-
Notifications
You must be signed in to change notification settings - Fork 41
/
nsp_bert_cloze_style.py
171 lines (149 loc) · 7.83 KB
/
nsp_bert_cloze_style.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
#! /usr/bin/env python
# -*- coding: utf-8 -*-
# __author__ = "Sponge_sy"
# Date: 2021/6/30
import numpy
from tqdm import tqdm
from sklearn import metrics
from bert4keras.tokenizers import Tokenizer
from bert4keras.models import build_transformer_model
from bert4keras.snippets import sequence_padding, DataGenerator
from utils import *
class data_generator(DataGenerator):
"""Data Generator"""
def __init__(self, is_pre=True, is_soft_pos=False, *args, **kwargs):
super(data_generator, self).__init__(*args, **kwargs)
self.is_pre = is_pre
self.is_soft_pos = is_soft_pos
def __iter__(self, random=False):
batch_token_ids, batch_segment_ids, batch_output_ids = [], [], []
batch_position_ids = []
for is_end, (text, candi) in self.sample(random):
text_1, text_2 = text.split("#idiom#")
if (self.is_soft_pos):
text = text.replace("#idiom#", "")
if (self.is_pre):
token_ids, segment_ids = tokenizer.encode(first_text=candi, second_text=text, maxlen=maxlen)
else:
token_ids, segment_ids = tokenizer.encode(first_text=text, second_text=candi, maxlen=maxlen)
position_ids = []
if (self.is_soft_pos):
tokens_1 = tokenizer._tokenize(text_1)
tokens_2 = tokenizer._tokenize(text_2)
tokens_c = tokenizer._tokenize(candi)
position_ids = [0] # "[CSL]
if (self.is_pre):
position_ids += [2 + len(tokens_1) + p_id for p_id in
range(len(tokens_c))] # "[CLS]" + TEXT1 + TEXT2 + "[SEP]" + CANDI
position_ids += [1] # "[CLS]" + TEXT1 + TEXT2 + "[SEP]"
position_ids += [position_ids[-1] + p_id + 1 for p_id in range(len(tokens_1))] # "[CLS]" + TEXT1
position_ids += [position_ids[-1] + len(tokens_c) + p_id + 1 for p_id in
range(len(tokens_2))] # "[CLS]" + TEXT1 +TEXT2
position_ids += [position_ids[-1] + 1]
else:
position_ids += [position_ids[-1] + p_id + 1 for p_id in range(len(tokens_1))] # "[CLS]" + TEXT1
position_ids += [position_ids[-1] + len(tokens_c) + p_id + 1 for p_id in
range(len(tokens_2))] # "[CLS]" + TEXT1 +TEXT2
position_ids += [position_ids[-1] + 1] # "[CLS]" + TEXT1 + TEXT2 + "[SEP]"
position_ids += [1 + len(tokens_1) + p_id for p_id in
range(len(tokens_c))] # "[CLS]" + TEXT1 + TEXT2 + "[SEP]" + CANDI
position_ids += [len(tokens_1 + tokens_2 + tokens_c) + 2]
# position_ids += [position_ids[-1] + 1]
source_ids, target_ids = token_ids[:], token_ids[:]
# label_ids = tokenizer.encode(label)[0][1:-1]
batch_token_ids.append(source_ids)
batch_segment_ids.append(segment_ids)
batch_position_ids.append(position_ids)
if len(batch_token_ids) == self.batch_size or is_end:
batch_token_ids = sequence_padding(batch_token_ids)
batch_segment_ids = sequence_padding(batch_segment_ids)
batch_position_ids = sequence_padding(batch_position_ids)
if (self.is_soft_pos):
yield [batch_token_ids, batch_segment_ids, batch_position_ids], None
else:
yield [batch_token_ids, batch_segment_ids], None
batch_token_ids, batch_segment_ids, batch_position_ids = [], [], []
def evaluate(data_generator_list, data, note=""):
print("\n*******************Start to Zero-Shot predict on 【{}】*******************".format(note), flush=True)
candidates_logits = [[] for _ in range(len(data[0][-1]))]
for i in range(len(data_generator_list)):
print("\nPattern{}".format(i), flush=True)
data_generator = data_generator_list[i]
counter = 0
for (x, _) in tqdm(data_generator):
outputs = model.predict(x)
for out in outputs:
logit_pos = out[0].T
candidates_logits[i].append(logit_pos)
counter += 1
# Evaluate the results
trues = [d[1] for d in data]
preds = []
for i in range(len(candidates_logits[0])):
pred = numpy.argmax([logits[i] for logits in candidates_logits])
preds.append(int(pred))
confusion_matrix = metrics.confusion_matrix(trues, preds, labels=None, sample_weight=None)
acc = metrics.accuracy_score(trues, preds, normalize=True, sample_weight=None)
print("Confusion Matrix:\n{}".format(confusion_matrix), flush=True)
print("Acc.:\t{:.4f}".format(acc), flush=True)
return acc
def depart_choze(data):
text_list = []
candidates_list = [[] for _ in range(len(data[0][-1]))]
for d in data:
text = d[0]
text_list.append(text)
candidates = d[-1]
for i, c in enumerate(candidates):
candidates_list[i].append(c)
return text_list, candidates_list
if __name__ == "__main__":
# Load the hyper-parameters-----------------------------------------------------------
maxlen = 128 # The max length 128 is used in our paper
batch_size = 40 # Will not influence the results
# Choose one of the models----------------------------------------------------------------------
# Recommend to use 'uer-mixed-bert-base'
# model_names = ['google-bert', 'google-bert-small', 'google-bert-zh',
# 'hfl-bert-wwm', 'hfl-bert-wwm-ext',
# 'uer-mixed-bert-tiny', 'uer-mixed-bert-small',
# 'uer-mixed-bert-base', 'uer-mixed-bert-large']
model_name = 'uer-mixed-bert-base'
# Choose a dataset----------------------------------------------------------------------
# dataset_names = ['chid']
dataset_name = 'chid'
# Load model and dataset class
bert_model = Model(model_name=model_name)
dataset = Datasets(dataset_name=dataset_name)
# Prefix or Suffix-------------------------------------------------------------------
is_pre = True
# if using soft position or not
is_soft_pos = True
# Load the dev set--------------------------------------------------------------------
# -1 for all the samples
dev_data = dataset.load_data(dataset.dev_path, sample_num=-1, is_shuffle=True)
text_list, candidates_list = depart_choze(dev_data)
dev_generator_list = []
for candidates in candidates_list:
dev_generator_list.append(data_generator(is_pre=is_pre, is_soft_pos=is_soft_pos, data=zip(text_list, candidates),
batch_size=batch_size))
# Load the test set--------------------------------------------------------------------
# -1 for all the samples-
test_data = dataset.load_data(dataset.test_path, sample_num=-1, is_shuffle=True)
text_list, candidates_list = depart_choze(test_data)
test_generator_list = []
for candidates in candidates_list:
test_generator_list.append(
data_generator(is_pre=is_pre, is_soft_pos=is_soft_pos, data=zip(text_list, candidates),
batch_size=batch_size))
# Build BERT model---------------------------------------------------------------------
tokenizer = Tokenizer(bert_model.dict_path, do_lower_case=True)
# Load BERET model with NSP head
model = build_transformer_model(
config_path=bert_model.config_path,
checkpoint_path=bert_model.checkpoint_path,
custom_position_ids=is_soft_pos,
with_nsp=True,
)
# Zero-Shot predict and evaluate-------------------------------------------------------
evaluate(dev_generator_list, dev_data, note="Dev Set")
evaluate(test_generator_list, test_data, note="Test Set")