-
Notifications
You must be signed in to change notification settings - Fork 9
/
layer.py
185 lines (153 loc) · 7.16 KB
/
layer.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
'''
@Author: tengfei ma
@Date: 2020-05-09 21:27:12
LastEditTime: 2021-05-29 07:00:00
LastEditors: Please set LastEditors
@Description: RGCN与共享
@FilePath: /Multi-task-pytorch/layer.py
'''
import dgl
import dgl.function as fn
import torch as th
import torch.nn.functional as F
from dgl import DGLGraph
from dgl.nn.pytorch import RelGraphConv
import numpy as np
import torch.nn as nn
#Cross-stitch https://arxiv.org/abs/1604.03539
class Cross_stitch(nn.Module):
def __init__(self):
super(Cross_stitch,self).__init__()
# self.out_dim=out_dim
self.w_aa = nn.Parameter(th.Tensor(1,))
self.w_aa.data=th.tensor(np.random.random(),requires_grad=True)
self.w_ab=nn.Parameter(th.Tensor(1,))
self.w_ab.data=th.tensor(np.random.random(),requires_grad=True)
self.w_ba=nn.Parameter(th.Tensor(1,))
self.w_ba.data=th.tensor(np.random.random(),requires_grad=True)
self.w_bb=nn.Parameter(th.Tensor(1,))
self.w_bb.data=th.tensor(np.random.random(),requires_grad=True)
# np.random.random()
print(self.w_aa)
def forward(self,drug_cnn,drug_kg):
drug_cnn_=self.w_aa*drug_cnn+self.w_ab*drug_kg
drug_kg_=self.w_ba*drug_cnn+self.w_bb*drug_kg
print('shared parameters: w_aa:{:.4f}, w_ab:{:.4f}, w_ba:{:.4f}, w_bb:{:.4f}'.format(self.w_aa,self.w_ab,self.w_ba,self.w_bb))
return drug_cnn_,drug_kg_
#非线性的参数共享
class Shared_Unit_NL(nn.Module):
def __init__(self,input_dim=200, variant='KG-MTL'):
super(Shared_Unit_NL,self).__init__()
self.out_dim=input_dim
self.variant=variant
self.w_aa = nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.w_aa)
self.w_ab=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_normal_(self.w_ab)
self.w_ba=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_normal_(self.w_ba)
self.w_bb=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_normal_(self.w_bb)
self.d_cnn_bias=nn.Parameter(th.Tensor(input_dim,1))
self.d_kg_bias=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_normal_(self.d_cnn_bias)
nn.init.xavier_normal_(self.d_kg_bias)
self.w_aa_ = nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.w_aa_)
self.w_ab_=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.w_ab_)
self.w_ba_=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.w_ba_)
self.w_bb_=nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.w_bb_)
#print(self.w_aa)
def forward(self,drug_cnn,drug_kg):
##### linear
if self.variant=='KG-MTL':
drug_cnn_=self.w_aa_.squeeze()*drug_cnn+self.w_ab_.squeeze()*drug_kg
drug_kg_=self.w_ba_.squeeze()*drug_cnn+self.w_bb_.squeeze()*drug_kg
# #### non-linear
drug_cnn=drug_cnn_.unsqueeze(2)
drug_kg=drug_kg_.unsqueeze(1)
c_mat=th.matmul(drug_cnn,drug_kg)
c_mat_t=c_mat.permute(0, 2, 1)
c_mat=c_mat.view(-1,self.out_dim)
c_mat_t=c_mat_t.view(-1,self.out_dim)
drug_cnn=(c_mat.matmul(self.w_aa)+c_mat_t.matmul(self.w_ab)).view(-1,self.out_dim)+self.d_cnn_bias.squeeze()
drug_kg=(c_mat.matmul(self.w_ba)+c_mat_t.matmul(self.w_bb)).view(-1,self.out_dim)+self.d_kg_bias.squeeze()
return drug_cnn, drug_kg
elif self.variant=='KG-MTL-L':
drug_cnn_=self.w_aa_.squeeze()*drug_cnn+self.w_ab_.squeeze()*drug_kg
drug_kg_=self.w_ba_.squeeze()*drug_cnn+self.w_bb_.squeeze()*drug_kg
return drug_cnn_, drug_kg_
elif self.variant=='KG-MTL-C':
drug_cnn=drug_cnn.unsqueeze(2)
drug_kg=drug_kg.unsqueeze(1)
c_mat=th.matmul(drug_cnn,drug_kg)
c_mat_t=c_mat.permute(0, 2, 1)
c_mat=c_mat.view(-1,self.out_dim)
c_mat_t=c_mat_t.view(-1,self.out_dim)
drug_cnn=(c_mat.matmul(self.w_aa)+c_mat_t.matmul(self.w_ab)).view(-1,self.out_dim)+self.d_cnn_bias.squeeze()
drug_kg=(c_mat.matmul(self.w_ba)+c_mat_t.matmul(self.w_bb)).view(-1,self.out_dim)+self.d_kg_bias.squeeze()
return drug_cnn, drug_kg
class AttentionUnit(nn.Module):
def __init__(self, input_dim=128):
super(AttentionUnit, self).__init__()
self.dim=input_dim
### shared parameters
self.W = nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.W)
self.W_cpi = nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.W_cpi)
self.W_dti = nn.Parameter(th.Tensor(input_dim,1))
nn.init.xavier_uniform_(self.W_dti)
self.W_a = nn.Parameter(th.Tensor(input_dim,2))
nn.init.xavier_uniform_(self.W_a)
#self.MLP_dti=nn.Linear(2*input_dim, input_dim)
self.MLP_cpi=nn.Linear(2*input_dim, input_dim)
def forward(self, drug_cnn, drug_kg):
drug_cnn_=drug_cnn.unsqueeze(1)
drug_kg_=drug_kg.unsqueeze(1)
features=drug_cnn_+drug_kg_
features=features.squeeze(1)
features=F.softmax(th.matmul(th.tanh(features), self.W_a))
features=features.unsqueeze(1)
drug_cnn= (features[:,:,0].unsqueeze(1)* drug_cnn_).squeeze()
drug_kg=(features[:,:,1].unsqueeze(1)*drug_kg_).squeeze()
features=th.cat((drug_cnn, drug_kg), dim=1)
features_cpi=self.MLP_cpi(features)
#features_dti=self.MLP_cpi(features)
return features_cpi, features_cpi
class SimpleUnit(nn.Module):
def __init__(self, input_dim=128):
super(SimpleUnit, self).__init__()
self.dim=input_dim
### shared parameters
# self.W = nn.Parameter(th.Tensor(input_dim,1))
# nn.init.xavier_uniform_(self.W)
# self.W_cpi = nn.Parameter(th.Tensor(input_dim,1))
# nn.init.xavier_uniform_(self.W_cpi)
# self.W_dti = nn.Parameter(th.Tensor(input_dim,1))
# nn.init.xavier_uniform_(self.W_dti)
# self.W_a = nn.Parameter(th.Tensor(input_dim,2))
# nn.init.xavier_uniform_(self.W_a)
#self.MLP_dti=nn.Linear(2*input_dim, input_dim)
self.MLP_cpi=nn.Linear(2*input_dim, input_dim)
def forward(self, drug_cnn, drug_kg):
# drug_cnn_=drug_cnn.unsqueeze(1)
# drug_kg_=drug_kg.unsqueeze(1)
# features=drug_cnn_+drug_kg_
# features=features.squeeze(1)
# features=F.softmax(th.matmul(th.tanh(features), self.W_a))
# features=features.unsqueeze(1)
# drug_cnn= (features[:,:,0].unsqueeze(1)* drug_cnn_).squeeze()
# drug_kg=(features[:,:,1].unsqueeze(1)*drug_kg_).squeeze()
features=th.cat((drug_cnn, drug_kg), dim=1)
features_cpi=self.MLP_cpi(features)
#features_dti=self.MLP_cpi(features)
return features_cpi, features_cpi
if __name__=='__main__':
au=AttentionUnit(input_dim=10)
drug_cnn=th.Tensor(1,10)
drug_kg=th.Tensor(1,10)
au(drug_cnn, drug_kg)