-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathModels.py
153 lines (121 loc) · 5.37 KB
/
Models.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
from Layers import AttentionLayerBag, SelfAttentionLayer, Attention
from torch.nn import Module, Linear, Dropout, Sequential, LeakyReLU, BatchNorm1d, ReLU, Tanh, Sigmoid
import torch
import torch.nn.functional as F
# 定义聚合包信息再聚合实例信息的网络
class Net(Module):
def __init__(self, ins_len, out_len, n_class, drop_rate=0.2):
super(Net, self).__init__()
# 包信息传播网络
self.bag_info = Sequential(
AttentionLayerBag(ins_len=ins_len, out_len=out_len),
LeakyReLU(negative_slope=0.2),
Linear(in_features=out_len, out_features=int(out_len / 2)),
LeakyReLU(negative_slope=0.2),
Dropout(p=drop_rate),
Linear(in_features=int(out_len / 2), out_features=n_class),
LeakyReLU(negative_slope=0.2),
)
# 实例信息传播
self.ins_info = Sequential(
Linear(in_features=ins_len, out_features=int(ins_len / 2)),
LeakyReLU(negative_slope=0.2),
Dropout(p=drop_rate),
Linear(in_features=int(ins_len / 2), out_features=int(ins_len / 4)),
LeakyReLU(negative_slope=0.2),
Linear(in_features=int(ins_len / 4), out_features=n_class),
LeakyReLU(negative_slope=0.2),
)
# 包信息与实例信息聚合
self.agg_info = Sequential(
SelfAttentionLayer(ins_len=n_class, out_len=2 * n_class),
LeakyReLU(negative_slope=0.2),
# 最后一层不用softmax，因为pytorch交叉熵损失函数会自动将输入softmax
Linear(in_features=2 * n_class, out_features=n_class)
)
def forward(self, bag):
bag = bag.float()
bag_info = self.bag_info(bag)
ins_info = self.ins_info(bag)
# 信息拼接
info = torch.cat((bag_info, ins_info), dim=0)
y = self.agg_info(info)
return y
# 定义功能同上，但包先过一层bn1的网络
class NetWithBN(Module):
def __init__(self, ins_len, out_len, n_class, drop_rate=0.2):
super(NetWithBN, self).__init__()
# 包信息传播网络
self.bag_info = Sequential(
BatchNorm1d(num_features=ins_len),
AttentionLayerBag(ins_len=ins_len, out_len=out_len),
LeakyReLU(negative_slope=0.2),
Linear(in_features=out_len, out_features=int(out_len / 2)),
LeakyReLU(negative_slope=0.2),
Dropout(p=drop_rate),
Linear(in_features=int(out_len / 2), out_features=n_class),
LeakyReLU(negative_slope=0.2),
)
# 实例信息传播
self.ins_info = Sequential(
BatchNorm1d(num_features=ins_len),
Linear(in_features=ins_len, out_features=int(ins_len / 2)),
LeakyReLU(negative_slope=0.2),
Dropout(p=drop_rate),
Linear(in_features=int(ins_len / 2), out_features=int(ins_len / 4)),
LeakyReLU(negative_slope=0.2),
Linear(in_features=int(ins_len / 4), out_features=n_class),
LeakyReLU(negative_slope=0.2)
)
# 包信息与实例信息聚合
self.agg_info = Sequential(
SelfAttentionLayer(ins_len=n_class, out_len=2 * n_class),
LeakyReLU(negative_slope=0.2),
# 最后一层不用softmax，因为pytorch交叉熵损失函数会自动将输入softmax
Linear(in_features=2 * n_class, out_features=n_class)
)
def forward(self, bag):
bag = bag.float()
bag_info = self.bag_info(bag)
ins_info = self.ins_info(bag)
# 信息拼接
info = torch.cat((bag_info, ins_info), dim=0)
y = self.agg_info(info)
return y
# 定义用attention-net的网络聚合包内信息的总网络
class NetWithAttention(Module):
def __init__(self, ins_len, n_class, drop_rate=0.2):
super(NetWithAttention, self).__init__()
# 包信息传播网络
self.bag_info = Attention(ins_len=ins_len, n_class=n_class)
# 实例信息传播
self.ins_info = Sequential(
Linear(in_features=ins_len, out_features=int(ins_len / 2)),
LeakyReLU(negative_slope=0.2),
Dropout(p=drop_rate),
Linear(in_features=int(ins_len / 2), out_features=int(ins_len / 4)),
LeakyReLU(negative_slope=0.2),
Linear(in_features=int(ins_len / 4), out_features=n_class),
LeakyReLU(negative_slope=0.2),
)
# 包信息与实例信息聚合
self.agg_info = Sequential(
SelfAttentionLayer(ins_len=n_class, out_len=2 * n_class),
LeakyReLU(negative_slope=0.2),
# 最后一层不用softmax，因为pytorch交叉熵损失函数会自动将输入softmax
Linear(in_features=2 * n_class, out_features=n_class)
)
def forward(self, bag):
bag = bag.float()
bag_info = self.bag_info(bag)
ins_info = self.ins_info(bag)
# 信息拼接
info = torch.cat((bag_info, ins_info), dim=0)
y = self.agg_info(info)
return y
if __name__ == '__main__':
input_ = torch.randn(3, 166)
net = NetWithAttention(ins_len=166, n_class=2, drop_rate=0.2)
out = net(input_)
print(out)
print(out.shape)