-
Notifications
You must be signed in to change notification settings - Fork 23
/
Copy pathlcm_exp_on_bert.py
161 lines (145 loc) · 6.3 KB
/
lcm_exp_on_bert.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
import datetime
import numpy as np
from utils import load_dataset, create_asy_noise_labels
from sklearn.utils import shuffle
from bert4keras.tokenizers import Tokenizer
from bert4keras.snippets import sequence_padding
from models import bert
import matplotlib.pyplot as plt
import time
#%%
# ========== parameters: ==========
maxlen = 128
hidden_size = 64
batch_size = 512
epochs = 150
# ========== bert config: ==========
# for English, use bert_tiny:
bert_type = 'bert'
config_path = 'bert_weights/bert_tiny_uncased_L-2_H-128_A-2/bert_config.json'
checkpoint_path = 'bert_weights/bert_tiny_uncased_L-2_H-128_A-2/bert_model.ckpt'
vocab_path = 'bert_weights/bert_tiny_uncased_L-2_H-128_A-2/vocab.txt'
# for Chinese, use albert_tiny:
# bert_type = 'albert'
# config_path = '../bert_weights/albert_tiny_google_zh_489k/albert_config.json'
# checkpoint_path = '../bert_weights/albert_tiny_google_zh_489k/albert_model.ckpt'
# vocab_path = '../bert_weights/albert_tiny_google_zh_489k/vocab.txt'
tokenizer = Tokenizer(vocab_path, do_lower_case=True)
# ========== dataset: ==========
dataset_name = '20NG'
group_noise_rate = 0.3
df,num_classes,label_groups = load_dataset(dataset_name)
# define log file name:
log_txt_name = '%s_BERT_log(group_noise=%s,comp+rec+talk)' % (dataset_name,group_noise_rate)
df = df.dropna(axis=0,how='any')
df = shuffle(df)[:50000]
print('data size:',len(df))
#%%
# ========== data preparation: ==========
labels = sorted(list(set(df.label)))
assert len(labels) == num_classes,'wrong num of classes!'
label2idx = {name:i for name,i in zip(labels,range(num_classes))}
#%%
print('start tokenizing...')
t = time.time()
X_token = []
X_seg = []
y = []
i = 0
for content,label in zip(list(df.content),list(df.label)):
i += 1
if i%1000 == 0:
print(i)
token_ids, seg_ids = tokenizer.encode(content, maxlen=maxlen)
X_token.append(token_ids)
X_seg.append(seg_ids)
y.append(label2idx[label])
# the sequences we obtained from above may have different length, so use Padding：
X_token = sequence_padding(X_token)
X_seg = sequence_padding(X_seg)
y = np.array(y)
print('tokenizing time cost:',time.time()-t,'s.')
#%%
# ========== model traing: ==========
old_list = []
ls_list = []
lcm_list = []
N = 5
for n in range(N):
# randomly split train and test each time:
np.random.seed(n) # 这样保证了每次试验的seed一致
random_indexs = np.random.permutation(range(len(X_token)))
train_size = int(len(X_token)*0.6)
val_size = int(len(X_token)*0.15)
X_token_train = X_token[random_indexs][:train_size]
X_token_val = X_token[random_indexs][train_size:train_size+val_size]
X_token_test = X_token[random_indexs][train_size+val_size:]
X_seg_train = X_seg[random_indexs][:train_size]
X_seg_val = X_seg[random_indexs][train_size:train_size + val_size]
X_seg_test = X_seg[random_indexs][train_size + val_size:]
y_train = y[random_indexs][:train_size]
y_val = y[random_indexs][train_size:train_size+val_size]
y_test = y[random_indexs][train_size+val_size:]
data_package = [X_token_train, X_seg_train, y_train, X_token_val, X_seg_val, y_val, X_token_test, X_seg_test, y_test]
# apply noise only on train set:
if group_noise_rate>0:
_, overall_noise_rate, y_train = create_asy_noise_labels(y_train,label_groups,label2idx,group_noise_rate)
data_package = [X_token_train, X_seg_train, y_train, X_token_val, X_seg_val, y_val, X_token_test, X_seg_test,
y_test]
with open('output/%s.txt' % log_txt_name, 'a') as f:
print('-'*30,'\nNOITCE: overall_noise_rate=%s'%round(overall_noise_rate,2), file=f)
with open('output/%s.txt'%log_txt_name,'a') as f:
print('\n',str(datetime.datetime.now()),file=f)
print('\n ROUND & SEED = ',n,'-'*20,file=f)
model_to_run = [""]
print('====Original:============')
model = bert.BERT_Basic(config_path,checkpoint_path,hidden_size,num_classes,bert_type)
train_score_list, val_socre_list, best_val_score, test_score = model.train_val(data_package,batch_size,epochs)
plt.plot(train_score_list, label='train')
plt.plot(val_socre_list, label='val')
plt.title('BERT')
plt.legend()
plt.show()
old_list.append(test_score)
with open('output/%s.txt'%log_txt_name,'a') as f:
print('\n*** Orig BERT ***:',file=f)
print('test acc:', str(test_score), file=f)
print('best val acc:',str(best_val_score),file=f)
print('train acc list:\n',str(train_score_list),file=f)
print('val acc list:\n',str(val_socre_list),'\n',file=f)
print('====Label Smooth:============')
ls_e = 0.1
model = bert.BERT_LS(config_path,checkpoint_path,hidden_size,num_classes,ls_e,bert_type)
train_score_list, val_socre_list, best_val_score, test_score = model.train_val(data_package,batch_size,epochs)
plt.plot(train_score_list, label='train')
plt.plot(val_socre_list, label='val')
plt.title('BERT with LS')
plt.legend()
plt.show()
old_list.append(test_score)
with open('output/%s.txt'%log_txt_name,'a') as f:
print('\n*** Orig BERT with LS (e=%s) ***:'%ls_e, file=f)
print('test acc:', str(test_score), file=f)
print('best val acc:', str(best_val_score), file=f)
print('train acc list:\n', str(train_score_list), file=f)
print('val acc list:\n', str(val_socre_list), '\n', file=f)
print('====LCM:============')
# alpha = 3
for alpha in [3,4,5]:
wvdim = 256
lcm_stop = 100
params_str = 'a=%s, wvdim=%s, lcm_stop=%s'%(alpha,wvdim,lcm_stop)
model = bert.BERT_LCM(config_path,checkpoint_path,hidden_size,num_classes,alpha,wvdim,bert_type)
train_score_list, val_socre_list, best_val_score, test_score = model.train_val(data_package, batch_size,epochs,lcm_stop)
plt.plot(train_score_list, label='train')
plt.plot(val_socre_list, label='val')
plt.title('BERT with LCM')
plt.legend()
plt.show()
old_list.append(test_score)
with open('output/%s.txt'%log_txt_name,'a') as f:
print('\n*** Orig BERT with LCM (%s) ***:'%params_str,file=f)
print('test acc:', str(test_score), file=f)
print('best val acc:', str(best_val_score), file=f)
print('train acc list:\n', str(train_score_list), file=f)
print('val acc list:\n', str(val_socre_list), '\n', file=f)