-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmain.py
191 lines (162 loc) · 7.46 KB
/
main.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
import torch
import numpy as np
from torch.optim import AdamW
from torch.nn import CrossEntropyLoss
from data_utils import get_data_list, data_preprocess, get_data_loader, calc_metrics
from model import MultimodalModel
import argparse
def init_argparse():
parser = argparse.ArgumentParser()
parser.add_argument('--text_only', action='store_true', help='仅使用文本')
parser.add_argument('--image_only', action='store_true', help='仅使用图片')
parser.add_argument('--do_test', action='store_true', help='使用训练后的模型对测试集进行预测')
parser.add_argument('--lr', default=5e-5, help='设置学习率', type=float)
parser.add_argument('--weight_decay', default=1e-2, help='设置权重衰减', type=float)
parser.add_argument('--epochs', default=10, help='设置训练轮数', type=int)
parser.add_argument('--seed', default=233, help='设置随机种子', type=int)
args = parser.parse_args()
return args
args = init_argparse()
print('args:', args)
"""text_only和image_only互斥"""
assert((args.text_only and args.image_only) == False)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
np.random.seed(args.seed)
"""在种子不变的情况下保证结果一致"""
torch.backends.cudnn.deterministic = True
def model_train():
"""训练模型并保存至./model.pth"""
train_data_list, test_data_list = get_data_list()
train_data_list, test_data_list = data_preprocess(train_data_list, test_data_list)
train_data_loader, valid_data_loader, test_data_loader = get_data_loader(train_data_list, test_data_list)
model = MultimodalModel.from_pretrained('bert-base-uncased')
model.to(device)
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [
{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
]
optimizer = AdamW(lr=args.lr, params=optimizer_grouped_parameters)
criterion = CrossEntropyLoss()
best_rate = 0
print('[START_OF_TRAINING_STAGE]')
for epoch in range(args.epochs):
total_loss = 0
correct = 0
total = 0
target_list = []
pred_list = []
model.train()
for idx, (guid, tag, image, text) in enumerate(train_data_loader):
tag = tag.to(device)
image = image.to(device)
text = text.to(device)
if args.text_only:
out = model(image_input=None, text_input=text)
elif args.image_only:
out = model(image_input=image, text_input=None)
else:
out = model(image_input=image, text_input=text)
loss = criterion(out, tag)
loss.backward()
optimizer.step()
optimizer.zero_grad()
total_loss += loss.item() * len(guid)
pred = torch.max(out, 1)[1]
total += len(guid)
correct += (pred == tag).sum()
target_list.extend(tag.cpu().tolist())
pred_list.extend(pred.cpu().tolist())
total_loss /= total
print('[EPOCH{:02d}]'.format(epoch + 1), end='')
print('[TRAIN] - LOSS:{:.6f}'.format(total_loss), end='')
rate = correct / total * 100
print(' ACC_RATE:{:.2f}%'.format(rate), end='')
metrics = calc_metrics(target_list, pred_list)
print(' WEIGHTED_ACC: {:.2f}% WEIGHTED_F1: {:.2f}% MAC_ACC: {:.2f}% MAC_F1: {:.2f}%'.format(metrics[0] * 100,
metrics[2] * 100,
metrics[3] * 100,
metrics[5] * 100))
total_loss = 0
correct = 0
total = 0
target_list = []
pred_list = []
model.eval()
for guid, tag, image, text in valid_data_loader:
tag = tag.to(device)
image = image.to(device)
text = text.to(device)
if args.text_only:
out = model(image_input=None, text_input=text)
elif args.image_only:
out = model(image_input=image, text_input=None)
else:
out = model(image_input=image, text_input=text)
loss = criterion(out, tag)
total_loss += loss.item() * len(guid)
pred = torch.max(out, 1)[1]
total += len(guid)
correct += (pred == tag).sum()
target_list.extend(tag.cpu().tolist())
pred_list.extend(pred.cpu().tolist())
total_loss /= total
print(' [EVAL] - LOSS:{:.6f}'.format(total_loss), end='')
rate = correct / total * 100
print(' ACC_RATE:{:.2f}%'.format(rate), end='')
metrics = calc_metrics(target_list, pred_list)
print(' WEIGHTED_ACC: {:.2f}% WEIGHTED_F1: {:.2f}% MAC_ACC: {:.2f}% MAC_F1: {:.2f}%'.format(metrics[0] * 100,
metrics[2] * 100,
metrics[3] * 100,
metrics[5] * 100))
if rate > best_rate:
best_rate = rate
print(' [SAVE] BEST ACC_RATE ON THE VALIDATION SET:{:.2f}%'.format(rate))
torch.save(model.state_dict(), 'model.pth')
print()
print('[END_OF_TRAINING_STAGE]')
def model_test():
"""利用训练好的./model.pth对测试集进行预测，结果保存至output/test_with_label.txt"""
train_data_list, test_data_list = get_data_list()
train_data_list, test_data_list = data_preprocess(train_data_list, test_data_list)
train_data_loader, valid_data_loader, test_data_loader = get_data_loader(train_data_list, test_data_list)
model = MultimodalModel.from_pretrained('bert-base-uncased')
model.load_state_dict(torch.load('model.pth'))
model.to(device)
print('[START_OF_TESTING_STAGE]')
guid_list = []
pred_list = []
model.eval()
for guid, tag, image, text in test_data_loader:
image = image.to(device)
text = text.to(device)
if args.text_only:
out = model(image_input=None, text_input=text)
elif args.image_only:
out = model(image_input=image, text_input=None)
else:
out = model(image_input=image, text_input=text)
pred = torch.max(out, 1)[1]
guid_list.extend(guid)
pred_list.extend(pred.cpu().tolist())
pred_mapped = {
0: 'negative',
1: 'neutral',
2: 'positive',
}
with open('output/test_with_label.txt', 'w', encoding='utf-8') as f:
f.write('guid,tag\n')
for guid, pred in zip(guid_list, pred_list):
f.write(f'{guid},{pred_mapped[pred]}\n')
f.close()
print('[PREDICTION] SAVE TO output/test_with_label.txt')
print('[END_OF_TESTING_STAGE]')
if __name__ == "__main__":
if args.do_test:
model_test()
else:
model_train()