-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_attention.py
192 lines (138 loc) · 5.73 KB
/
model_attention.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
186
187
188
189
190
191
192
import dgl
import torch
import torch.nn.functional as F
import numpy as np
import pickle
import gensim
class Model(torch.nn.Module):
def __init__(self,
class_num,
hidden_size_node,
vocab,
n_gram,
drop_out,
edges_num,
edges_matrix,
max_length=300,
trainable_edges=True,
pmi=None,
cuda=True,
):
super(Model, self).__init__()
self.is_cuda = cuda
self.vocab = vocab
print(edges_num)
self.node_hidden = torch.nn.Embedding(len(vocab), hidden_size_node)
self.node_eta = torch.nn.Embedding.from_pretrained(torch.rand(len(vocab), 1), freeze=False)
# self.seq_edge_w = torch.nn.Embedding.from_pretrained(pmi, freeze=True)
self.edges_num = edges_num
# self.seq_edge_w = torch.nn.Embedding.from_pretrained(torch.rand(edges_num, 1), freeze=False)
self.hidden_size_node = hidden_size_node
self.node_hidden.weight.data.copy_(torch.tensor(self.load_word2vec('/content/glove.6B.300d.txt')))
self.node_hidden.weight.requires_grad = True
self.len_vocab = len(vocab)
self.ngram = n_gram
self.d = dict(zip(self.vocab, range(len(self.vocab))))
self.max_length = max_length
self.edges_matrix = edges_matrix
self.dropout = torch.nn.Dropout(p=drop_out,inplace=True)
self.activation = torch.nn.LeakyReLU(inplace=True)
self.attn_fc = torch.nn.Linear(2 * hidden_size_node, 1, bias=True)
self.Linear = torch.nn.Linear(hidden_size_node, class_num, bias=True)
def word2id(self, word):
try:
result = self.d[word]
except KeyError:
result = self.d['UNK']
return result
def load_word2vec(self, word2vec_file):
model = gensim.models.KeyedVectors.load_word2vec_format(word2vec_file, binary=False)
embedding_matrix = []
for word in self.vocab:
try:
embedding_matrix.append(model[word])
except KeyError:
# print(word)
embedding_matrix.append(np.random.uniform(-0.1,0.1,300))
embedding_matrix = np.array(embedding_matrix)
return embedding_matrix
def add_seq_edges(self, doc_ids: list, old_to_new: dict):
edges = []
old_edge_id = []
for index, src_word_old in enumerate(doc_ids):
src = old_to_new[src_word_old]
for i in range(max(0, index - self.ngram), min(index + self.ngram, len(doc_ids))):
dst_word_old = doc_ids[i]
dst = old_to_new[dst_word_old]
# - first connect the new sub_graph
edges.append([src, dst])
# - then get the hidden from parent_graph
try :
old_edge_id.append(self.edges_matrix[(src_word_old, dst_word_old)])
except KeyError:
old_edge_id.append(np.random.randint(0,self.edges_num))
# self circle
# edges.append([src, src])
# old_edge_id.append(self.edges_matrix[src_word_old, src_word_old])
return edges, old_edge_id
def seq_to_graph(self, doc_ids: list) -> dgl.DGLGraph():
if len(doc_ids) > self.max_length:
doc_ids = doc_ids[:self.max_length]
local_vocab = set(doc_ids)
old_to_new = dict(zip(local_vocab, range(len(local_vocab))))
if self.is_cuda:
local_vocab = torch.tensor(list(local_vocab)).cuda()
else:
local_vocab = torch.tensor(list(local_vocab))
sub_graph = dgl.DGLGraph()
sub_graph.add_nodes(len(local_vocab))
local_node_hidden = self.node_hidden(local_vocab)
sub_graph.ndata['h'] = local_node_hidden
sub_graph.ndata['eta'] = self.node_eta(local_vocab)
seq_edges, seq_old_edges_id = self.add_seq_edges(doc_ids, old_to_new)
edges, old_edge_id = [], []
# edges = []
edges.extend(seq_edges)
old_edge_id.extend(seq_old_edges_id)
if self.is_cuda:
old_edge_id = torch.LongTensor(old_edge_id).cuda()
else:
old_edge_id = torch.LongTensor(old_edge_id)
srcs, dsts = zip(*edges)
sub_graph.add_edges(srcs, dsts)
'''
try:
seq_edges_w = self.seq_edge_w(old_edge_id)
except RuntimeError:
print(old_edge_id)
sub_graph.edata['w'] = seq_edges_w
'''
return sub_graph
def edge_attention(self, edges):
# edge UDF for equation (2)
z2 = torch.cat([edges.src['h'], edges.dst['h']], dim=1)
a = self.attn_fc(z2)
return {'w': F.leaky_relu(a)}
def gcn_msg(self,edge):
return {'m': edge.src['h'], 'w': edge.data['w']}
def gcn_reduce(self,node):
w = F.softmax(node.mailbox['w'], dim=1)
# new_hidden = torch.sum(w * node.mailbox['m'], dim=1)
# w = node.mailbox['w']
new_hidden = torch.mul(w, node.mailbox['m'])
new_hidden,_ = torch.max(new_hidden, 1)
node_eta = node.data['eta']
new_hidden = node_eta * node.data['h'] + (1 - node_eta) * new_hidden
return {'h': new_hidden}
def forward(self, doc_ids):
sub_graphs = [self.seq_to_graph(doc) for doc in doc_ids]
batch_graph = dgl.batch(sub_graphs)
batch_graph.apply_edges(self.edge_attention)
batch_graph.update_all(
self.gcn_msg,self.gcn_reduce
)
h1 = dgl.sum_nodes(batch_graph, feat='h')
drop1 = self.dropout(h1)
act1 = self.activation(drop1)
l = self.Linear(act1)
return l