-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclassification_config.py
143 lines (127 loc) · 6.87 KB
/
classification_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
import os
import argparse
root_path = "/mnt/inspurfs/user-fs/wxy/glyphCRM/" # your project pwd
data_path = "/mnt/inspurfs/user-fs/wxy/glyphCRM/"
def get_path(p, is_data=False):
if is_data:
return os.path.join(data_path, p)
else:
return os.path.join(root_path, p)
config = {
#"device": "0",
"seed": 2048, #random seed
"device": "3", #gpu device
"num_labels": 2, #classification labels
"epoch": 10,
"batch_size": 32,
"batch_expand_times": 1,
"warm_up": 0.2,
"weight_decay": 0,
"steps_eval": 0.2,
"start_eval_epoch": 3,
"exp_times": 3,
"lr":3e-5,
"bmp_path": get_path("data/bmp48/", is_data=True), #The preprocess bmp image path, preprocessed with a given vocab
"vocab_path": get_path("data/vocab.txt", is_data=True),#The given vocab.
#"vocab_path": get_path("data/vocab_bmp.txt", is_data=True),
"bert_config_path": get_path("pretrained_model/config.json"),
"preprocessing": False,
"use_res2bert": True,
"cnn_and_embed_mat": True,
"parallel": None,
"vocab_size": 18612,
"json_data": True,
#"save_root": "./downstream/save",
"state_dict": None, # state_dict is not pretrained_model_path
}
parser = argparse.ArgumentParser()
parser.add_argument('--device', default=config.get('device'), type=str, required=False)
parser.add_argument('--epoch', default=config.get('epoch'), type=int, required=False)
parser.add_argument('--batch_size', default=config.get('batch_size'), type=int, required=False)
parser.add_argument('--batch_expand_times', default=config.get('batch_expand_times'), type=int, required=False)
parser.add_argument('--warm_up', default=config.get('warm_up'), type=float, required=False)
parser.add_argument('--weight_decay', default=config.get('weight_decay'), type=float, required=False)
parser.add_argument('--lr', default=config.get('lr'), type=float, required=False)
parser.add_argument('--steps_eval', default=config.get('steps_eval'), type=float, required=False)
parser.add_argument('--start_eval_epoch', default=config.get('start_eval_epoch'), type=int, required=False)
parser.add_argument('--dataset_name', type=str, required=True)
parser.add_argument('--exp_times', default=config.get('exp_times'), type=int, required=False)
parser.add_argument('--pretrained_model_path', type=str, required=True)
parser.add_argument('--use_res2bert', default=config.get('use_res2bert'), type=bool, required=False)
parser.add_argument('--cnn_and_embed_mat', default=config.get('cnn_and_embed_mat'), type=bool, required=False)
parser.add_argument('--state_dict', default=config.get('state_dict'), type=str, required=False)
parser.add_argument('--save_root', default=config.get('save_root'), type=str, required=False)
args = vars(parser.parse_args())
for k in args.keys():
if k not in config.keys():
print("add new config key: {}={}".format(k, args[k]))
config[k] = args[k]
config['save_root']="./downstream/"+config['dataset_name']
os.environ["CUDA_VISIBLE_DEVICES"] = config['device']
if config['dataset_name'] == "chnsenti" and config['json_data']==True:
config.update({
"train_data_path": get_path("data/downstream_data/senti_train.json", is_data=True),
"dev_data_path": get_path("data/downstream_data/senti_dev.json", is_data=True),
"test_data_path": get_path("data/downstream_data/senti_test.json", is_data=True),
})
# elif config['dataset_name'] == "hotel" and config['json_data']==True:
# config.update({
# "train_data_path": get_path("./data/classification_zy_data/hotel_train.json", is_data=True),
# "dev_data_path": get_path("./data/classification_zy_data/hotel_dev.json", is_data=True),
# "test_data_path": get_path("./data/classification_zy_data/hotel_test.json", is_data=True),
# })
# elif config['dataset_name'] == "onlinesenti_cls" and config['json_data']==True:
# config.update({
# "train_data_path": get_path("./data/classification_zy_data/onlinesenti_train.json", is_data=True),
# "dev_data_path": get_path("./data/classification_zy_data/onlinesenti_dev.json", is_data=True),
# "test_data_path": get_path("./data/classification_zy_data/onlinesenti_test.json", is_data=True),
# })
# elif config['dataset_name'] == "cnews" and config['json_data']==True:
# config.update({
# "train_data_path": get_path("./data/classification_zy_data/cnews_train.json", is_data=True),
# "dev_data_path": get_path("./data/classification_zy_data/cnews_dev.json", is_data=True),
# "test_data_path": get_path("./data/classification_zy_data/cnews_test.json", is_data=True),
# })
# elif config['dataset_name'] == "toutiao" and config['json_data']==True:
# config.update({
# "train_data_path": get_path("./data/classification_zy_data/toutiao_train.json", is_data=True),
# "dev_data_path": get_path("./data/classification_zy_data/toutiao_dev.json", is_data=True),
# "test_data_path": get_path("./data/classification_zy_data/toutiao_test.json", is_data=True),
# })
# elif config['dataset_name'] == "medical" and config['json_data']==True:
# config.update({
# "train_data_path": get_path("./data/classification_zy_data/medical_train.json", is_data=True),
# "dev_data_path": get_path("./data/classification_zy_data/medical_dev.json", is_data=True),
# "test_data_path": get_path("./data/classification_zy_data/medical_test.json", is_data=True),
# })
# elif config['dataset_name'] == "afqmc":
# config.update({
# "train_data_path": get_path("Baseline/afqmc_public/afqmc_train_preprocess.pkl", is_data=True),
# "dev_data_path": get_path("Baseline/afqmc_public/afqmc_dev_preprocess.pkl", is_data=True),
# "test_data_path": get_path("Baseline/afqmc_public/afqmc_dev_preprocess.pkl", is_data=True),
# })
#
# elif config['dataset_name'] == "iflytek":
# config.update({
# "train_data_path": get_path("Baseline/iflytek_public/iflytek_train_preprocess.pkl", is_data=True),
# "dev_data_path": get_path("Baseline/iflytek_public/iflytek_dev_preprocess.pkl", is_data=True),
# "test_data_path": get_path("Baseline/iflytek_public/iflytek_dev_preprocess.pkl", is_data=True),
# })
#
# elif config['dataset_name'] == "tnews":
# config.update({
# "train_data_path": get_path("Baseline/tnews_public/tnews_train_preprocess.pkl", is_data=True),
# "dev_data_path": get_path("Baseline/tnews_public/tnews_dev_preprocess.pkl", is_data=True),
# "test_data_path": get_path("Baseline/tnews_public/tnews_dev_preprocess.pkl", is_data=True),
# })
print(config)
print("")
# ./save/AddBertResPos3-epoch2-loss1.03873.pt
""" 参数记录
onlinesenti:
--lr=3e-5 --epoch=10 --steps_eval=0.1 --start_eval_epoch=2 --batch_size=8 --warm_up=0.2
hotel:
--lr=2e-5 --epoch=10 --steps_eval=0.2 --start_eval_epoch=3 --batch_size=32 --warm_up=0.1
chnsenti:
--lr=2e-5 --epoch=10 --steps_eval=0.2 --start_eval_epoch=3 --batch_size=32 --warm_up=0.1
"""