-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel_trans.py
193 lines (168 loc) · 8.57 KB
/
model_trans.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
import numpy as np
import dynet_config
dynet_config.set(
mem=2048,
random_seed=1,
# autobatch=True
)
import dynet as dy
import math
import pickle
class TNLM:
def __init__(self, vocab_size, char_size, char_embedding_dim, char_hidden_size,
word_embedding_dim, hidden_dim, pos_size, pos_embeddings_size, label_size,
pattern_hidden_dim, pattern_embeddings_dim, rule_size, max_rule_length,
lstm_num_layers, pretrained):
self.vocab_size = vocab_size
self.char_size = char_size
self.word_embedding_dim = word_embedding_dim
self.char_embedding_dim = char_embedding_dim
self.hidden_dim = hidden_dim
self.model = dy.Model()
self.trainer = dy.SimpleSGDTrainer(self.model)
self.label_size = label_size
self.char_hidden_size = char_hidden_size
self.pos_size = pos_size
self.pos_embeddings_size = pos_embeddings_size
self.pretrained = pretrained
self.pattern_hidden_dim = pattern_hidden_dim
self.pattern_embeddings_dim = pattern_embeddings_dim
self.rule_size = rule_size
self.max_rule_length = max_rule_length
self.lstm_num_layers = lstm_num_layers
if np.any(self.pretrained):
self.word_embeddings = self.model.lookup_parameters_from_numpy(self.pretrained)
else:
self.word_embeddings = self.model.add_lookup_parameters((self.vocab_size, self.word_embedding_dim))
self.pos_embeddings = self.model.add_lookup_parameters((self.pos_size, self.pos_embeddings_size))
self.char_embeddings = self.model.add_lookup_parameters((self.char_size, self.char_embedding_dim))
self.character_lstm = dy.BiRNNBuilder(
self.lstm_num_layers,
self.char_embedding_dim,
self.char_hidden_size,
self.model,
dy.VanillaLSTMBuilder,
)
self.weight_eq = self.model.add_parameters((2 * self.hidden_dim,
self.word_embedding_dim + self.char_hidden_size + self.pos_embeddings_size))
self.weight_ek = self.model.add_parameters((2 * self.hidden_dim,
self.word_embedding_dim + self.char_hidden_size + self.pos_embeddings_size))
self.weight_ev = self.model.add_parameters((2 * self.hidden_dim,
self.word_embedding_dim + self.char_hidden_size + self.pos_embeddings_size))
self.eff = self.model.add_parameters((self.hidden_dim, 2 * self.hidden_dim))
self.eff_bias = self.model.add_parameters((self.hidden_dim))
self.pattern_embeddings = self.model.add_lookup_parameters((self.rule_size, self.pattern_embeddings_dim))
self.weight_dq = self.model.add_parameters((2 * self.hidden_dim, self.pattern_embeddings_dim))
self.weight_dk = self.model.add_parameters((2 * self.hidden_dim, self.pattern_embeddings_dim))
self.weight_dv = self.model.add_parameters((2 * self.hidden_dim, self.pattern_embeddings_dim))
self.dff = self.model.add_parameters((self.hidden_dim, 2 * self.hidden_dim))
self.dff_bias = self.model.add_parameters((self.hidden_dim))
self.weight_q = self.model.add_parameters((self.pattern_hidden_dim, self.hidden_dim))
self.weight_k = self.model.add_parameters((self.pattern_hidden_dim, self.hidden_dim))
self.weight_v = self.model.add_parameters((self.pattern_hidden_dim, self.hidden_dim))
self.pt = self.model.add_parameters((self.rule_size, self.pattern_hidden_dim))
self.pt_bias = self.model.add_parameters((self.rule_size))
self.lb = self.model.add_parameters((self.hidden_dim, 2 * self.hidden_dim))
self.lb_bias = self.model.add_parameters((self.hidden_dim))
self.lb2 = self.model.add_parameters((1, self.hidden_dim))
self.lb2_bias = self.model.add_parameters((1))
self.self_attention_weight = self.model.add_parameters((1, self.hidden_dim))
def save(self, name):
params = (
self.vocab_size, self.char_size, self.char_embedding_dim, self.char_hidden_size,
self.word_embedding_dim, self.hidden_dim, self.pos_size, self.pos_embeddings_size,
self.label_size, self.pattern_hidden_dim, self.pattern_embeddings_dim,
self.rule_size, self.max_rule_length, self.lstm_num_layers, self.pretrained
)
# save model
self.model.save(f'{name}.model')
# save pickle
with open(f'{name}.pickle', 'wb') as f:
pickle.dump(params, f)
@staticmethod
def load(name):
with open(f'{name}.pickle', 'rb') as f:
params = pickle.load(f)
parser = LSTMLM(*params)
parser.model.populate(f'{name}.model')
return parser
def char_encode(self, word):
c_seq = [self.char_embeddings[c] for c in word]
return self.character_lstm.transduce(c_seq)[-1]
def encode_sentence(self, sentence, pos, chars):
embeds_sent = [dy.concatenate([self.word_embeddings[sentence[i]], self.char_encode(chars[i]), self.pos_embeddings[pos[i]]])
for i in range(len(sentence))]
embeds = dy.concatenate_cols(embeds_sent)
scores = []
for embed in embeds_sent:
q = self.weight_eq.expr() * embed
k = self.weight_ek.expr() * embed
scores.append(q*dy.transpose(k)/math.sqrt(self.hidden_dim * 2))
scores = dy.softmax(dy.concatenate_cols(scores))
V = self.weight_ev.expr() * embeds
features = dy.cmult(scores, V)
features = self.eff.expr() * features + self.eff_bias.expr()
return features
def self_attend(self, H_e):
H_e = dy.concatenate_cols(H_e)
S = self.self_attention_weight.expr() * H_e
S = dy.transpose(S)
A = dy.softmax(S)
context_vector = H_e * A
return A, context_vector
def decoder(features, pres):
encode = dy.concatenate_cols(features)
decoded = [self.pattern_embeddings[p] for p in pres]
decoded2 = dy.concatenate_cols(pres)
scores = []
for embed in decoded:
q = self.weight_eq.expr() * embed
k = self.weight_ek.expr() * embed
scores.append(q*dy.transpose(k)/math.sqrt(self.hidden_dim * 2))
scores = dy.softmax(dy.concatenate_cols(scores))
V = self.weight_dv.expr() * decoded2
Q2 = self.dff.expr() * (scores * V) + self.dff_bias.expr()
Q2 = self.weight_q.expr() * Q2
K2 = self.weight_k.expr() * encode
V2 = self.weight_v.expr() * encode
output = dy.softmax(Q*dy.transpose(K)/math.sqrt(self.hidden_dim * 2)) * V
output = self.pt.expr() * output + self.pt_bias.expr()
return dy.softmax(output)
def train(self, trainning_set):
for sentence, eid, entity, trigger, label, pos, chars, rule in trainning_set:
features = self.encode_sentence(sentence, pos, chars)
loss = []
entity_embeds = features[entity]
attention, context = self.self_attend(features)
ty = dy.vecInput(len(sentence))
ty.set([0 if i!=trigger else 1 for i in range(len(sentence))])
loss.append(dy.binary_log_loss(dy.reshape(attention,(len(sentence),)), ty))
h_t = dy.concatenate([context, entity_embeds])
hidden = dy.tanh(self.lb.expr() * h_t + self.lb_bias.expr())
out_vector = dy.reshape(dy.logistic(self.lb2.expr() * hidden + self.lb2_bias.expr()), (1,))
label = dy.scalarInput(label)
loss.append(dy.binary_log_loss(out_vector, label))
pres = [0]
for pattern in rule:
probs = self.decoder(features, pres)
loss.append(-dy.log(dy.pick(probs, pattern)))
pres.append(pattern)
loss = dy.esum(loss)
loss.backward()
self.trainer.update()
dy.renew_cg()
def get_pred(self, sentence, pos, chars, entity):
features = self.encode_sentence(sentence, pos, chars)
entity_embeds = features[entity]
attention, context = self.self_attend(features)
attention = attention.vec_value()
# pred_trigger = attention.index(max(attention))
h_t = dy.concatenate([context, entity_embeds])
hidden = dy.tanh(self.lb.expr() * h_t + self.lb_bias.expr())
out_vector = dy.reshape(dy.logistic(self.lb2.expr() * hidden + self.lb2_bias.expr()), (1,))
res = 1 if out_vector.npvalue() > 0.0005 else 0
rule = [0]
while rule[-1] != 0:
probs = self.decoder(features, rule)
rule.append(probs.index(max(probs)))
return attention, res, out_vector.npvalue(), rule[1:]