Spaces:
Paused
Paused
yjwtheonly
commited on
Commit
·
2d06d0e
1
Parent(s):
6ecb301
midification
Browse files- DiseaseSpecific/__pycache__/attack.cpython-38.pyc +0 -0
- DiseaseSpecific/__pycache__/model.cpython-38.pyc +0 -0
- DiseaseSpecific/__pycache__/utils.cpython-38.pyc +0 -0
- Openai/__pycache__/chat.cpython-38.pyc +0 -0
- Parameters.py +4 -4
- __pycache__/Parameters.cpython-38.pyc +0 -0
- __pycache__/model.cpython-38.pyc +0 -0
- __pycache__/server.cpython-38.pyc +0 -0
- __pycache__/utils.cpython-38.pyc +0 -0
- model.py +504 -0
- server/server.py → server.py +8 -10
- server/__init__.py +0 -0
- server/__pycache__/__init__.cpython-38.pyc +0 -0
- utils.py +195 -0
DiseaseSpecific/__pycache__/attack.cpython-38.pyc
CHANGED
|
Binary files a/DiseaseSpecific/__pycache__/attack.cpython-38.pyc and b/DiseaseSpecific/__pycache__/attack.cpython-38.pyc differ
|
|
|
DiseaseSpecific/__pycache__/model.cpython-38.pyc
CHANGED
|
Binary files a/DiseaseSpecific/__pycache__/model.cpython-38.pyc and b/DiseaseSpecific/__pycache__/model.cpython-38.pyc differ
|
|
|
DiseaseSpecific/__pycache__/utils.cpython-38.pyc
CHANGED
|
Binary files a/DiseaseSpecific/__pycache__/utils.cpython-38.pyc and b/DiseaseSpecific/__pycache__/utils.cpython-38.pyc differ
|
|
|
Openai/__pycache__/chat.cpython-38.pyc
CHANGED
|
Binary files a/Openai/__pycache__/chat.cpython-38.pyc and b/Openai/__pycache__/chat.cpython-38.pyc differ
|
|
|
Parameters.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
from audioop import reverse
|
| 2 |
|
| 3 |
-
GNBRfile = '
|
| 4 |
-
PubTatorfile = '
|
| 5 |
-
UMLSfile = '
|
| 6 |
-
Pubmedfile = '
|
| 7 |
|
| 8 |
edge_type_dict = {
|
| 9 |
'chemical-gene':(['A+', 'A-', 'B', 'E+', 'E-', 'E', 'N', 'O', 'K', 'Z'],
|
|
|
|
| 1 |
from audioop import reverse
|
| 2 |
|
| 3 |
+
GNBRfile = 'GNBRdata/'
|
| 4 |
+
PubTatorfile = 'pubtator/'
|
| 5 |
+
UMLSfile = 'umls/META/'
|
| 6 |
+
Pubmedfile = 'pubmed/'
|
| 7 |
|
| 8 |
edge_type_dict = {
|
| 9 |
'chemical-gene':(['A+', 'A-', 'B', 'E+', 'E-', 'E', 'N', 'O', 'K', 'Z'],
|
__pycache__/Parameters.cpython-38.pyc
ADDED
|
Binary file (3.23 kB). View file
|
|
|
__pycache__/model.cpython-38.pyc
ADDED
|
Binary file (11.4 kB). View file
|
|
|
__pycache__/server.cpython-38.pyc
ADDED
|
Binary file (18.8 kB). View file
|
|
|
__pycache__/utils.cpython-38.pyc
ADDED
|
Binary file (7.81 kB). View file
|
|
|
model.py
ADDED
|
@@ -0,0 +1,504 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from torch.nn import functional as F, Parameter
|
| 3 |
+
from torch.autograd import Variable
|
| 4 |
+
from torch.nn.init import xavier_normal_, xavier_uniform_
|
| 5 |
+
from torch.nn.utils.rnn import pack_padded_sequence, pad_packed_sequence
|
| 6 |
+
|
| 7 |
+
class Distmult(torch.nn.Module):
|
| 8 |
+
def __init__(self, args, num_entities, num_relations):
|
| 9 |
+
super(Distmult, self).__init__()
|
| 10 |
+
|
| 11 |
+
if args.max_norm:
|
| 12 |
+
self.emb_e = torch.nn.Embedding(num_entities, args.embedding_dim, max_norm=1.0)
|
| 13 |
+
self.emb_rel = torch.nn.Embedding(num_relations, args.embedding_dim)
|
| 14 |
+
else:
|
| 15 |
+
self.emb_e = torch.nn.Embedding(num_entities, args.embedding_dim, padding_idx=None)
|
| 16 |
+
self.emb_rel = torch.nn.Embedding(num_relations, args.embedding_dim, padding_idx=None)
|
| 17 |
+
|
| 18 |
+
self.inp_drop = torch.nn.Dropout(args.input_drop)
|
| 19 |
+
self.loss = torch.nn.CrossEntropyLoss()
|
| 20 |
+
|
| 21 |
+
self.init()
|
| 22 |
+
|
| 23 |
+
def init(self):
|
| 24 |
+
xavier_normal_(self.emb_e.weight)
|
| 25 |
+
xavier_normal_(self.emb_rel.weight)
|
| 26 |
+
|
| 27 |
+
def score_sr(self, sub, rel, sigmoid = False):
|
| 28 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 29 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 30 |
+
|
| 31 |
+
#sub_emb = self.inp_drop(sub_emb)
|
| 32 |
+
#rel_emb = self.inp_drop(rel_emb)
|
| 33 |
+
|
| 34 |
+
pred = torch.mm(sub_emb*rel_emb, self.emb_e.weight.transpose(1,0))
|
| 35 |
+
if sigmoid:
|
| 36 |
+
pred = torch.sigmoid(pred)
|
| 37 |
+
return pred
|
| 38 |
+
|
| 39 |
+
def score_or(self, obj, rel, sigmoid = False):
|
| 40 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 41 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 42 |
+
|
| 43 |
+
#obj_emb = self.inp_drop(obj_emb)
|
| 44 |
+
#rel_emb = self.inp_drop(rel_emb)
|
| 45 |
+
|
| 46 |
+
pred = torch.mm(obj_emb*rel_emb, self.emb_e.weight.transpose(1,0))
|
| 47 |
+
if sigmoid:
|
| 48 |
+
pred = torch.sigmoid(pred)
|
| 49 |
+
return pred
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
def forward(self, sub_emb, rel_emb, mode='rhs', sigmoid=False):
|
| 53 |
+
'''
|
| 54 |
+
When mode is 'rhs' we expect (s,r); for 'lhs', we expect (o,r)
|
| 55 |
+
For distmult, computations for both modes are equivalent, so we do not need if-else block
|
| 56 |
+
'''
|
| 57 |
+
sub_emb = self.inp_drop(sub_emb)
|
| 58 |
+
rel_emb = self.inp_drop(rel_emb)
|
| 59 |
+
|
| 60 |
+
pred = torch.mm(sub_emb*rel_emb, self.emb_e.weight.transpose(1,0))
|
| 61 |
+
|
| 62 |
+
if sigmoid:
|
| 63 |
+
pred = torch.sigmoid(pred)
|
| 64 |
+
|
| 65 |
+
return pred
|
| 66 |
+
|
| 67 |
+
def score_triples(self, sub, rel, obj, sigmoid=False):
|
| 68 |
+
'''
|
| 69 |
+
Inputs - subject, relation, object
|
| 70 |
+
Return - score
|
| 71 |
+
'''
|
| 72 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 73 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 74 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 75 |
+
|
| 76 |
+
pred = torch.sum(sub_emb*rel_emb*obj_emb, dim=-1)
|
| 77 |
+
|
| 78 |
+
if sigmoid:
|
| 79 |
+
pred = torch.sigmoid(pred)
|
| 80 |
+
|
| 81 |
+
return pred
|
| 82 |
+
|
| 83 |
+
def score_emb(self, emb_s, emb_r, emb_o, sigmoid=False):
|
| 84 |
+
'''
|
| 85 |
+
Inputs - embeddings of subject, relation, object
|
| 86 |
+
Return - score
|
| 87 |
+
'''
|
| 88 |
+
pred = torch.sum(emb_s*emb_r*emb_o, dim=-1)
|
| 89 |
+
|
| 90 |
+
if sigmoid:
|
| 91 |
+
pred = torch.sigmoid(pred)
|
| 92 |
+
|
| 93 |
+
return pred
|
| 94 |
+
|
| 95 |
+
def score_triples_vec(self, sub, rel, obj, sigmoid=False):
|
| 96 |
+
'''
|
| 97 |
+
Inputs - subject, relation, object
|
| 98 |
+
Return - a vector score for the triple instead of reducing over the embedding dimension
|
| 99 |
+
'''
|
| 100 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 101 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 102 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 103 |
+
|
| 104 |
+
pred = sub_emb*rel_emb*obj_emb
|
| 105 |
+
|
| 106 |
+
if sigmoid:
|
| 107 |
+
pred = torch.sigmoid(pred)
|
| 108 |
+
|
| 109 |
+
return pred
|
| 110 |
+
|
| 111 |
+
class Complex(torch.nn.Module):
|
| 112 |
+
def __init__(self, args, num_entities, num_relations):
|
| 113 |
+
super(Complex, self).__init__()
|
| 114 |
+
|
| 115 |
+
if args.max_norm:
|
| 116 |
+
self.emb_e = torch.nn.Embedding(num_entities, 2*args.embedding_dim, max_norm=1.0)
|
| 117 |
+
self.emb_rel = torch.nn.Embedding(num_relations, 2*args.embedding_dim)
|
| 118 |
+
else:
|
| 119 |
+
self.emb_e = torch.nn.Embedding(num_entities, 2*args.embedding_dim, padding_idx=None)
|
| 120 |
+
self.emb_rel = torch.nn.Embedding(num_relations, 2*args.embedding_dim, padding_idx=None)
|
| 121 |
+
|
| 122 |
+
self.inp_drop = torch.nn.Dropout(args.input_drop)
|
| 123 |
+
self.loss = torch.nn.CrossEntropyLoss()
|
| 124 |
+
|
| 125 |
+
self.init()
|
| 126 |
+
|
| 127 |
+
def init(self):
|
| 128 |
+
xavier_normal_(self.emb_e.weight)
|
| 129 |
+
xavier_normal_(self.emb_rel.weight)
|
| 130 |
+
|
| 131 |
+
def score_sr(self, sub, rel, sigmoid = False):
|
| 132 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 133 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 134 |
+
|
| 135 |
+
s_real, s_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 136 |
+
rel_real, rel_img = torch.chunk(sub_emb, 2, dim=-1)
|
| 137 |
+
emb_e_real, emb_e_img = torch.chunk(self.emb_e.weight, 2, dim=-1)
|
| 138 |
+
|
| 139 |
+
realo_realreal = s_real*rel_real
|
| 140 |
+
realo_imgimg = s_img*rel_img
|
| 141 |
+
realo = realo_realreal - realo_imgimg
|
| 142 |
+
real = torch.mm(realo, emb_e_real.transpose(1,0))
|
| 143 |
+
|
| 144 |
+
imgo_realimg = s_real*rel_img
|
| 145 |
+
imgo_imgreal = s_img*rel_real
|
| 146 |
+
imgo = imgo_realimg + imgo_imgreal
|
| 147 |
+
img = torch.mm(imgo, emb_e_img.transpose(1,0))
|
| 148 |
+
|
| 149 |
+
pred = real + img
|
| 150 |
+
|
| 151 |
+
if sigmoid:
|
| 152 |
+
pred = torch.sigmoid(pred)
|
| 153 |
+
return pred
|
| 154 |
+
|
| 155 |
+
|
| 156 |
+
def score_or(self, obj, rel, sigmoid = False):
|
| 157 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 158 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 159 |
+
|
| 160 |
+
rel_real, rel_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 161 |
+
o_real, o_img = torch.chunk(obj_emb, 2, dim=-1)
|
| 162 |
+
emb_e_real, emb_e_img = torch.chunk(self.emb_e.weight, 2, dim=-1)
|
| 163 |
+
|
| 164 |
+
#rel_real = self.inp_drop(rel_real)
|
| 165 |
+
#rel_img = self.inp_drop(rel_img)
|
| 166 |
+
#o_real = self.inp_drop(o_real)
|
| 167 |
+
#o_img = self.inp_drop(o_img)
|
| 168 |
+
|
| 169 |
+
# complex space bilinear product (equivalent to HolE)
|
| 170 |
+
# realrealreal = torch.mm(rel_real*o_real, emb_e_real.transpose(1,0))
|
| 171 |
+
# realimgimg = torch.mm(rel_img*o_img, emb_e_real.transpose(1,0))
|
| 172 |
+
# imgrealimg = torch.mm(rel_real*o_img, emb_e_img.transpose(1,0))
|
| 173 |
+
# imgimgreal = torch.mm(rel_img*o_real, emb_e_img.transpose(1,0))
|
| 174 |
+
# pred = realrealreal + realimgimg + imgrealimg - imgimgreal
|
| 175 |
+
|
| 176 |
+
reals_realreal = rel_real*o_real
|
| 177 |
+
reals_imgimg = rel_img*o_img
|
| 178 |
+
reals = reals_realreal + reals_imgimg
|
| 179 |
+
real = torch.mm(reals, emb_e_real.transpose(1,0))
|
| 180 |
+
|
| 181 |
+
imgs_realimg = rel_real*o_img
|
| 182 |
+
imgs_imgreal = rel_img*o_real
|
| 183 |
+
imgs = imgs_realimg - imgs_imgreal
|
| 184 |
+
img = torch.mm(imgs, emb_e_img.transpose(1,0))
|
| 185 |
+
|
| 186 |
+
pred = real + img
|
| 187 |
+
|
| 188 |
+
if sigmoid:
|
| 189 |
+
pred = torch.sigmoid(pred)
|
| 190 |
+
return pred
|
| 191 |
+
|
| 192 |
+
|
| 193 |
+
def forward(self, sub_emb, rel_emb, mode='rhs', sigmoid=False):
|
| 194 |
+
'''
|
| 195 |
+
When mode is 'rhs' we expect (s,r); for 'lhs', we expect (o,r)
|
| 196 |
+
|
| 197 |
+
'''
|
| 198 |
+
if mode == 'lhs':
|
| 199 |
+
rel_real, rel_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 200 |
+
o_real, o_img = torch.chunk(sub_emb, 2, dim=-1)
|
| 201 |
+
emb_e_real, emb_e_img = torch.chunk(self.emb_e.weight, 2, dim=-1)
|
| 202 |
+
|
| 203 |
+
rel_real = self.inp_drop(rel_real)
|
| 204 |
+
rel_img = self.inp_drop(rel_img)
|
| 205 |
+
o_real = self.inp_drop(o_real)
|
| 206 |
+
o_img = self.inp_drop(o_img)
|
| 207 |
+
|
| 208 |
+
reals_realreal = rel_real*o_real
|
| 209 |
+
reals_imgimg = rel_img*o_img
|
| 210 |
+
reals = reals_realreal + reals_imgimg
|
| 211 |
+
real = torch.mm(reals, emb_e_real.transpose(1,0))
|
| 212 |
+
|
| 213 |
+
imgs_realimg = rel_real*o_img
|
| 214 |
+
imgs_imgreal = rel_img*o_real
|
| 215 |
+
imgs = imgs_realimg - imgs_imgreal
|
| 216 |
+
img = torch.mm(imgs, emb_e_img.transpose(1,0))
|
| 217 |
+
|
| 218 |
+
pred = real + img
|
| 219 |
+
|
| 220 |
+
else:
|
| 221 |
+
s_real, s_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 222 |
+
rel_real, rel_img = torch.chunk(sub_emb, 2, dim=-1)
|
| 223 |
+
emb_e_real, emb_e_img = torch.chunk(self.emb_e.weight, 2, dim=-1)
|
| 224 |
+
|
| 225 |
+
s_real = self.inp_drop(s_real)
|
| 226 |
+
s_img = self.inp_drop(s_img)
|
| 227 |
+
rel_real = self.inp_drop(rel_real)
|
| 228 |
+
rel_img = self.inp_drop(rel_img)
|
| 229 |
+
|
| 230 |
+
realo_realreal = s_real*rel_real
|
| 231 |
+
realo_imgimg = s_img*rel_img
|
| 232 |
+
realo = realo_realreal - realo_imgimg
|
| 233 |
+
real = torch.mm(realo, emb_e_real.transpose(1,0))
|
| 234 |
+
|
| 235 |
+
imgo_realimg = s_real*rel_img
|
| 236 |
+
imgo_imgreal = s_img*rel_real
|
| 237 |
+
imgo = imgo_realimg + imgo_imgreal
|
| 238 |
+
img = torch.mm(imgo, emb_e_img.transpose(1,0))
|
| 239 |
+
|
| 240 |
+
pred = real + img
|
| 241 |
+
|
| 242 |
+
if sigmoid:
|
| 243 |
+
pred = torch.sigmoid(pred)
|
| 244 |
+
|
| 245 |
+
return pred
|
| 246 |
+
|
| 247 |
+
def score_triples(self, sub, rel, obj, sigmoid=False):
|
| 248 |
+
'''
|
| 249 |
+
Inputs - subject, relation, object
|
| 250 |
+
Return - score
|
| 251 |
+
'''
|
| 252 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 253 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 254 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 255 |
+
|
| 256 |
+
s_real, s_img = torch.chunk(sub_emb, 2, dim=-1)
|
| 257 |
+
rel_real, rel_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 258 |
+
o_real, o_img = torch.chunk(obj_emb, 2, dim=-1)
|
| 259 |
+
|
| 260 |
+
realrealreal = torch.sum(s_real*rel_real*o_real, dim=-1)
|
| 261 |
+
realimgimg = torch.sum(s_real*rel_img*o_img, axis=-1)
|
| 262 |
+
imgrealimg = torch.sum(s_img*rel_real*o_img, axis=-1)
|
| 263 |
+
imgimgreal = torch.sum(s_img*rel_img*o_real, axis=-1)
|
| 264 |
+
|
| 265 |
+
pred = realrealreal + realimgimg + imgrealimg - imgimgreal
|
| 266 |
+
|
| 267 |
+
if sigmoid:
|
| 268 |
+
pred = torch.sigmoid(pred)
|
| 269 |
+
|
| 270 |
+
return pred
|
| 271 |
+
|
| 272 |
+
def score_emb(self, emb_s, emb_r, emb_o, sigmoid=False):
|
| 273 |
+
'''
|
| 274 |
+
Inputs - embeddings of subject, relation, object
|
| 275 |
+
Return - score
|
| 276 |
+
'''
|
| 277 |
+
|
| 278 |
+
s_real, s_img = torch.chunk(emb_s, 2, dim=-1)
|
| 279 |
+
rel_real, rel_img = torch.chunk(emb_r, 2, dim=-1)
|
| 280 |
+
o_real, o_img = torch.chunk(emb_o, 2, dim=-1)
|
| 281 |
+
|
| 282 |
+
realrealreal = torch.sum(s_real*rel_real*o_real, dim=-1)
|
| 283 |
+
realimgimg = torch.sum(s_real*rel_img*o_img, axis=-1)
|
| 284 |
+
imgrealimg = torch.sum(s_img*rel_real*o_img, axis=-1)
|
| 285 |
+
imgimgreal = torch.sum(s_img*rel_img*o_real, axis=-1)
|
| 286 |
+
|
| 287 |
+
pred = realrealreal + realimgimg + imgrealimg - imgimgreal
|
| 288 |
+
|
| 289 |
+
if sigmoid:
|
| 290 |
+
pred = torch.sigmoid(pred)
|
| 291 |
+
|
| 292 |
+
return pred
|
| 293 |
+
|
| 294 |
+
def score_triples_vec(self, sub, rel, obj, sigmoid=False):
|
| 295 |
+
'''
|
| 296 |
+
Inputs - subject, relation, object
|
| 297 |
+
Return - a vector score for the triple instead of reducing over the embedding dimension
|
| 298 |
+
'''
|
| 299 |
+
sub_emb = self.emb_e(sub).squeeze(dim=1)
|
| 300 |
+
rel_emb = self.emb_rel(rel).squeeze(dim=1)
|
| 301 |
+
obj_emb = self.emb_e(obj).squeeze(dim=1)
|
| 302 |
+
|
| 303 |
+
s_real, s_img = torch.chunk(sub_emb, 2, dim=-1)
|
| 304 |
+
rel_real, rel_img = torch.chunk(rel_emb, 2, dim=-1)
|
| 305 |
+
o_real, o_img = torch.chunk(obj_emb, 2, dim=-1)
|
| 306 |
+
|
| 307 |
+
realrealreal = s_real*rel_real*o_real
|
| 308 |
+
realimgimg = s_real*rel_img*o_img
|
| 309 |
+
imgrealimg = s_img*rel_real*o_img
|
| 310 |
+
imgimgreal = s_img*rel_img*o_real
|
| 311 |
+
|
| 312 |
+
pred = realrealreal + realimgimg + imgrealimg - imgimgreal
|
| 313 |
+
|
| 314 |
+
if sigmoid:
|
| 315 |
+
pred = torch.sigmoid(pred)
|
| 316 |
+
|
| 317 |
+
return pred
|
| 318 |
+
|
| 319 |
+
class Conve(torch.nn.Module):
|
| 320 |
+
|
| 321 |
+
#Too slow !!!!
|
| 322 |
+
|
| 323 |
+
def __init__(self, args, num_entities, num_relations):
|
| 324 |
+
super(Conve, self).__init__()
|
| 325 |
+
|
| 326 |
+
if args.max_norm:
|
| 327 |
+
self.emb_e = torch.nn.Embedding(num_entities, args.embedding_dim, max_norm=1.0)
|
| 328 |
+
self.emb_rel = torch.nn.Embedding(num_relations, args.embedding_dim)
|
| 329 |
+
else:
|
| 330 |
+
self.emb_e = torch.nn.Embedding(num_entities, args.embedding_dim, padding_idx=None)
|
| 331 |
+
self.emb_rel = torch.nn.Embedding(num_relations, args.embedding_dim, padding_idx=None)
|
| 332 |
+
|
| 333 |
+
self.inp_drop = torch.nn.Dropout(args.input_drop)
|
| 334 |
+
self.hidden_drop = torch.nn.Dropout(args.hidden_drop)
|
| 335 |
+
self.feature_drop = torch.nn.Dropout2d(args.feat_drop)
|
| 336 |
+
|
| 337 |
+
self.embedding_dim = args.embedding_dim #default is 200
|
| 338 |
+
self.num_filters = args.num_filters # default is 32
|
| 339 |
+
self.kernel_size = args.kernel_size # default is 3
|
| 340 |
+
self.stack_width = args.stack_width # default is 20
|
| 341 |
+
self.stack_height = args.embedding_dim // self.stack_width
|
| 342 |
+
|
| 343 |
+
self.bn0 = torch.nn.BatchNorm2d(1)
|
| 344 |
+
self.bn1 = torch.nn.BatchNorm2d(self.num_filters)
|
| 345 |
+
self.bn2 = torch.nn.BatchNorm1d(args.embedding_dim)
|
| 346 |
+
|
| 347 |
+
self.conv1 = torch.nn.Conv2d(1, out_channels=self.num_filters,
|
| 348 |
+
kernel_size=(self.kernel_size, self.kernel_size),
|
| 349 |
+
stride=1, padding=0, bias=args.use_bias)
|
| 350 |
+
#self.conv1 = torch.nn.Conv2d(1, 32, (3, 3), 1, 0, bias=args.use_bias) # <-- default
|
| 351 |
+
|
| 352 |
+
flat_sz_h = int(2*self.stack_width) - self.kernel_size + 1
|
| 353 |
+
flat_sz_w = self.stack_height - self.kernel_size + 1
|
| 354 |
+
self.flat_sz = flat_sz_h*flat_sz_w*self.num_filters
|
| 355 |
+
self.fc = torch.nn.Linear(self.flat_sz, args.embedding_dim)
|
| 356 |
+
|
| 357 |
+
self.register_parameter('b', Parameter(torch.zeros(num_entities)))
|
| 358 |
+
self.loss = torch.nn.CrossEntropyLoss()
|
| 359 |
+
|
| 360 |
+
self.init()
|
| 361 |
+
|
| 362 |
+
def init(self):
|
| 363 |
+
xavier_normal_(self.emb_e.weight)
|
| 364 |
+
xavier_normal_(self.emb_rel.weight)
|
| 365 |
+
|
| 366 |
+
def concat(self, e1_embed, rel_embed, form='plain'):
|
| 367 |
+
if form == 'plain':
|
| 368 |
+
e1_embed = e1_embed. view(-1, 1, self.stack_width, self.stack_height)
|
| 369 |
+
rel_embed = rel_embed.view(-1, 1, self.stack_width, self.stack_height)
|
| 370 |
+
stack_inp = torch.cat([e1_embed, rel_embed], 2)
|
| 371 |
+
|
| 372 |
+
elif form == 'alternate':
|
| 373 |
+
e1_embed = e1_embed. view(-1, 1, self.embedding_dim)
|
| 374 |
+
rel_embed = rel_embed.view(-1, 1, self.embedding_dim)
|
| 375 |
+
stack_inp = torch.cat([e1_embed, rel_embed], 1)
|
| 376 |
+
stack_inp = torch.transpose(stack_inp, 2, 1).reshape((-1, 1, 2*self.stack_width, self.stack_height))
|
| 377 |
+
|
| 378 |
+
else: raise NotImplementedError
|
| 379 |
+
return stack_inp
|
| 380 |
+
|
| 381 |
+
def conve_architecture(self, sub_emb, rel_emb):
|
| 382 |
+
stacked_inputs = self.concat(sub_emb, rel_emb)
|
| 383 |
+
stacked_inputs = self.bn0(stacked_inputs)
|
| 384 |
+
x = self.inp_drop(stacked_inputs)
|
| 385 |
+
x = self.conv1(x)
|
| 386 |
+
x = self.bn1(x)
|
| 387 |
+
x = F.relu(x)
|
| 388 |
+
x = self.feature_drop(x)
|
| 389 |
+
#x = x.view(x.shape[0], -1)
|
| 390 |
+
x = x.view(-1, self.flat_sz)
|
| 391 |
+
x = self.fc(x)
|
| 392 |
+
x = self.hidden_drop(x)
|
| 393 |
+
x = self.bn2(x)
|
| 394 |
+
x = F.relu(x)
|
| 395 |
+
|
| 396 |
+
return x
|
| 397 |
+
|
| 398 |
+
def score_sr(self, sub, rel, sigmoid = False):
|
| 399 |
+
sub_emb = self.emb_e(sub)
|
| 400 |
+
rel_emb = self.emb_rel(rel)
|
| 401 |
+
|
| 402 |
+
x = self.conve_architecture(sub_emb, rel_emb)
|
| 403 |
+
|
| 404 |
+
pred = torch.mm(x, self.emb_e.weight.transpose(1,0))
|
| 405 |
+
pred += self.b.expand_as(pred)
|
| 406 |
+
|
| 407 |
+
if sigmoid:
|
| 408 |
+
pred = torch.sigmoid(pred)
|
| 409 |
+
return pred
|
| 410 |
+
|
| 411 |
+
def score_or(self, obj, rel, sigmoid = False):
|
| 412 |
+
obj_emb = self.emb_e(obj)
|
| 413 |
+
rel_emb = self.emb_rel(rel)
|
| 414 |
+
|
| 415 |
+
x = self.conve_architecture(obj_emb, rel_emb)
|
| 416 |
+
pred = torch.mm(x, self.emb_e.weight.transpose(1,0))
|
| 417 |
+
pred += self.b.expand_as(pred)
|
| 418 |
+
|
| 419 |
+
if sigmoid:
|
| 420 |
+
pred = torch.sigmoid(pred)
|
| 421 |
+
return pred
|
| 422 |
+
|
| 423 |
+
|
| 424 |
+
def forward(self, sub_emb, rel_emb, mode='rhs', sigmoid=False):
|
| 425 |
+
'''
|
| 426 |
+
When mode is 'rhs' we expect (s,r); for 'lhs', we expect (o,r)
|
| 427 |
+
For conve, computations for both modes are equivalent, so we do not need if-else block
|
| 428 |
+
'''
|
| 429 |
+
x = self.conve_architecture(sub_emb, rel_emb)
|
| 430 |
+
|
| 431 |
+
pred = torch.mm(x, self.emb_e.weight.transpose(1,0))
|
| 432 |
+
pred += self.b.expand_as(pred)
|
| 433 |
+
|
| 434 |
+
if sigmoid:
|
| 435 |
+
pred = torch.sigmoid(pred)
|
| 436 |
+
|
| 437 |
+
return pred
|
| 438 |
+
|
| 439 |
+
def score_triples(self, sub, rel, obj, sigmoid=False):
|
| 440 |
+
'''
|
| 441 |
+
Inputs - subject, relation, object
|
| 442 |
+
Return - score
|
| 443 |
+
'''
|
| 444 |
+
sub_emb = self.emb_e(sub)
|
| 445 |
+
rel_emb = self.emb_rel(rel)
|
| 446 |
+
obj_emb = self.emb_e(obj)
|
| 447 |
+
x = self.conve_architecture(sub_emb, rel_emb)
|
| 448 |
+
|
| 449 |
+
pred = torch.mm(x, obj_emb.transpose(1,0))
|
| 450 |
+
#print(pred.shape)
|
| 451 |
+
pred += self.b[obj].expand_as(pred) #taking the bias value for object embedding
|
| 452 |
+
# above works fine for single input triples;
|
| 453 |
+
# but if input is batch of triples, then this is a matrix of (num_trip x num_trip) where diagonal is scores
|
| 454 |
+
# so use torch.diagonal() after calling this function
|
| 455 |
+
pred = torch.diagonal(pred)
|
| 456 |
+
# or could have used : pred= torch.sum(x*obj_emb, dim=-1)
|
| 457 |
+
|
| 458 |
+
if sigmoid:
|
| 459 |
+
pred = torch.sigmoid(pred)
|
| 460 |
+
|
| 461 |
+
return pred
|
| 462 |
+
|
| 463 |
+
def score_emb(self, emb_s, emb_r, emb_o, sigmoid=False):
|
| 464 |
+
'''
|
| 465 |
+
Inputs - embeddings of subject, relation, object
|
| 466 |
+
Return - score
|
| 467 |
+
'''
|
| 468 |
+
x = self.conve_architecture(emb_s, emb_r)
|
| 469 |
+
|
| 470 |
+
pred = torch.mm(x, emb_o.transpose(1,0))
|
| 471 |
+
#pred += self.b[obj].expand_as(pred) #taking the bias value for object embedding - don't know which obj
|
| 472 |
+
# above works fine for single input triples;
|
| 473 |
+
# but if input is batch of triples, then this is a matrix of (num_trip x num_trip) where diagonal is scores
|
| 474 |
+
# so use torch.diagonal() after calling this function
|
| 475 |
+
pred = torch.diagonal(pred)
|
| 476 |
+
# or could have used : pred= torch.sum(x*obj_emb, dim=-1)
|
| 477 |
+
|
| 478 |
+
if sigmoid:
|
| 479 |
+
pred = torch.sigmoid(pred)
|
| 480 |
+
|
| 481 |
+
return pred
|
| 482 |
+
|
| 483 |
+
def score_triples_vec(self, sub, rel, obj, sigmoid=False):
|
| 484 |
+
'''
|
| 485 |
+
Inputs - subject, relation, object
|
| 486 |
+
Return - a vector score for the triple instead of reducing over the embedding dimension
|
| 487 |
+
'''
|
| 488 |
+
sub_emb = self.emb_e(sub)
|
| 489 |
+
rel_emb = self.emb_rel(rel)
|
| 490 |
+
obj_emb = self.emb_e(obj)
|
| 491 |
+
|
| 492 |
+
x = self.conve_architecture(sub_emb, rel_emb)
|
| 493 |
+
|
| 494 |
+
#pred = torch.mm(x, obj_emb.transpose(1,0))
|
| 495 |
+
pred = x*obj_emb
|
| 496 |
+
#print(pred.shape, self.b[obj].shape) #shapes are [7,200] and [7]
|
| 497 |
+
#pred += self.b[obj].expand_as(pred) #taking the bias value for object embedding - can't add scalar to vector
|
| 498 |
+
|
| 499 |
+
#pred = sub_emb*rel_emb*obj_emb
|
| 500 |
+
|
| 501 |
+
if sigmoid:
|
| 502 |
+
pred = torch.sigmoid(pred)
|
| 503 |
+
|
| 504 |
+
return pred
|
server/server.py → server.py
RENAMED
|
@@ -9,7 +9,7 @@ import numpy as np
|
|
| 9 |
import json
|
| 10 |
import networkx as nx
|
| 11 |
import spacy
|
| 12 |
-
os.system("python -m spacy download en-core-web-sm")
|
| 13 |
import pickle as pkl
|
| 14 |
#%%
|
| 15 |
|
|
@@ -17,14 +17,12 @@ from torch.nn.modules.loss import CrossEntropyLoss
|
|
| 17 |
from transformers import AutoTokenizer
|
| 18 |
from transformers import BioGptForCausalLM, BartForConditionalGeneration
|
| 19 |
|
| 20 |
-
import server_utils
|
| 21 |
|
| 22 |
-
sys.path.append("..")
|
| 23 |
import Parameters
|
| 24 |
from Openai.chat import generate_abstract
|
| 25 |
-
|
| 26 |
-
import
|
| 27 |
-
from attack import calculate_edge_bound, get_model_loss_without_softmax
|
| 28 |
|
| 29 |
|
| 30 |
specific_model = None
|
|
@@ -51,8 +49,8 @@ np.set_printoptions(precision=5)
|
|
| 51 |
cudnn.benchmark = False
|
| 52 |
|
| 53 |
model_name = '{0}_{1}_{2}_{3}_{4}'.format(args.model, args.embedding_dim, args.input_drop, args.hidden_drop, args.feat_drop)
|
| 54 |
-
model_path = '
|
| 55 |
-
data_path = os.path.join('
|
| 56 |
data = utils.load_data(os.path.join(data_path, 'all.txt'))
|
| 57 |
|
| 58 |
n_ent, n_rel, ent_to_id, rel_to_id = utils.generate_dicts(data_path)
|
|
@@ -596,11 +594,11 @@ def specific_func(start_entity, end_entity):
|
|
| 596 |
o_name = entity_raw_name[id_to_entity[str(o)]]
|
| 597 |
attack_data = np.array([[s, r, o]])
|
| 598 |
path_list = []
|
| 599 |
-
with open(f'
|
| 600 |
for line in fl.readlines():
|
| 601 |
line.replace('\n', '')
|
| 602 |
path_list.append(line)
|
| 603 |
-
with open(f'
|
| 604 |
sentence_dict = json.load(fl)
|
| 605 |
dpath = []
|
| 606 |
for k, v in sentence_dict.items():
|
|
|
|
| 9 |
import json
|
| 10 |
import networkx as nx
|
| 11 |
import spacy
|
| 12 |
+
# os.system("python -m spacy download en-core-web-sm")
|
| 13 |
import pickle as pkl
|
| 14 |
#%%
|
| 15 |
|
|
|
|
| 17 |
from transformers import AutoTokenizer
|
| 18 |
from transformers import BioGptForCausalLM, BartForConditionalGeneration
|
| 19 |
|
| 20 |
+
from server import server_utils
|
| 21 |
|
|
|
|
| 22 |
import Parameters
|
| 23 |
from Openai.chat import generate_abstract
|
| 24 |
+
from DiseaseSpecific import utils, attack
|
| 25 |
+
from DiseaseSpecific.attack import calculate_edge_bound, get_model_loss_without_softmax
|
|
|
|
| 26 |
|
| 27 |
|
| 28 |
specific_model = None
|
|
|
|
| 49 |
cudnn.benchmark = False
|
| 50 |
|
| 51 |
model_name = '{0}_{1}_{2}_{3}_{4}'.format(args.model, args.embedding_dim, args.input_drop, args.hidden_drop, args.feat_drop)
|
| 52 |
+
model_path = 'DiseaseSpecific/saved_models/{0}_{1}.model'.format(args.data, model_name)
|
| 53 |
+
data_path = os.path.join('DiseaseSpecific/processed_data', args.data)
|
| 54 |
data = utils.load_data(os.path.join(data_path, 'all.txt'))
|
| 55 |
|
| 56 |
n_ent, n_rel, ent_to_id, rel_to_id = utils.generate_dicts(data_path)
|
|
|
|
| 594 |
o_name = entity_raw_name[id_to_entity[str(o)]]
|
| 595 |
attack_data = np.array([[s, r, o]])
|
| 596 |
path_list = []
|
| 597 |
+
with open(f'DiseaseSpecific/generate_abstract/path/random_{args.reasonable_rate}_path.json', 'r') as fl:
|
| 598 |
for line in fl.readlines():
|
| 599 |
line.replace('\n', '')
|
| 600 |
path_list.append(line)
|
| 601 |
+
with open(f'DiseaseSpecific/generate_abstract/random_{args.reasonable_rate}_sentence.json', 'r') as fl:
|
| 602 |
sentence_dict = json.load(fl)
|
| 603 |
dpath = []
|
| 604 |
for k, v in sentence_dict.items():
|
server/__init__.py
ADDED
|
File without changes
|
server/__pycache__/__init__.cpython-38.pyc
ADDED
|
Binary file (137 Bytes). View file
|
|
|
utils.py
ADDED
|
@@ -0,0 +1,195 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
'''
|
| 2 |
+
A file modified on https://github.com/PeruBhardwaj/AttributionAttack/blob/main/KGEAttack/ConvE/utils.py
|
| 3 |
+
'''
|
| 4 |
+
#%%
|
| 5 |
+
import logging
|
| 6 |
+
import time
|
| 7 |
+
from tqdm import tqdm
|
| 8 |
+
import io
|
| 9 |
+
import pandas as pd
|
| 10 |
+
import numpy as np
|
| 11 |
+
import os
|
| 12 |
+
import json
|
| 13 |
+
|
| 14 |
+
import argparse
|
| 15 |
+
import torch
|
| 16 |
+
import random
|
| 17 |
+
|
| 18 |
+
from yaml import parse
|
| 19 |
+
|
| 20 |
+
from model import Conve, Distmult, Complex
|
| 21 |
+
|
| 22 |
+
logger = logging.getLogger(__name__)
|
| 23 |
+
#%%
|
| 24 |
+
def generate_dicts(data_path):
|
| 25 |
+
with open (os.path.join(data_path, 'entities_dict.json'), 'r') as f:
|
| 26 |
+
ent_to_id = json.load(f)
|
| 27 |
+
with open (os.path.join(data_path, 'relations_dict.json'), 'r') as f:
|
| 28 |
+
rel_to_id = json.load(f)
|
| 29 |
+
n_ent = len(list(ent_to_id.keys()))
|
| 30 |
+
n_rel = len(list(rel_to_id.keys()))
|
| 31 |
+
|
| 32 |
+
return n_ent, n_rel, ent_to_id, rel_to_id
|
| 33 |
+
|
| 34 |
+
def save_data(file_name, data):
|
| 35 |
+
with open(file_name, 'w') as fl:
|
| 36 |
+
for item in data:
|
| 37 |
+
fl.write("%s\n" % "\t".join(map(str, item)))
|
| 38 |
+
|
| 39 |
+
def load_data(file_name, drop = True):
|
| 40 |
+
df = pd.read_csv(file_name, sep='\t', header=None, names=None, dtype=str)
|
| 41 |
+
if drop:
|
| 42 |
+
df = df.drop_duplicates()
|
| 43 |
+
else:
|
| 44 |
+
pass
|
| 45 |
+
return df.values
|
| 46 |
+
|
| 47 |
+
def seed_all(seed=1):
|
| 48 |
+
random.seed(seed)
|
| 49 |
+
np.random.seed(seed)
|
| 50 |
+
torch.manual_seed(seed)
|
| 51 |
+
torch.cuda.manual_seed_all(seed)
|
| 52 |
+
os.environ['PYTHONHASHSEED'] = str(seed)
|
| 53 |
+
torch.backends.cudnn.deterministic = True
|
| 54 |
+
|
| 55 |
+
def add_model(args, n_ent, n_rel):
|
| 56 |
+
if args.model is None:
|
| 57 |
+
model = Distmult(args, n_ent, n_rel)
|
| 58 |
+
elif args.model == 'distmult':
|
| 59 |
+
model = Distmult(args, n_ent, n_rel)
|
| 60 |
+
elif args.model == 'complex':
|
| 61 |
+
model = Complex(args, n_ent, n_rel)
|
| 62 |
+
elif args.model == 'conve':
|
| 63 |
+
model = Conve(args, n_ent, n_rel)
|
| 64 |
+
else:
|
| 65 |
+
raise Exception("Unknown model!")
|
| 66 |
+
|
| 67 |
+
return model
|
| 68 |
+
|
| 69 |
+
def load_model(model_path, args, n_ent, n_rel, device):
|
| 70 |
+
# add a model and load the pre-trained params
|
| 71 |
+
model = add_model(args, n_ent, n_rel)
|
| 72 |
+
model.to(device)
|
| 73 |
+
logger.info('Loading saved model from {0}'.format(model_path))
|
| 74 |
+
state = torch.load(model_path)
|
| 75 |
+
model_params = state['state_dict']
|
| 76 |
+
params = [(key, value.size(), value.numel()) for key, value in model_params.items()]
|
| 77 |
+
for key, size, count in params:
|
| 78 |
+
logger.info('Key:{0}, Size:{1}, Count:{2}'.format(key, size, count))
|
| 79 |
+
|
| 80 |
+
model.load_state_dict(model_params)
|
| 81 |
+
model.eval()
|
| 82 |
+
logger.info(model)
|
| 83 |
+
|
| 84 |
+
return model
|
| 85 |
+
|
| 86 |
+
def add_eval_parameters(parser):
|
| 87 |
+
|
| 88 |
+
# parser.add_argument('--eval-mode', type = str, default = 'all', help = 'Method to evaluate the attack performance. Default: all. (all or single)')
|
| 89 |
+
parser.add_argument('--cuda-name', type = str, required = True, help = 'Start a main thread on each cuda.')
|
| 90 |
+
parser.add_argument('--direct', action='store_true', help = 'Directly add edge or not.')
|
| 91 |
+
parser.add_argument('--seperate', action='store_true', help = 'Evaluate seperatly or not')
|
| 92 |
+
parser.add_argument('--mode', type = str, default = '', help = ' '' or '' ')
|
| 93 |
+
parser.add_argument('--mask-ratio', type=str, default='', help='Mask ratio for Fig4b')
|
| 94 |
+
return parser
|
| 95 |
+
|
| 96 |
+
def add_attack_parameters(parser):
|
| 97 |
+
|
| 98 |
+
# parser.add_argument('--target-split', type=str, default='0_100_1', help='Ranks to use for target set. Values are 0 for ranks==1; 1 for ranks <=10; 2 for ranks>10 and ranks<=100. Default: 1')
|
| 99 |
+
parser.add_argument('--target-split', type=str, default='min', help='Methods for target triple selection. Default: min. (min or top_?, top means top_0.1)')
|
| 100 |
+
parser.add_argument('--target-size', type=int, default=50, help='Number of target triples. Default: 50')
|
| 101 |
+
parser.add_argument('--target-existed', action='store_true', help='Whether the targeted s_?_o already exists.')
|
| 102 |
+
|
| 103 |
+
# parser.add_argument('--budget', type=int, default=1, help='Budget for each target triple for each corruption side')
|
| 104 |
+
|
| 105 |
+
parser.add_argument('--attack-goal', type = str, default='single', help='Attack goal. Default: single. (single or global)')
|
| 106 |
+
parser.add_argument('--neighbor-num', type = int, default=20, help='Max neighbor num for each side. Default: 20')
|
| 107 |
+
parser.add_argument('--candidate-mode', type = str, default='quadratic', help = 'The method to generate candidate edge. Default: quadratic. (quadratic or linear)')
|
| 108 |
+
parser.add_argument('--reasonable-rate', type = float, default=0.7, help = 'The added edge\'s existance rank prob greater than this rate')
|
| 109 |
+
parser.add_argument('--added-edge-num', type = str, default='', help = 'How many edges to add for each target edge. Default: '' means 1.')
|
| 110 |
+
# parser.add_argument('--neighbor-num', type = int, default=200, help='Max neighbor num for each side. Default: 200')
|
| 111 |
+
# parser.add_argument('--candidate-mode', type = str, default='linear', help = 'The method to generate candidate edge. Default: quadratic. (quadratic or linear)')
|
| 112 |
+
parser.add_argument('--attack-batch-size', type=int, default=256, help='Batch size for processing neighbours of target')
|
| 113 |
+
parser.add_argument('--template-mode', type=str, default = 'manual', help = 'Template mode for transforming edge to single sentense. Default: manual. (manual or auto)')
|
| 114 |
+
|
| 115 |
+
parser.add_argument('--update-lissa', action='store_true', help = 'Update lissa cache or not.')
|
| 116 |
+
|
| 117 |
+
parser.add_argument('--GPT-batch-size', type=int, default = 64, help = 'Batch size for GPT2 when calculating LM score. Default: 64')
|
| 118 |
+
parser.add_argument('--LM-softmax', action='store_true', help = 'Use a softmax head on LM prob or not.')
|
| 119 |
+
parser.add_argument('--LMprob-mode', type=str, default='relative', help = 'Use the absolute LM score or calculate the destruction score when target word is replaced. Default: absolute. (absolute or relative)')
|
| 120 |
+
|
| 121 |
+
parser.add_argument('--load-existed', action='store_true', help = 'Use cached intermidiate results or not, when only --reasonable-rate changed, set this param to True')
|
| 122 |
+
|
| 123 |
+
return parser
|
| 124 |
+
|
| 125 |
+
def get_argument_parser():
|
| 126 |
+
'''Generate an argument parser'''
|
| 127 |
+
parser = argparse.ArgumentParser(description='Graph embedding')
|
| 128 |
+
|
| 129 |
+
parser.add_argument('--seed', type=int, default=1, metavar='S', help='Random seed (default: 1)')
|
| 130 |
+
|
| 131 |
+
parser.add_argument('--data', type=str, default='GNBR', help='Dataset to use: { GNBR }')
|
| 132 |
+
parser.add_argument('--model', type=str, default='distmult', help='Choose from: {distmult, conve, complex}')
|
| 133 |
+
|
| 134 |
+
parser.add_argument('--transe-margin', type=float, default=0.0, help='Margin value for TransE scoring function. Default:0.0')
|
| 135 |
+
parser.add_argument('--transe-norm', type=int, default=2, help='P-norm value for TransE scoring function. Default:2')
|
| 136 |
+
|
| 137 |
+
parser.add_argument('--epochs', type=int, default=100, help='Number of epochs to train (default: 100)')
|
| 138 |
+
parser.add_argument('--lr', type=float, default=0.001, help='Learning rate (default: 0.001)')
|
| 139 |
+
parser.add_argument('--lr-decay', type=float, default=0.0, help='Weight decay value to use in the optimizer. Default: 0.0')
|
| 140 |
+
parser.add_argument('--max-norm', action='store_true', help='Option to add unit max norm constraint to entity embeddings')
|
| 141 |
+
|
| 142 |
+
parser.add_argument('--train-batch-size', type=int, default=64, help='Batch size for train split (default: 128)')
|
| 143 |
+
parser.add_argument('--test-batch-size', type=int, default=128, help='Batch size for test split (default: 128)')
|
| 144 |
+
parser.add_argument('--valid-batch-size', type=int, default=128, help='Batch size for valid split (default: 128)')
|
| 145 |
+
parser.add_argument('--KG-valid-rate', type = float, default=0.1, help='Validation rate during KG embedding training. (default: 0.1)')
|
| 146 |
+
|
| 147 |
+
parser.add_argument('--save-influence-map', action='store_true', help='Save the influence map during training for gradient rollback.')
|
| 148 |
+
parser.add_argument('--add-reciprocals', action='store_true')
|
| 149 |
+
|
| 150 |
+
parser.add_argument('--embedding-dim', type=int, default=128, help='The embedding dimension (1D). Default: 128')
|
| 151 |
+
parser.add_argument('--stack-width', type=int, default=16, help='The first dimension of the reshaped/stacked 2D embedding. Second dimension is inferred. Default: 20')
|
| 152 |
+
#parser.add_argument('--stack_height', type=int, default=10, help='The second dimension of the reshaped/stacked 2D embedding. Default: 10')
|
| 153 |
+
parser.add_argument('--hidden-drop', type=float, default=0.3, help='Dropout for the hidden layer. Default: 0.3.')
|
| 154 |
+
parser.add_argument('--input-drop', type=float, default=0.2, help='Dropout for the input embeddings. Default: 0.2.')
|
| 155 |
+
parser.add_argument('--feat-drop', type=float, default=0.3, help='Dropout for the convolutional features. Default: 0.2.')
|
| 156 |
+
parser.add_argument('-num-filters', default=32, type=int, help='Number of filters for convolution')
|
| 157 |
+
parser.add_argument('-kernel-size', default=3, type=int, help='Kernel Size for convolution')
|
| 158 |
+
|
| 159 |
+
parser.add_argument('--use-bias', action='store_true', help='Use a bias in the convolutional layer. Default: True')
|
| 160 |
+
|
| 161 |
+
parser.add_argument('--reg-weight', type=float, default=5e-2, help='Weight for regularization. Default: 5e-2')
|
| 162 |
+
parser.add_argument('--reg-norm', type=int, default=3, help='Norm for regularization. Default: 2')
|
| 163 |
+
# parser.add_argument('--resume', action='store_true', help='Restore a saved model.')
|
| 164 |
+
# parser.add_argument('--resume-split', type=str, default='test', help='Split to evaluate a restored model')
|
| 165 |
+
# parser.add_argument('--reproduce-results', action='store_true', help='Use the hyperparameters to reproduce the results.')
|
| 166 |
+
# parser.add_argument('--original-data', type=str, default='FB15k-237', help='Dataset to use; this option is needed to set the hyperparams to reproduce the results for training after attack, default: FB15k-237')
|
| 167 |
+
return parser
|
| 168 |
+
|
| 169 |
+
def set_hyperparams(args):
|
| 170 |
+
if args.model == 'distmult':
|
| 171 |
+
args.lr = 0.005
|
| 172 |
+
args.train_batch_size = 1024
|
| 173 |
+
args.reg_norm = 3
|
| 174 |
+
elif args.model == 'complex':
|
| 175 |
+
args.lr = 0.005
|
| 176 |
+
args.reg_norm = 3
|
| 177 |
+
args.input_drop = 0.4
|
| 178 |
+
args.train_batch_size = 1024
|
| 179 |
+
elif args.model == 'conve':
|
| 180 |
+
args.lr = 0.005
|
| 181 |
+
args.train_batch_size = 1024
|
| 182 |
+
args.reg_weight = 0.0
|
| 183 |
+
|
| 184 |
+
# args.damping = 0.01
|
| 185 |
+
# args.lissa_repeat = 1
|
| 186 |
+
# args.lissa_depth = 1
|
| 187 |
+
# args.scale = 500
|
| 188 |
+
# args.lissa_batch_size = 100
|
| 189 |
+
|
| 190 |
+
args.damping = 0.01
|
| 191 |
+
args.lissa_repeat = 1
|
| 192 |
+
args.lissa_depth = 1
|
| 193 |
+
args.scale = 400
|
| 194 |
+
args.lissa_batch_size = 300
|
| 195 |
+
return args
|