-
Notifications
You must be signed in to change notification settings - Fork 0
/
algorithm-ag.py
232 lines (170 loc) · 8.37 KB
/
algorithm-ag.py
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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
import igraph as ig
import numpy as np
from sklearn.metrics import euclidean_distances
from sklearn.model_selection import train_test_split
def pagerank(graph):
return ig.Graph.pagerank(graph)
def create_representation(X_train, y_train, q_value):
euclidean_dist = euclidean_distances(X_train)
np.fill_diagonal(euclidean_dist, np.inf)
neighbors = np.argsort(euclidean_dist, axis=1)[:, :q_value]
mask = np.zeros((len(X_train), q_value)).astype(int)
for i in range(len(neighbors)):
mask[i] = (y_train[neighbors[i]] == y_train[i])
return neighbors, mask
def generate_ind(genome_size, q_value):
if q_value == 5:
return [[np.random.randint(2), np.random.randint(2), np.random.randint(2), np.random.randint(2), np.random.randint(2)] for _ in range(genome_size)]
return [[np.random.randint(2), np.random.randint(2), np.random.randint(2)] for _ in range(genome_size)]
def create_population(genome_size, q_value, TP, CR):
Pop = np.zeros((TP + CR, genome_size, q_value)).astype(int)
Fit = np.zeros(TP + CR)
Pop[:TP] = [generate_ind(genome_size, q_value) for _ in range(TP)]
return Pop, Fit
def validate_pop(pop, actual_classes):
# os que nao percecem a claase nao recebem a conecao
not_connect = np.argwhere(actual_classes == 0)
for value in not_connect:
pop[:, value[0]][:, value[1]] = 0
def generate_graph(X_train, individuo, neighbors, q_value):
euclidean_dist = euclidean_distances(X_train)
sources = np.arange(0, len(individuo))
graph = ig.Graph(n=len(sources), directed=True)
for i in sources:
for ids_dest in range(q_value):
# decide estocasticamente se deve haver a conexao
if individuo[i][ids_dest] == 1:
graph.add_edge(i, neighbors[i][ids_dest], weight=euclidean_dist[i, neighbors[i][ids_dest]])
return graph
# requer a passagem de um grafo ponderado como parametro
def efficiency_flow(graph):
global edge_sources
eff = np.zeros(graph.vcount())
try:
edge_sources = np.asarray(graph.get_edgelist()).T[0]
except:
print("An exception occurred")
for j, i in enumerate(edge_sources):
eff[i] += graph.es['weight'][j]
count = np.bincount(edge_sources)
for i in range(len(count)):
if count[i] != 0:
eff[i] = eff[i] / count[i]
comps = ig.Graph.components(graph, mode=WEAK)
for i in range(len(comps)):
comp_ids = np.array(comps[i])
if len(comp_ids) > 1:
eff[comp_ids] = np.sum(eff[comp_ids]) / (1. * len(comp_ids))
return eff
def evaluate_config(X_train, y_train, X_val, y_val, pop, neighbor, actual_classes, alpha, q_value):
fit = []
graphs = []
# os que nao forem da msm classe nao conecta
validate_pop(pop, actual_classes)
for individual in pop:
graph = generate_graph(X_train, individual, neighbor, q_value)
graphs.append(graph)
# se gerou um grafo totalmente desconexo
if len(graph.get_edgelist()) == 0:
fit = np.append(fit, 0)
else:
eff = efficiency_flow(graph)
fdist = euclidean_distances(X_val, X_train)
I = np.asarray(pagerank(graph))
for z1, a in enumerate(alpha):
prob_class = np.zeros((len(X_val), len(np.unique(y_val))))
for i in range(len(X_val)):
f = eff[i] * a - fdist[i, :]
ids = np.where(f >= 0.)[0]
if len(ids) == 0:
ids = np.where(f == max(f))[0]
for j in ids:
prob_class[i, y_train[j]] += I[j]
predicted = np.argmax(prob_class, axis=1)
ac = np.mean(predicted == y_val)
fit = np.append(fit, ac)
return fit, graphs
def tournament(Fit, CR, TP, TOURNAMENT_SIZE):
parents = [do_tournament(Fit, TP, TOURNAMENT_SIZE) for _ in range(0, CR, 2)]
return np.array(parents).reshape(1, -1)[0]
def roulette(Fit, TP, CR):
fit_invertido = (max(Fit[:TP]) + 1) - Fit[:TP]
max_v = sum(fit_invertido).astype(np.float64)
parents = np.random.choice(TP, CR, p=(None if max_v == 0 else fit_invertido / max_v))
return parents
def do_tournament(Fit, TP, TOURNAMENT_SIZE):
random_parents = np.random.randint(TP, size=TOURNAMENT_SIZE)
parent1 = random_parents[np.argmax([Fit[random_parents]])]
random_parents = np.random.randint(TP, size=TOURNAMENT_SIZE)
parent2 = random_parents[np.argmax([Fit[random_parents]])]
return parent1, parent2
def apply_mutation(pop, genome_size, q_value, TP, CR, PMUT):
ind_to_mutate = np.random.randint(low=TP, high=TP + CR, size=int(TP * PMUT))
for individual in ind_to_mutate:
genes = np.random.choice(genome_size, size=int(genome_size * PMUT))
for gene in genes:
# mutando co certeza
for qi in range(q_value):
pop[individual][gene][qi] = 1 if genes[0] == 0 else 0
def ordered_reinsertion(Pop, Fit, TP):
aux_pop = np.zeros(Pop.shape).astype(int)
fit_sorted = np.argsort(-Fit)[:TP]
aux_pop[:TP] = Pop[fit_sorted]
return aux_pop
def pure_reinsertion(Pop, Fit, TP, CR):
aux_pop = np.zeros(Pop.shape).astype(int)
fit_sorted = np.argsort(Fit[:TP])[:TP - CR]
aux_pop[:TP - CR] = np.copy(Pop[fit_sorted])
aux_pop[TP - CR:TP] = np.copy(Pop[TP:TP + CR])
return aux_pop
def two_points_crossover(parent1, parent2, genome_size):
gene1, gene2 = np.random.choice(genome_size, 2, replace=False)
new_child1 = np.copy(parent1)
new_child2 = np.copy(parent2)
aux = np.copy(parent1)
if gene1 > gene2:
gene1, gene2 = gene2, gene1
new_child1[range(gene1, gene2)] = np.copy(new_child2[range(gene1, gene2)])
new_child2[range(gene1, gene2)] = np.copy(aux[range(gene1, gene2)])
return new_child1, new_child2
def mask_crossover(parent1, parent2, genome_size):
mask = np.random.randint(low=0, high=2, size=genome_size)
new_child1 = np.copy(parent1)
new_child2 = np.copy(parent2)
new_child1[np.where(mask == 1)] = parent2[np.where(mask == 1)]
new_child2[np.where(mask == 0)] = parent1[np.where(mask == 0)]
return new_child1, new_child2
def one_point_crossover(parent1, parent2, genome_size):
gene = np.random.choice(genome_size, replace=False)
new_child1 = np.copy(parent1)
new_child2 = np.copy(parent2)
aux = np.copy(parent1)
new_child1[range(0, gene)] = np.copy(new_child2[range(0, gene)])
new_child2[range(gene, genome_size)] = np.copy(aux[range(gene, genome_size)])
return new_child1, new_child2
def alg(config, X, y, random, alpha, q_value, EXECUTIONS, TP, CR, GEN, PMUT):
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=np.random.seed(random * 100))
X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.25, random_state=np.random.seed(random * 100))
neighbors, actual_classes = create_representation(X_train, y_train, q_value)
genome_size = X_train.shape[0]
for execution in range(EXECUTIONS):
pop, fit = create_population(genome_size, q_value, TP, CR)
# testando a opcao de fazer o cut somente qndo avalia a populacao
fit[:TP], graphs = evaluate_config(X_train, y_train, X_val, y_val, pop[:TP], neighbors, actual_classes, alpha, q_value)
for generation in range(GEN):
parents = config[3](fit[:TP])
for i in range(0, CR - 1, 2):
# pensar como fazer o crossover d forma simples
pop[TP + i], pop[TP + i + 1] = config[5](pop[parents[i]], pop[parents[i + 1]], genome_size)
# pensar na mutacao
apply_mutation(pop, genome_size, q_value, TP, CR, PMUT)
# avaliacao
fit, graphs = evaluate_config(X_train, y_train, X_val, y_val, pop, neighbors, actual_classes, alpha, q_value)
# Re-insercao
pop = np.copy(config[4](pop, fit))
fit, graphs = evaluate_config(X_train, y_train, X_val, y_val, pop[:TP], neighbors, actual_classes, alpha, q_value)
acc_val = fit[np.argmax(fit[:TP])]
acc_teste, graph_final = evaluate_config(X_train, y_train, X_test, y_test,
pop[np.argmax(fit[:TP])].reshape(1, len(pop[np.argmax(fit[:TP])]), -1), neighbors,
actual_classes, alpha, q_value)
return acc_val, acc_teste[0], graph_final[0]