-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathbase_roberta_single.py
170 lines (135 loc) · 7.27 KB
/
base_roberta_single.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
# -*- coding:utf-8 -*-
import json
import pandas as pd
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torch.cuda.amp import autocast, GradScaler
from tqdm import tqdm
import random
import os
import time
from sklearn.model_selection import StratifiedKFold, train_test_split
from transformers import BertTokenizer, RobertaTokenizer, RobertaModel, BertModel, BertConfig
from transformers import BertForMultipleChoice, RobertaForMultipleChoice
from transformers import get_cosine_schedule_with_warmup
from transformers import AdamW
import BertModelsCustom
import utils
Param = { # 训练的参数配置
'fold_num': 5, # 五折交叉验证
'seed': 42,
'model': '/data2/roberta/RoBERTa_zh_L12',
'path_data': '/home/zuoyuhui/DataGame/haihuai_RC/data/',
'max_len': 256, # 文本截断的最大长度
'epochs': 8,
'train_bs': 16, # batch_size，可根据自己的显存调整
'valid_bs': 16,
'lr': 1e-5, # 学习率
'num_workers': 2,
'accum_iter': 2, # 梯度累积，相当于将batch_size*2
'weight_decay': 1e-5, # 权重衰减，防止过拟合
'device': 0,
}
tokenizer = BertTokenizer.from_pretrained(Param['model'])
def collate_fn(data): # 将文章问题选项拼在一起后，得到分词后的数字id，输出的size是(batch,n_choices,max_len
input_ids, attention_mask, token_type_ids = [], [], []
for x in data:
text = tokenizer(x[1],
text_pair=x[0],
padding='max_length', # 填充到使用参数max_length指定的最大长度，或者填充到模型的最大可接受输入长度(如果未提供该参数)。
truncation=True,
# TRUE或‘LIMEST_FIRST’：截断到使用参数max_length指定的最大长度，或者截断到模型的最大可接受输入长度(如果没有提供该参数)。这将逐个令牌截断令牌，如果提供了一对序列(或一批对)，则从该对中最长的序列中删除一个令牌。
max_length=Param['max_len'],
return_tensors='pt') # 返回pytorch tensor格式
input_ids.append(text['input_ids'].tolist())
attention_mask.append(text['attention_mask'].tolist())
token_type_ids.append(text['token_type_ids'].tolist())
input_ids = torch.tensor(input_ids)
attention_mask = torch.tensor(attention_mask)
token_type_ids = torch.tensor(token_type_ids)
label = torch.tensor([x[-1] for x in data])
return input_ids, attention_mask, token_type_ids, label
def main():
utils.seed_everything(Param['seed'])
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
train_df = pd.read_csv(Param['path_data'] + 'train.csv')
test_df = pd.read_csv(Param['path_data'] + 'test.csv')
train_df['label'] = train_df['Answer'].apply(lambda x: ['A', 'B', 'C', 'D'].index(x))
test_df['label'] = 0
model = BertModelsCustom.RobertaForMultipleChoice.from_pretrained(Param['model']).to(device)
folds = StratifiedKFold(n_splits=Param['fold_num'], shuffle=True, random_state=Param['seed']) \
.split(np.arange(train_df.shape[0]), train_df.label.values) # 五折交叉验证
cv = [] # 保存每折的最佳准确率
for fold, (trn_idx, val_idx) in enumerate(folds):
train = train_df.loc[trn_idx]
val = train_df.loc[val_idx]
train_set = utils.MyDataset(train)
val_set = utils.MyDataset(val)
train_loader = DataLoader(train_set, batch_size=Param['train_bs'], collate_fn=collate_fn, shuffle=True,
num_workers=Param['num_workers'])
val_loader = DataLoader(val_set, batch_size=Param['valid_bs'], collate_fn=collate_fn, shuffle=False,
num_workers=Param['num_workers'])
best_acc = 0
scaler = GradScaler()
optimizer = AdamW(model.parameters(), lr=Param['lr'], weight_decay=Param['weight_decay']) # AdamW优化器
criterion = nn.CrossEntropyLoss()
scheduler = get_cosine_schedule_with_warmup(optimizer, len(train_loader) // Param['accum_iter'],
Param['epochs'] * len(train_loader) // Param['accum_iter'])
for epoch in range(Param['epochs']):
print('epoch:', epoch)
time.sleep(0.2)
# train_loss, train_acc = train_model(model, train_loader)
losses = utils.AverageMeter1()
accs = utils.AverageMeter1()
optimizer.zero_grad()
tk = tqdm(train_loader, total=len(train_loader), position=0, leave=True)
for step, (input_ids, attention_mask, token_type_ids, y) in enumerate(tk):
input_ids, attention_mask, token_type_ids, y = input_ids.to(device), attention_mask.to(
device), token_type_ids.to(device), y.to(device).long()
# https://zhuanlan.zhihu.com/p/165152789
with autocast(): # 使用半精度训练
output = model(input_ids, attention_mask, token_type_ids)
output = output.logits
loss = criterion(output, y) / Param['accum_iter']
scaler.scale(loss).backward()
if ((step + 1) % Param['accum_iter'] == 0) or ((step + 1) == len(train_loader)): # 梯度累加
scaler.step(optimizer)
scaler.update()
optimizer.zero_grad()
scheduler.step()
acc = (output.argmax(1) == y).sum().item() / y.size(0)
losses.update(loss.item() * Param['accum_iter'], y.size(0))
accs.update(acc, y.size(0))
tk.set_postfix(loss=losses.avg, acc=accs.avg)
# val_loss, val_acc = test_model(model, val_loader)
model.eval()
losses = utils.AverageMeter1()
accs = utils.AverageMeter1()
y_truth, y_pred = [], []
with torch.no_grad():
tk = tqdm(val_loader, total=len(val_loader), position=0, leave=True)
for idx, (input_ids, attention_mask, token_type_ids, y) in enumerate(tk):
input_ids, attention_mask, token_type_ids, y = input_ids.to(device), attention_mask.to(
device), token_type_ids.to(device), y.to(device).long()
output = model(input_ids, attention_mask, token_type_ids).logits
y_truth.extend(y.cpu().numpy())
y_pred.extend(output.argmax(1).cpu().numpy())
loss = criterion(output, y)
acc = (output.argmax(1) == y).sum().item() / y.size(0)
losses.update(loss.item(), y.size(0))
accs.update(acc, y.size(0))
tk.set_postfix(loss=losses.avg, acc=accs.avg)
val_acc = accs.avg
val_loss = losses.avg
if val_acc > best_acc:
best_acc = val_acc
print(val_acc)
torch.save(model.state_dict(),
'{}_fold_{}_robert_single.pt'.format(Param['model'].split('/')[-1], fold))
cv.append(best_acc)
print("cv:", np.mean(cv))
if __name__ == '__main__':
main()