-
Notifications
You must be signed in to change notification settings - Fork 19
/
run_multi_label.py
144 lines (131 loc) · 5.91 KB
/
run_multi_label.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
# coding: UTF-8
import os
import time
import torch
import torch.nn.functional as F
import numpy as np
from tools.losses import cross_entropy_multi_label, multi_label_circle_loss
from tools.train_eval import train, init_network
from importlib import import_module
import argparse
# 基本参数
os.environ["CUDA_VISIBLE_DEVICES"] = "0" # gpu
seed = 1
# 1.命令行传参：出于简洁性考虑，刻意控制了命令行参数的数量，可根据需要修改
parser = argparse.ArgumentParser(description='Text Classification')
parser.add_argument('--lng', type=str, default="cn",
help='choose language: en for English, cn for Chinese, multi for multi-language')
parser.add_argument('--architecture', type=str, default="DNN",
help='choose model architecture: shallow for traditional method, DNN for deep neural networks, transformers for bert-type pretrained models')
parser.add_argument('--model', type=str, default="TextCNN",
help='choose a model: TextCNN, TextRNN, FastText, TextRCNN, TextRNN_Att, DPCNN, Transformer, Bert, ERNIE')
parser.add_argument('--embedding', default='random', type=str, help='random, inner, outer')
parser.add_argument('--word', default=False, type=bool, help='True for word, False for char')
parser.add_argument('--pretrained_name_or_path', type=str, default="cn",
help='pretrained model/word vector name or path')
args = parser.parse_args()
# 2.直接在代码中指定参数：提供了DNN和transformers的示例
# 数据
data_path = "./data/"
dataset = 'THUCNews' # 数据集
sep = "\t"
# dataset = 'dbpedia' # 数据集
# sep = ","
# dataset = 'nlp-with-disaster-tweets' # 数据集
# sep = "\t"
data_path = "./data/"
dataset = 'rcv1' # 数据集
sep = "\t"
dataset = data_path + dataset
# transformers类模型
n_samples = None # None for full data, n for debug or big file
args.lng = "en" # en, cn, multi
args.architecture = "transformers"
args.model = "bert" # 名字需要与相应的module即py文件对齐
# args.model = "ERNIE" # 名字需要与相应的module即py文件对齐
args.embedding = "inner" # embedding下载需要时间
# args.pretrained_name_or_path = "bert-base-chinese"
# args.pretrained_name_or_path = "bert-base-uncased"
args.pretrained_name_or_path = "bert-base-cased"
# args.pretrained_name_or_path = "nghuyong/ernie-1.0-base-zh"
# args.pretrained_name_or_path = "swtx/ernie-2.0-base-chinese"
# args.pretrained_name_or_path = "nghuyong/ernie-3.0-base-zh"
# args.pretrained_name_or_path = "nghuyong/ernie-2.0-base-en"
# DNN类模型
# n_samples = None
# args.lng = "en" # en, cn, multi
# args.architecture = "DNN"
# args.model = "TextCNN" # 名字需要与相应的module即py文件对齐
# args.embedding = "outer" # embedding下载需要时间
# args.word = True
# en word vectors
# args.pretrained_name_or_path = "glove.840B.300d.txt"
# args.pretrained_name_or_path = "wiki-news-300d-1M.vec" # fasttext
# cn word vectors
# args.pretrained_name_or_path = "sgns.sogounews.bigram-char"
# args.pretrained_name_or_path = "cc.zh.300.vec" # fasttext
# 超参数：大部分参数在config中设置
lr = 2e-5
batch_size = 128 # 默认128
multi_label = True
loss_fn = cross_entropy_multi_label # multi-hot ce
loss_fn = F.binary_cross_entropy_with_logits # bce
loss_fn = multi_label_circle_loss # 苏剑林版多标签分类loss
eval_activate = None
cls_threshold = 0
epochs = 20
if __name__ == '__main__':
embedding = 'embedding_SougouNews.npz' if args.lng == "cn" else "glove.840B.300d.txt"
if args.embedding == 'random':
embedding = 'random'
model_name = args.model
if model_name == 'FastText':
from tools.utils_fasttext import build_dataset, build_iterator, get_time_dif
embedding = 'random'
elif args.architecture == "transformers":
from tools.utils_transformers import build_dataset, build_iterator, get_time_dif
else:
from tools.utils import build_dataset, build_iterator, get_time_dif
x = import_module('models.' + model_name)
if args.embedding == "inner":
# 预训练模型使用自带的embeddings
config = x.Config(dataset, args.pretrained_name_or_path)
else:
config = x.Config(dataset, args.pretrained_name_or_path)
# single/multi label
config.multi_label = multi_label
# 超参数设置
config.learning_rate = lr
config.batch_size = batch_size
config.num_epochs = epochs
# 保存路径+时间
config.save_path = config.save_path + "_" + time.strftime('%m-%d_%H.%M', time.localtime())
np.random.seed(1)
torch.manual_seed(1)
torch.cuda.manual_seed_all(1)
torch.backends.cudnn.deterministic = True # 保证每次结果一样
start_time = time.time()
print("Loading data...")
if args.architecture == "transformers":
train_data, dev_data, test_data = build_dataset(config, n_samples=n_samples, sep=sep, multi_label=multi_label)
else:
vocab, train_data, dev_data, test_data = build_dataset(config, args.word, n_samples=n_samples, lng=args.lng,
sep=sep, multi_label=multi_label)
config.vocab = vocab
config.n_vocab = len(vocab)
train_iter = build_iterator(train_data, config)
dev_iter = build_iterator(dev_data, config)
test_iter = build_iterator(test_data, config)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# train
model = x.Model(config)
if args.architecture == 'DNN':
init_network(model)
print(model.parameters)
print("Training...")
start_time = time.time()
train(config, model, train_iter, dev_iter, test_iter, loss_fn=loss_fn, multi_label=multi_label,
cls_threshold=cls_threshold, eval_activate=eval_activate)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)