-
Notifications
You must be signed in to change notification settings - Fork 2
/
Network.py
183 lines (159 loc) · 6.44 KB
/
Network.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
import torch.nn as nn
import torch.nn.functional as F
from Resnet_imagenet import ResNet, ResNet_blocks, ResNet_cf
import numpy as np
from torch.nn.parameter import Parameter
import torch
import math
from Config import opt
from torch.distributions.normal import Normal
import time
class D_Z(nn.Module):
def __init__(self, label_num, dz):
super(D_Z, self).__init__()
self.model = nn.Sequential(
nn.Linear(dz, label_num),
nn.Sigmoid()
)
self.params = list(filter(self.__filter_func__,
self.parameters()))
def __filter_func__(self, item):
if id(item) not in list(map(id, [])):
return True
else:
return False
def forward(self, z, y):
if self.training is True:
mask = torch.ByteTensor(self.model(z).size()).cuda()
mask.zero_()
mask.scatter_(1, torch.unsqueeze(y, -1), 1)
validity = torch.masked_select(self.model(z), mask)
return validity
else:
print('In test stage, this component is not callable.')
class Y_Z(nn.Module):
def __init__(self, label_num, in_features=256):
super(Y_Z, self).__init__()
self.resnet_cf = nn.Linear(in_features=in_features, out_features=label_num)
self.params = list(filter(self.__filter_func__,
self.parameters()))
def __filter_func__(self, item):
if id(item) not in list(map(id, [])):
return True
else:
return False
def forward(self, z):
if self.training is True:
logit = self.resnet_cf(z)
# y = nn.Softmax(logit)
else:
logit = self.resnet_cf(z)
# y = logit
return logit
class Z_Y(nn.Module):
def __init__(self, gamma, class_num, mode=None, img_chns=1, img_res=None, dz=64):
super(Z_Y, self).__init__()
self.mode = mode
self.alpha = Parameter(torch.Tensor(dz))
self.beta = Parameter(torch.Tensor(dz))
self.gamma = gamma
self.params = list(filter(self.__filter_func__,
self.parameters()))
self.reset_parameters()
self.dz = dz
def reset_parameters(self):
stdv = 1. / math.sqrt(self.alpha.size(0))
self.alpha.data.uniform_(opt.a-stdv, opt.a+stdv)
self.beta.data.uniform_(opt.b-stdv, opt.b+stdv)
def __filter_func__(self, item):
if id(item) not in list(map(id, [])):
return True
else:
return False
def forward(self, embeds=None, x=None):
"""
:param input_instance: ->(batch_size, feature_dim)
:return:
"""
if embeds is not None and x is None:
embeds.detach()
if self.training is True:
mu_z_y = torch.mul(embeds, self.alpha).squeeze()
ones = torch.ones(embeds.size()).cuda()
sigma_z_y = torch.mul(ones, self.beta).squeeze()
gaussian_distribution = Normal(mu_z_y, sigma_z_y)
z_ys = gaussian_distribution.sample([self.gamma]).reshape(-1, self.dz)
if self.mode is not None:
return torch.squeeze(z_ys).cuda(), mu_z_y
else:
y_z_y = self.full_connect(z_ys)
return torch.squeeze(z_ys).cuda(), mu_z_y, y_z_y
else:
print('In test stage, this component is not callable.')
if x is not None and embeds is None:
if self.training is True:
y_x, embed = self.mu_resnet(x)
mu_z_y = torch.mul(embed.detach(), self.alpha).squeeze()
ones = torch.ones(embed.size()).cuda()
sigma_z_y = torch.mul(ones, self.beta).squeeze()
z_ys = list()
for i in range(self.gamma):
z_y_i = torch.normal(mu_z_y, sigma_z_y)
z_ys.append(z_y_i)
z_ys = torch.cat(z_ys, 0)
if self.mode is not None:
return torch.squeeze(z_ys).cuda(), mu_z_y
else:
y_z_y = self.full_connect(z_ys)
return torch.squeeze(z_ys).cuda(), mu_z_y, y_z_y, y_x, embed
else:
print('In test stage, this component is not callable.')
def extra_repr(self):
return 'in_features={}, out_features={}'.format(
self.feature_dim, self.code_len
)
class Z_X(nn.Module):
def __init__(self, gamma, img_chns, img_res):
super(Z_X, self).__init__()
self.gamma = gamma
if 'ResNet' in opt.ResNet_blocks:
if opt.dataset.upper() == 'TINYIMAGENET'.upper():
print("from Resnet_imagenet import ResNet_blocks")
from Resnet_imagenet import ResNet_blocks
self.mu_resblock = ResNet_blocks(18, img_chns, img_res)
self.sigma_resblock = ResNet_blocks(18, img_chns, img_res)
elif opt.dataset.upper() in ['MNIST'.upper(), 'NORB'.upper(), 'CIFAR10'.upper()]:
print("from Resnet import ResNet_blocks")
from Resnet import ResNet_blocks
self.mu_resblock = ResNet_blocks(5, img_chns, img_res)
self.sigma_resblock = ResNet_blocks(5, img_chns, img_res)
self.mu_full_connect = nn.Sequential(
nn.Linear(self.mu_resblock.dz, self.mu_resblock.dz),
)
self.sigma_full_connect = nn.Sequential(
nn.Linear(self.mu_resblock.dz, self.mu_resblock.dz),
)
self.params = list(filter(self.__filter_func__,
self.parameters()))
def __filter_func__(self, item):
if id(item) not in list(map(id, [])):
return True
else:
return False
def forward(self, x):
if self.training is True:
mu_z_x = self.mu_resblock(x)
sigma_z_x = self.sigma_resblock(x)
gaussian_distribution = Normal(mu_z_x, sigma_z_x)
sampled_LVs = gaussian_distribution.sample([self.gamma]).reshape(-1, self.mu_resblock.dz)
return sampled_LVs, mu_z_x
else:
mu_z_x = self.mu_resblock(x)
return mu_z_x
def extra_repr(self):
return 'in_features={}, out_features={}'.format(
self.feature_dim, self.code_len
)
def Y_Augmentation(y, gamma):
ys = y.repeat(gamma, 1).reshape(-1)
return ys