-
Notifications
You must be signed in to change notification settings - Fork 9
/
config.py
342 lines (311 loc) · 14.9 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
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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
# -*- coding: utf-8 -*-
import os
import sys
import datetime
import numpy as np
from tqdm import tqdm
import sklearn.metrics
import torch
import torch.optim as optim
from torch.autograd import Variable
import ipdb
def to_var(x):
return Variable(torch.from_numpy(x).cuda())
class Accuracy(object):
def __init__(self):
self.correct = 0
self.total = 0
def add(self, is_correct):
self.total += 1
if is_correct:
self.correct += 1
def get(self):
if self.total == 0:
return 0.0
else:
return float(self.correct) / self.total
def clear(self):
self.correct = 0
self.total = 0
class Config(object):
def __init__(self, batch_size=64, max_epoch=15, num_class=15, max_length=120, drop_prob=0.5, is_training=True):
self.acc_NA = Accuracy()
self.acc_not_NA = Accuracy()
self.acc_total = Accuracy()
self.data_path = './_data' # path to save data
self.checkpoint_dir = './_checkpoint' # path to save models
self.test_result_dir = './_test_result' # path to save results
self.use_bag = True
self.use_gpu = True
self.trainModel = None
self.testModel = None
self.is_training = is_training
self.pretrain_model = None # pre-trained model path
self.num_classes = num_class
self.batch_size = batch_size
self.max_length = max_length
self.word_size = 200
self.pos_num = 2 * self.max_length
self.pos_size = 5
self.window_size = 3
self.hidden_size = 200
self.opt_method = "SGD"
self.optimizer = None
self.learning_rate = 0.5
self.weight_decay = 1e-5
self.lr_decay = 0.0
self.drop_prob = drop_prob
self.max_epoch = max_epoch
self.save_epoch = 1 # save training model in every save_epoch
self.valid_epoch = 1 # do validation in every valid_epoch
self.epoch_range = None
def set_pretrain_model(self, pretrain_model):
self.pretrain_model = pretrain_model
def set_epoch_range(self, epoch_range):
self.epoch_range = epoch_range
def load_train_data(self):
print("Reading training data...")
self.data_word_vec = np.load(os.path.join(self.data_path, 'vec.npy'))
self.data_train_word = np.load(os.path.join(self.data_path, 'train_word.npy'))
self.data_train_pos1 = np.load(os.path.join(self.data_path, 'train_pos1.npy'))
self.data_train_pos2 = np.load(os.path.join(self.data_path, 'train_pos2.npy'))
self.data_train_mask = np.load(os.path.join(self.data_path, 'train_mask.npy'))
if self.use_bag:
self.data_query_label = np.load(os.path.join(self.data_path, 'train_ins_label.npy'))
self.data_train_label = np.load(os.path.join(self.data_path, 'train_bag_label.npy'))
self.data_train_scope = np.load(os.path.join(self.data_path, 'train_bag_scope.npy'))
else:
self.data_train_label = np.load(os.path.join(self.data_path, 'train_ins_label.npy'))
self.data_train_scope = np.load(os.path.join(self.data_path, 'train_ins_scope.npy'))
print("Finish reading")
self.train_order = list(range(len(self.data_train_label)))
self.train_batches = len(self.data_train_label) / self.batch_size
if len(self.data_train_label) % self.batch_size != 0:
self.train_batches += 1
def load_test_data(self):
print("Reading testing data...")
self.data_word_vec = np.load(os.path.join(self.data_path, 'vec.npy'))
self.data_test_word = np.load(os.path.join(self.data_path, 'test_word.npy'))
self.data_test_pos1 = np.load(os.path.join(self.data_path, 'test_pos1.npy'))
self.data_test_pos2 = np.load(os.path.join(self.data_path, 'test_pos2.npy'))
self.data_test_mask = np.load(os.path.join(self.data_path, 'test_mask.npy'))
if self.use_bag:
self.data_test_label = np.load(os.path.join(self.data_path, 'test_bag_label.npy'))
self.data_test_scope = np.load(os.path.join(self.data_path, 'test_bag_scope.npy'))
else:
self.data_test_label = np.load(os.path.join(self.data_path, 'test_ins_label.npy'))
self.data_test_scope = np.load(os.path.join(self.data_path, 'test_ins_scope.npy'))
print("Finish reading")
self.test_batches = len(self.data_test_label) / self.batch_size
if len(self.data_test_label) % self.batch_size != 0:
self.test_batches += 1
self.total_recall = self.data_test_label[:, 1:].sum()
def load_predict_data(self, d, embeddings):
self.data_word_vec = embeddings
self.data_test_word = d.get("word")
self.data_test_pos1 = d.get("pos1")
self.data_test_pos2 = d.get("pos2")
self.data_test_mask = d.get("mask")
if self.use_bag:
self.data_test_label = d.get("bag_label")
self.data_test_scope = d.get("bag_scope")
else:
self.data_test_label = d.get("ins_label")
self.data_test_scope = d.get("ins_scope")
self.test_batches = len(self.data_test_label) / self.batch_size
if len(self.data_test_label) % self.batch_size != 0:
self.test_batches += 1
def set_train_model(self, model):
print("Initializing training model...")
self.model = model
self.trainModel = self.model(config=self)
if self.pretrain_model != None:
self.trainModel.load_state_dict(torch.load(self.pretrain_model))
self.trainModel.cuda()
if self.optimizer != None:
pass
elif self.opt_method == "Adagrad" or self.opt_method == "adagrad":
self.optimizer = optim.Adagrad(self.trainModel.parameters(), lr = self.learning_rate, lr_decay = self.lr_decay, weight_decay = self.weight_decay)
elif self.opt_method == "Adadelta" or self.opt_method == "adadelta":
self.optimizer = optim.Adadelta(self.trainModel.parameters(), lr = self.learning_rate, weight_decay = self.weight_decay)
elif self.opt_method == "Adam" or self.opt_method == "adam":
self.optimizer = optim.Adam(self.trainModel.parameters(), lr = self.learning_rate, weight_decay = self.weight_decay)
else:
self.optimizer = optim.SGD(self.trainModel.parameters(), lr = self.learning_rate, weight_decay = self.weight_decay)
print("Finish initializing")
def set_test_model(self, model):
print("Initializing test model...")
self.model = model
self.testModel = self.model(config=self)
self.testModel.cuda()
self.testModel.eval()
print("Finish initializing")
def set_predict_model(self, model, epoch=None):
print("Initializing predict model...")
self.model = model
self.testModel = self.model(config=self)
self.testModel.cuda()
self.testModel.eval()
path = os.path.join(self.checkpoint_dir, self.model.__name__ + '-' + str(epoch))
if not os.path.isfile(path):
raise Exception("[ERROR] pre-trained model doesn't exist.")
self.testModel.load_state_dict(torch.load(path))
print("Finish initializing")
def get_train_batch(self, batch):
input_scope = np.take(self.data_train_scope,
self.train_order[batch * self.batch_size: (batch + 1) * self.batch_size], axis=0)
index = []
scope = [0]
for num in input_scope:
index = index + list(range(num[0], num[1] + 1))
scope.append(scope[len(scope) - 1] + num[1] - num[0] + 1)
self.batch_word = self.data_train_word[index, :]
self.batch_pos1 = self.data_train_pos1[index, :]
self.batch_pos2 = self.data_train_pos2[index, :]
self.batch_mask = self.data_train_mask[index, :]
self.batch_label = np.take(self.data_train_label,
self.train_order[batch * self.batch_size: (batch + 1) * self.batch_size], axis=0)
self.batch_attention_query = self.data_query_label[index]
self.batch_scope = scope
def get_test_batch(self, batch):
input_scope = self.data_test_scope[batch * self.batch_size: (batch + 1) * self.batch_size]
index = []
scope = [0]
for num in input_scope:
index = index + list(range(num[0], num[1] + 1))
scope.append(scope[len(scope) - 1] + num[1] - num[0] + 1)
self.batch_word = self.data_test_word[index, :]
self.batch_pos1 = self.data_test_pos1[index, :]
self.batch_pos2 = self.data_test_pos2[index, :]
self.batch_mask = self.data_test_mask[index, :]
self.batch_scope = scope
def train_one_step(self):
self.trainModel.embedding.word = to_var(self.batch_word)
self.trainModel.embedding.pos1 = to_var(self.batch_pos1)
self.trainModel.embedding.pos2 = to_var(self.batch_pos2)
self.trainModel.encoder.mask = to_var(self.batch_mask)
self.trainModel.selector.scope = self.batch_scope
self.trainModel.selector.attention_query = to_var(self.batch_attention_query)
self.trainModel.selector.label = to_var(self.batch_label)
self.trainModel.classifier.label = to_var(self.batch_label)
self.optimizer.zero_grad()
loss, _output = self.trainModel()
loss.backward()
self.optimizer.step()
for i, prediction in enumerate(_output):
if self.batch_label[i] == 0:
self.acc_NA.add(prediction == self.batch_label[i])
else:
self.acc_not_NA.add(prediction == self.batch_label[i])
self.acc_total.add(prediction == self.batch_label[i])
return loss.data.item()
def test_one_step(self):
self.testModel.embedding.word = to_var(self.batch_word)
self.testModel.embedding.pos1 = to_var(self.batch_pos1)
self.testModel.embedding.pos2 = to_var(self.batch_pos2)
self.testModel.encoder.mask = to_var(self.batch_mask)
self.testModel.selector.scope = self.batch_scope
return self.testModel.test()
def train(self):
if not os.path.exists(self.checkpoint_dir):
os.mkdir(self.checkpoint_dir)
best_auc = 0.0
best_p = None
best_r = None
best_epoch = 0
for epoch in range(self.max_epoch):
print('Epoch ' + str(epoch) + ' starts...')
self.acc_NA.clear()
self.acc_not_NA.clear()
self.acc_total.clear()
np.random.shuffle(self.train_order)
for batch in range(int(self.train_batches)):
self.get_train_batch(batch)
loss = self.train_one_step()
time_str = datetime.datetime.now().isoformat()
sys.stdout.write("epoch %d step %d time %s | loss: %f, NA accuracy: %f, not NA accuracy: %f, total accuracy: %f\r" % (epoch, batch, time_str, loss, self.acc_NA.get(), self.acc_not_NA.get(), self.acc_total.get()))
sys.stdout.flush()
if (epoch + 1) % self.save_epoch == 0:
print('Epoch ' + str(epoch) + ' has finished')
print('Saving model...')
path = os.path.join(self.checkpoint_dir, self.model.__name__ + '-' + str(epoch))
torch.save(self.trainModel.state_dict(), path)
print('Have saved model to ' + path)
if (epoch + 1) % self.valid_epoch == 0:
self.testModel = self.trainModel
auc, pr_x, pr_y = self.test_one_epoch()
if auc > best_auc:
best_auc = auc
best_p = pr_x
best_r = pr_y
best_epoch = epoch
print("Finish training")
print("Best epoch = %d | auc = %f" % (best_epoch, best_auc))
print("Storing best result...")
if not os.path.isdir(self.test_result_dir):
os.mkdir(self.test_result_dir)
np.save(os.path.join(self.test_result_dir, self.model.__name__ + '_x.npy'), best_p)
np.save(os.path.join(self.test_result_dir, self.model.__name__ + '_y.npy'), best_r)
print("Finish storing")
def test_one_epoch(self):
test_score = []
for batch in tqdm(range(int(self.test_batches))):
self.get_test_batch(batch)
batch_score = self.test_one_step()
test_score = test_score + batch_score
test_result = []
for i in range(len(test_score)):
for j in range(1, len(test_score[i])):
test_result.append([self.data_test_label[i][j], test_score[i][j]]) # ith sample jth class
test_result = sorted(test_result, key=lambda x: x[1])
test_result = test_result[::-1]
pr_x = []
pr_y = []
correct = 0
for i, item in enumerate(test_result):
correct += item[0]
pr_y.append(float(correct) / (i + 1))
pr_x.append(float(correct) / self.total_recall)
auc = sklearn.metrics.auc(x=pr_x, y=pr_y)
print("auc: ", auc)
return auc, pr_x, pr_y
def test(self):
best_epoch = None
best_auc = 0.0
best_p = None
best_r = None
for epoch in self.epoch_range:
path = os.path.join(self.checkpoint_dir, self.model.__name__ + '-' + str(epoch))
if not os.path.exists(path):
continue
print("Start testing epoch %d" % epoch)
self.testModel.load_state_dict(torch.load(path))
auc, p, r = self.test_one_epoch()
if auc > best_auc:
best_auc = auc
best_epoch = epoch
best_p = p
best_r = r
print("Finish testing epoch %d" % epoch)
print("Best epoch = %d | auc = %f" % (best_epoch, best_auc))
print("Storing best result...")
if not os.path.isdir(self.test_result_dir):
os.mkdir(self.test_result_dir)
np.save(os.path.join(self.test_result_dir, self.model.__name__ + '_x.npy'), best_p)
np.save(os.path.join(self.test_result_dir, self.model.__name__ + '_y.npy'), best_r)
print("Finish storing")
def predict(self, id2rel):
# input: id2rel = {class_name: index}
# output: lst_pre = [sample_1_pre, sample_2_pre, ...], sample_i_pre = (class_name, score)
test_score = list()
for batch in tqdm(range(int(self.test_batches))):
self.get_test_batch(batch)
batch_score = self.test_one_step() # batch_score = [sample_1, sample_2, ...], where sample = np.array(num_class,), i.e., sample is the predicted vector for one-hot
test_score.extend(batch_score) # test_score = [sample_1, sample_2, ...], where sample = np.array(num_class,), i.e., sample is the predicted vector for one-hot
# make prediction
lst_pre = list()
for sample in test_score:
i_pre = int(np.argmax(sample))
pre = (id2rel.get(i_pre), sample[i_pre])
lst_pre.append(pre)
return lst_pre