-
Notifications
You must be signed in to change notification settings - Fork 10
/
config.py
164 lines (144 loc) · 6.42 KB
/
config.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
import pickle
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
from sklearn.metrics import f1_score, precision_score, recall_score
from dmcnn import dmcnn_t
from loader import load_word2vec
from loader import Batch_tri, Batch_arg
from loader import load_tri_sentences, load_arg_sentences
def to_var(x):
return Variable(torch.from_numpy(x).long().cuda())
class Config(object):
def __init__(self):
self.gpu = "4"
self.path_t = 'data/tri.train'
self.path_a = 'data/arg.train'
self.path_test_t = 'data/tri.test'
self.path_test_a = 'data/arg.test'
self.path_modelt = 'data/modelt'
self.path_debug = 'data/debug'
self.lr = 1
self.weight_decay = 1e-5
self.epoch = 75 # training epoches
self.epoch_save = 1
self.sen = 80 # sentence length
self.char_dim = 100 # length of word embedding tensor
self.num_char = 20136 # total num of word2vec model
self.batch_t = 170 # num of sentences in one batch
self.batch_a = 20
self.num_t = 34 # num of triggers
self.num_a = 36
self.pf_t = 5 # dim of pf in trigger classification
self.pf_a = 5
self.ef_a = 5
self.window_t = 3 # window size in cnn
self.window_a = 3
self.feature_t = 200 # num of features in cnn
self.feature_a = 300
def load_traint_data(self):
print("Reading training data...")
train_t = load_tri_sentences(self.path_t)
self.train_t_b = Batch_tri(train_t, self.batch_t, self.sen)
self.emb_weights = load_word2vec("data/100.utf8", 100, self.num_char, self.char_dim)
print("finish reading")
def load_testt_data(self):
print("Reading testing data...")
test_t = load_tri_sentences(self.path_test_t)
self.test_t_b = Batch_tri(test_t, self.batch_t, self.sen)
print("finish reading")
def set_traint_model(self):
print("Initializing training model...")
self.modelt = dmcnn_t(config=self)
self.optimizer_t = optim.Adadelta(self.modelt.parameters(), lr=self.lr, rho=0.95, eps=1e-6, weight_decay=self.weight_decay)
self.modelt.cuda()
for param_tensor in self.modelt.state_dict():
print(param_tensor, "\t", self.modelt.state_dict()[param_tensor].size())
print("Finish initializing")
def set_testt_model(self):
print("Initializing testing model...")
self.model_test_t = dmcnn_t(config=self)
self.model_test_t.cuda()
self.model_test_t.eval()
print("finish initializing")
def train_one_step(self, batch):
self.modelt.char_inputs = to_var(np.array(batch[0]))
self.modelt.trigger_inputs = to_var(np.array(batch[1]))
self.modelt.pf_inputs = to_var(np.array(batch[2]))
self.modelt.lxl_inputs = to_var(np.array(batch[3]))
self.modelt.masks = to_var(np.array(batch[4]))
self.modelt.cuts = to_var(np.array(batch[5]))
self.optimizer_t.zero_grad()
loss, maxes= self.modelt()
loss.backward()
self.optimizer_t.step()
return loss.data, maxes
def test_one_step(self, batch):
self.model_test_t.char_inputs = to_var(np.array(batch[0]))
self.model_test_t.trigger_inputs = to_var(np.array(batch[1]))
self.model_test_t.pf_inputs = to_var(np.array(batch[2]))
self.model_test_t.lxl_inputs = to_var(np.array(batch[3]))
self.model_test_t.masks = to_var(np.array(batch[4]))
self.model_test_t.cuts = to_var(np.array(batch[5]))
loss, maxes = self.model_test_t()
return loss, maxes
def train(self):
for epoch in range(self.epoch):
losses = 0
tru = pre = None
i = 0
print("epoch: ", epoch)
for batch in self.train_t_b.iter_batch():
loss, maxes = self.train_one_step(batch)
losses += loss
if i == 0:
tru = self.modelt.trigger_inputs
pre = maxes
else:
tru = torch.cat((tru, self.modelt.trigger_inputs), dim=0)
pre = torch.cat((pre, maxes), dim=0)
i += 1
tru = tru.cpu()
pre = pre.cpu()
prec = precision_score(tru, pre, labels=list(range(1, 34)), average='micro')
rec = recall_score(tru, pre, labels=list(range(1, 34)), average='micro')
f1 = f1_score(tru, pre, labels=list(range(1, 34)), average='micro')
i = 0
if epoch % self.epoch_save == 0:
torch.save(self.modelt.state_dict(), self.path_modelt)
print("loss_average:", losses/i)
print("Precision: ", prec)
print("Recall: ", rec)
print("FMeasure", f1)
def test(self):
self.model_test_t.load_state_dict(torch.load(self.path_modelt))
tru = pre = None
i = 0
with open(self.path_debug, 'w') as f:
losses = 0
for batch in self.test_t_b.iter_batch():
loss, maxes = self.test_one_step(batch)
losses += loss
if i == 0:
tru = self.model_test_t.trigger_inputs
pre = maxes
else:
tru = torch.cat((tru, self.model_test_t.trigger_inputs), dim=0)
pre = torch.cat((pre, maxes), dim=0)
i += 1
tru = tru.cpu()
pre = pre.cpu()
tru_n = tru.numpy()
pre_n = pre.numpy()
for p in range(self.batch_t*self.test_t_b.len_data):
if tru_n[p] != pre_n[p]:
f.write(str(tru_n[p]) + ':' + str(pre_n[p]) + '\n')
prec = precision_score(tru, pre, labels=list(range(1, 34)), average='micro')
rec = recall_score(tru, pre, labels=list(range(1, 34)), average='micro')
f1 = f1_score(tru, pre, labels=list(range(1, 34)), average='micro')
print("loss_average: ", losses/i)
print("Precision: ", prec)
print("Recall: ", rec)
print("FMeasure", f1)