forked from erika1203/dbnPytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
DBN.py
executable file
·188 lines (150 loc) · 6.55 KB
/
DBN.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
import torch
from torch.autograd import Variable
import torch.nn as nn
import torch.nn.functional as F
from RBM import RBM
import time
class DBN(nn.Module):
def __init__(self,
visible_units = 4, #可视层节点
hidden_units = [11,6,11,], #隐藏层节点
k = 2, #采样步数
learning_rate = 1e-3, #学习率
momentum_coefficient = 0.9, #动量系数
weight_decay = 1e-4, #权重衰减
use_gpu = False,
_activation = 'sigmoid',): #损失函数
super(DBN,self).__init__()
self.n_layers = len(hidden_units) #隐含层数
self.rbm_layers =[] #rbm
self.rbm_nodes = []
# 构建不同的RBM层
for i in range(self.n_layers ):
if i==0:
input_size = visible_units
else:
input_size = hidden_units[i-1]
rbm = RBM(visible_units = input_size,
hidden_units = hidden_units[i],
k= k,
learning_rate = learning_rate,
momentum_coefficient = momentum_coefficient,
weight_decay = weight_decay,
use_gpu=use_gpu,
_activation = _activation)
self.rbm_layers.append(rbm)
# rbm_layers = [RBM(rbn_nodes[i-1] , rbm_nodes[i],use_gpu=use_cuda) for i in range(1,len(rbm_nodes))]
self.W_rec = [nn.Parameter(self.rbm_layers[i].weight.data.clone()) for i in range(self.n_layers-1)]
self.W_gen = [nn.Parameter(self.rbm_layers[i].weight.data) for i in range(self.n_layers-1)]
self.bias_rec = [nn.Parameter(self.rbm_layers[i].c.data.clone()) for i in range(self.n_layers-1)]
self.bias_gen = [nn.Parameter(self.rbm_layers[i].b.data) for i in range(self.n_layers-1)]
self.W_mem = nn.Parameter(self.rbm_layers[-1].weight.data)
self.v_bias_mem = nn.Parameter(self.rbm_layers[-1].b.data)
self.h_bias_mem = nn.Parameter(self.rbm_layers[-1].c.data)
for i in range(self.n_layers-1):
self.register_parameter('W_rec%i'%i, self.W_rec[i])
self.register_parameter('W_gen%i'%i, self.W_gen[i])
self.register_parameter('bias_rec%i'%i, self.bias_rec[i])
self.register_parameter('bias_gen%i'%i, self.bias_gen[i])
self.BPNN=nn.Sequential( #用作分类和反向微调参数
torch.nn.Linear(11, 11),
torch.nn.ReLU(),
torch.nn.Dropout(0.5),
torch.nn.Linear(11,3),
)
def forward(self , input_data):
'''
前馈
'''
v = input_data
for i in range(len(self.rbm_layers)):
v = v.view((v.shape[0] , -1)).type(torch.FloatTensor)#flatten
p_v,v = self.rbm_layers[i].forward(v)
# print('p_v:', p_v.shape,p_v)
# print('v:',v.shape,v)
out=self.BPNN(p_v)
# print('out',out.shape,out)
# print(self.BPNN(p_v))
return out
def train_static(self, train_data,train_labels,num_epochs,batch_size):
'''
逐层贪婪训练RBM,固定上一层
'''
tmp = train_data
for i in range(len(self.rbm_layers)):
print("-"*20)
print("Training the {} st rbm layer".format(i+1))
tensor_x = tmp.type(torch.FloatTensor)
tensor_y = train_labels.type(torch.FloatTensor)
_dataset = torch.utils.data.TensorDataset(tensor_x,tensor_y)
_dataloader = torch.utils.data.DataLoader(_dataset)
self.rbm_layers[i].trains(_dataloader,num_epochs,batch_size)
print(type(_dataloader))
# print(train_data.shape)
v = tmp.view((tmp.shape[0] , -1)).type(torch.FloatTensor)
v,_ = self.rbm_layers[i].forward(v)
tmp = v
# print(v.shape)
return
def train_ith(self, train_data,num_epochs,batch_size,ith_layer,rbm_layers):
'''
只训练某一层，可用作调优
'''
if(ith_layer>len(rbm_layers)):
return
v = train_data
for ith in range(ith_layer):
v,out_ = self.rbm_layers[ith].forward(v)
self.rbm_layers[ith_layer].trains(v, num_epochs,batch_size)
return
def trainBP(self,trainloader):
optimizer = torch.optim.SGD(self.BPNN.parameters(), lr=0.005, momentum=0.7)
loss_func = torch.nn.CrossEntropyLoss()
for epoch in range(0):
for step,(x,y) in enumerate(trainloader):
bx = Variable(x)
by = Variable(y)
out=self.forward(bx)[1]
# print(out)
loss=loss_func(out,by)
optimizer.zero_grad()
loss.backward()
optimizer.step()
if step % 10 == 0:
print('Epoch: ', epoch, 'step:', step, '| train loss: %.4f' % loss.data.numpy())
def train_and_test(traind,trainl,testdat,testlabel,loader):
print(type(traind),type(trainl),type(traind),type(trainl),type(loader))
start_time = time.time()
dbn=DBN()
dbn.train()
dbn.train_static(train_data=traind,train_labels=trainl,num_epochs=0,batch_size=20)
optimizer = torch.optim.SGD(dbn.parameters(), lr=0.001, momentum=0.9)
loss_func = torch.nn.CrossEntropyLoss()
train_loader = loader
dbn.trainBP(train_loader)
for epoch in range(0):
for step,(x,y) in enumerate(train_loader):
# print(x.data.numpy(),y.data.numpy())
b_x=Variable(x)
b_y=Variable(y)
output=dbn(b_x)
# print(output)
# print(prediction);print(output);print(b_y)
loss=loss_func(output,b_y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
if step%10==0:
print('Epoch: ', epoch, 'step:',step,'| train loss: %.4f' % loss.data.numpy())
duration=time.time()-start_time
dbn.eval()
test_x = Variable(testdat);test_y = Variable(testlabel)
test_out = dbn(test_x)
# print(test_out)
test_pred = torch.max(test_out, 1)[1]
pre_val = test_pred.data.squeeze().numpy()
y_val = test_y.data.squeeze().numpy()
print('prediciton:',pre_val);print('true value:',y_val)
accuracy = float((pre_val == y_val).astype(int).sum()) / float(test_y.size(0))
print('test accuracy: %.2f' % accuracy,'duration:%.4f' % duration)
return accuracy, duration