx]
sort_fit = 1/fitness[sort_idx]
fit_sum = np.sum(sort_fit)
rou = np.zeros(pop.shape[0])
rou[0] = sort_fit[0]/fit_sum
for i in range(1, pop.shape[0]):
rou[i] = rou[i-1] + sort_fit[i]/fit_sum
rou[-1] = 1
new_idx = []
for i in range(n):
rand = np.random.uniform()
for j in range(pop.shape[0]):
if rou[j] >= rand:
new_idx.append(j)
break
new_pop = sort_pop[new_idx]
new_fit = fitness[sort_idx][new_idx]
new_idx = None
if self.mm == 'max':
new_idx = np.argsort(new_fit)
elif self.mm == 'min':
new_idx = np.argsort(-new_fit)
new_pop = new_pop[new_idx]
new_fit = new_fit[new_idx]
self.best_i.append(self.b2d(new_pop[-1], self.lb, self.ub))
self.best_if.append(new_fit[-1])
return new_pop, new_fit
def best(self, pop, fitness):
new_pop = None
new_fit = None
if self.mm == 'min':
idx = np.argsort(-fitness)
new_pop = pop[idx][:(self.pop_n+1):-1]
new_fit = fitness[idx][:(self.pop_n+1):-1]
if self.mm == 'max':
idx = np.argsort(fitness)
new_pop = pop[idx][:(self.pop_n+1):-1]
new_fit = fitness[idx][:(self.pop_n+1):-1]
self.best_i.append(self.b2d(new_pop[-1], self.lb, self.ub))
self.best_if.append(new_fit[-1])
return new_pop, new_fit
def select(self, pop, fitness):
if self.sf == 'roulette':
return self.roulette(pop, fitness)
elif self.sf == 'best':
return self.best(pop, fitness)
def crossover(self, pop):
n = pop.shape[0]
d = pop.shape[1]
new_pop = []
for i in range(n):
idx_1 = np.arange(n)
rand = np.random.uniform()
if rand <= self.pc:
idx_1 = np.delete(idx_1, i)
k = np.random.choice(idx_1)
idx_2 = np.arange(1, self.len)
ran_p = np.random.choice(idx_2, 2)
cp1, cp2 = ran_p[0], ran_p[1]
cp1, cp2 = np.minimum(cp1, cp2), np.maximum(cp1, cp2)
new_pop1, new_pop2 = [], []
for j in range(d):
temp1, temp2 = [], []
temp1.extend(pop[i, j][0:cp1])
temp1.extend(pop[k, j][cp1:cp2])
temp1.extend(pop[i, j][cp2:])
temp2.extend(pop[k, j][0:cp1])
temp2.extend(pop[i, j][cp1:cp2])
temp2.extend(pop[k, j][cp2:])
new_pop1.append(temp1)
new_pop2.append(temp2)
new_pop.append(new_pop1)
new_pop.append(new_pop2)
new_pop = np.array(new_pop)
return new_pop
def mutation(self, pop):
n = pop.shape[0]
d = pop.shape[1]
new_pop = pop.copy()
for i in range(n):
for j in range(d):
rand = np.random.uniform()
if rand <= self.pm:
cp = np.random.randint(0, self.len)
new_pop[i, j, cp] = 1 - pop[i, j, cp]
return new_pop
def run(self):
pop = self.init()
for g in tqdm(range(self.max_iter), desc="GA_Processing"):
pop1 = self.crossover(pop)
pop2 = self.mutation(pop1)
pop2 = np.unique(pop2, axis=0)
pop3 = self.b2d(pop2, self.lb, self.ub)
fit = self.fan(pop3)
pop, fit = self.select(pop2, fit)
return self.best_i, self.best_if
def get_fitness(pop):
res = np.sum(pop**2, axis=1)
return res
if __name__ == "__main__":
dim = 10
ga_init = {
'max_iter': 100,
'pop_n': 100,
'pop_d': dim,
'len': 10,
'lb': -5.12 * np.ones(dim),
'ub': 5.12 * np.ones(dim),
'mm': ['min', 'max'][0],
'gf': get_fitness,
'sf': ['roulette', 'best'][0],
}
ga = GA(ga_init)
ga.run()
x = np.arange(ga_init['max_iter'])
y = ga.best_if
plt.figure()
plt.plot(x, y)
plt.show()
|
