forked from wutong8023/AAAI-CML
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel.py
211 lines (180 loc) · 9.48 KB
/
model.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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
'''
This code is based on the Pytorch Orientaion:
https://pytorch.org/tutorials/beginner/nlp/sequence_models_tutorial.html#sphx-glr-beginner-nlp-sequence-models-tutorial-py
Original Author: Robert Guthrie
'''
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
torch.manual_seed(1)
class BiLSTM(nn.Module):
def __init__(self, embedding_dim, hidden_dim, vocab_size, vocab_embedding,
batch_size, device):
super(BiLSTM, self).__init__()
self.hidden_dim = hidden_dim
self.batch_size = batch_size
# The LSTM takes word embeddings as inputs, and outputs hidden states
# with dimensionality hidden_dim.
self.lstm = nn.LSTM(embedding_dim, hidden_dim, bidirectional=True).to(device)
#self.maxpool = nn.MaxPool1d(hidden_dim*2)
self.hidden = self.build_hidden()
def build_hidden(self, batch_size = 1):
# Before we've done anything, we dont have any hidden state.
# Refer to the Pytorch documentation to see exactly
# why they have this dimensionality.
# The axes semantics are (num_layers, minibatch_size, hidden_dim)
#print(batch_size)
return [torch.zeros(2, batch_size, self.hidden_dim),
torch.zeros(2, batch_size, self.hidden_dim)]
def init_hidden(self, device='cpu', batch_size = 1):
# Before we've done anything, we dont have any hidden state.
# Refer to the Pytorch documentation to see exactly
# why they have this dimensionality.
# The axes semantics are (num_layers, minibatch_size, hidden_dim)
#print(batch_size)
self.hidden = (torch.zeros(2, batch_size, self.hidden_dim).to(device),
torch.zeros(2, batch_size, self.hidden_dim).to(device))
#self.hidden[0] = self.hidden[0].to(device)
#self.hidden[1] = self.hidden[1].to(device)
def forward(self, packed_embeds):
# 句子和关系的编码方法
#print(packed_embeds)
#print(self.hidden)
lstm_out, self.hidden = self.lstm(packed_embeds, self.hidden)
#maxpool_hidden = self.maxpool(lstm_out.view(1,len(sentence), -1))
permuted_hidden = self.hidden[0].permute([1,0,2]).contiguous()
#print(permuted_hidden.size())
return permuted_hidden.view(-1, self.hidden_dim*2)
class SimilarityModel(nn.Module):
def __init__(self, embedding_dim, hidden_dim, vocab_size, vocab_embedding,
batch_size, device):
super(SimilarityModel, self).__init__()
self.batch_size = batch_size
self.word_embeddings = nn.Embedding(vocab_size, embedding_dim)
self.word_embeddings.weight.data.copy_(torch.from_numpy(vocab_embedding))
self.word_embeddings = self.word_embeddings.to(device)
self.word_embeddings.weight.requires_grad = False
self.sentence_biLstm = BiLSTM(embedding_dim, hidden_dim, vocab_size,
vocab_embedding, batch_size, device)
self.relation_biLstm = BiLSTM(embedding_dim, hidden_dim, vocab_size,
vocab_embedding, batch_size, device)
def init_hidden(self, device, batch_size=1):
self.sentence_biLstm.init_hidden(device, batch_size)
self.relation_biLstm.init_hidden(device, batch_size)
def init_embedding(self, vocab_embedding):
#print(self.word_embeddings(torch.tensor([27]).cuda()))
self.word_embeddings.weight.data.copy_(torch.from_numpy(vocab_embedding))
#print(self.word_embeddings(torch.tensor([27]).cuda()))
def ranking_sequence(self, sequence):
word_lengths = torch.tensor([len(sentence) for sentence in sequence])
rankedi_word, indexs = word_lengths.sort(descending = True)
ranked_indexs, inverse_indexs = indexs.sort()
#print(indexs)
sequence = [sequence[i] for i in indexs]
return sequence, inverse_indexs
def compute_que_embed(self, question_list, question_lengths,
reverse_question_indexs, reverse_model,
before_reverse=False):
question_embeds = self.word_embeddings(question_list)
question_packed = \
torch.nn.utils.rnn.pack_padded_sequence(question_embeds,
question_lengths)
question_embedding = self.sentence_biLstm(question_packed)
question_embedding = question_embedding[reverse_question_indexs]
if reverse_model is not None and not before_reverse:
return reverse_model(question_embedding).detach()
else:
return question_embedding.detach()
def compute_rel_embed(self, relation_list, relation_lengths,
reverse_relation_indexs, reverse_model,
before_reverse=False):
relation_embeds = self.word_embeddings(relation_list)
relation_packed = \
torch.nn.utils.rnn.pack_padded_sequence(relation_embeds,
relation_lengths)
relation_embedding = self.relation_biLstm(relation_packed)
relation_embedding = relation_embedding[reverse_relation_indexs]
if reverse_model is not None and not before_reverse:
return reverse_model(relation_embedding).detach()
else:
return relation_embedding.detach()
def forward(self, seq_list_1, seq_list_2, device,
seq_list_1_idx, seq_list_2_idx,
seq_list_1_lengths, seq_list_2_lengths, reverse_model=None, contrastive=False):
# shape of question_list: (36, 128) 36 is the maximum length of questions, 128 is the batch size
seq_list_1_embeds = self.word_embeddings(seq_list_1)
seq_list_2_embeds = self.word_embeddings(seq_list_2)
#print(question_lengths)
seq_list_1_packed = \
torch.nn.utils.rnn.pack_padded_sequence(seq_list_1_embeds,
seq_list_1_lengths)
seq_list_2_packed = \
torch.nn.utils.rnn.pack_padded_sequence(seq_list_2_embeds,
seq_list_2_lengths)
seq_list_1_embedding = self.sentence_biLstm(seq_list_1_packed) # shape
if contrastive:
seq_list_2_embedding = self.sentence_biLstm(seq_list_2_packed)
else:
seq_list_2_embedding = self.relation_biLstm(seq_list_2_packed)
seq_list_1_embedding = seq_list_1_embedding[seq_list_1_idx]
seq_list_2_embedding = seq_list_2_embedding[seq_list_2_idx]
if reverse_model is not None:
reverse_seq_list_1_embedding = reverse_model(seq_list_1_embedding)
reverse_seq_list_2_embedding = reverse_model(seq_list_2_embedding)
cos = nn.CosineSimilarity(dim=1)
origin_score = cos(seq_list_1_embedding, seq_list_2_embedding)
reverse_score = cos(reverse_seq_list_1_embedding,
reverse_seq_list_2_embedding)
avg_pooling = torch.mean(torch.stack((origin_score,
reverse_score)),0)
return reverse_score
#return max_pooling
else:
cos = nn.CosineSimilarity(dim=1)
return cos(seq_list_1_embedding, seq_list_2_embedding)
class PCNN_Encoder(nn.Module):
def __init__(self):
super(PCNN_Encoder, self).__init__()
class PCNNModel(nn.Module):
def __init__(self, embedding_dim, hidden_dim, vocab_size, vocab_embedding,
batch_size, device, pos_limit=60, pos_dim=10):
super(PCNNModel, self).__init__()
# hyperparams
# set glove embeddings
self.word_embeddings = nn.Embedding(vocab_size, embedding_dim, padding_idx=0)
self.word_embeddings.weight.data.copy_(torch.from_numpy(vocab_embedding))
self.word_embeddings = self.word_embeddings.to(device)
self.word_embeddings.weight.requires_grad = False
# set pos embeddings
# pos_size = 2 * pos_limit + 2, 0 for padding
self.headPosEmbed = nn.Embedding(2 * pos_limit + 2, pos_dim, padding_idx=0) # pos_size = max_length
self.tailPosEmbed = nn.Embedding(2 * pos_limit + 2, pos_dim, padding_idx=0)
self.conv = nn.Conv1d(embedding_dim + pos_dim * 2, 100, 3)
self.pool = nn.MaxPool1d(120)
# set mask embeddings
self.mask_embedding = nn.Embedding(4, 3)
self.mask_embedding.weight.data.copy_(torch.FloatTensor([[0, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]]))
self.mask_embedding.weight.requires_grad = False
self._minus = -100
# relation encoder use BiLSTM
self.relation_biLstm = BiLSTM(embedding_dim, hidden_dim, vocab_size,
vocab_embedding, batch_size, device)
def init_hidden(self, device, batch_size=1):
pass
def init_embedding(self, vocab_embedding):
pass
def ranking_sequence(self, sequence):
pass
def compute_que_embed(self, question_list, question_lengths,
reverse_question_indexs, reverse_model,
before_reverse=False):
pass
def compute_rel_embed(self, relation_list, relation_lengths,
reverse_relation_indexs, reverse_model,
before_reverse=False):
pass
def forward(self, question_list, relation_list, device,
reverse_question_indexs, reverse_relation_indexs,
question_lengths, relation_lengths, reverse_model=None):
pass