-
Notifications
You must be signed in to change notification settings - Fork 1
/
ensemble_feature.py
177 lines (150 loc) · 6.04 KB
/
ensemble_feature.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
172
173
174
175
176
177
# -*- coding: utf-8 -*-
import json
import argparse
import numpy as np
from collections import OrderedDict
import sys
from tqdm import tqdm
import os
import re
puc = set(',.，。！？~-=? ')
left_adj_words = ["大于","接近","超过","不超过","最高","最低","约为","约","大约"]
right_adj_words = ["左右","上下","以内","内","以上"]
def check_overlap_QA(question,answer_tuple,topN):
'''
这个函数为了过滤到一些和问题重叠度太高的答案。具体可以参考下面"fun1"中的例子
question: str
answer_tuple: ((answer_text, leijia_prob), (answer_text, leijia_prob) ...)
答案按照leijia概率从高到低排序排序
'''
top1_answer = answer_tuple[0][0]
for i,(answer, prob) in enumerate(answer_tuple[1:topN]):
diff_words = set(top1_answer) - set(answer)
if len(diff_words & set(question)) > 0.7 * len(diff_words) and len(diff_words & set(question)) > 2:
return answer
def Num_adj(answer_tuple,question,topN=3):
'''
避免一些数字的修饰词被忽略的情况。比如"大于5cm" vs "5cm"; "10min左右" vs "10min"
使用func2进行测试
answer_tuple: ((answer_text, leijia_prob), (answer_text, leijia_prob) ...)
答案按照leijia概率从高到低排序排序
'''
def is_number(text):
if re.findall(r"[1-9]",text):
return True
if re.findall(r"[几千百万亿十]",text):
return True
return False
top1_answer = answer_tuple[0][0]
if not is_number(top1_answer):
return None
for i,(answer, prob) in enumerate(answer_tuple[1:topN]):
for w in left_adj_words:
if w + top1_answer == answer and len(set(w) & set(question))<1:
return answer
for w in right_adj_words:
if top1_answer + w == answer and len(set(w) & set(question))<1:
return answer
def multi_sub_answer(answer_tuple,question,topN=3):
top1_answer = answer_tuple[0][0].lower()
question = question.lower()
best_answer = top1_answer
best_answer_topkens = re.split(r"[;、，,\s]",best_answer)
for i,(answer, prob) in enumerate(answer_tuple[1:topN]):
answer = answer.lower()
if best_answer not in answer:
continue
else:
answer_tokens = re.split(r"[;、，,]",answer)
flag = True
for t in best_answer_topkens:
if t not in answer_tokens:
flag = False
if not flag:
continue
if len(answer_tokens) <= len(best_answer_topkens):
continue
for t in answer_tokens:
if t >= '0' and t <= '9':
break
if "。" in t:
break
if len(t) > 5:
break
if len(t.strip()) == 0:
break
if len(t) - max([len(w) for w in best_answer_topkens]) >=2:
break
if t not in best_answer and len(set(t) & set(question)) > 1:
break
else:
if len(set(answer_tokens) - set(best_answer_topkens)) > 0:
best_answer = answer
return best_answer if best_answer!=top1_answer else None
def some_cases(char): # 哪些
return char in set('、或和')
def has_puc(char): # 短答案在长答案里面，然后其中有标点符号的不要
return char in set('。，！~')
question_set = set()
def keep_answer(answers, question, debug, which=False):
if len(answers) == 1:
return answers[0]
answer1 = answers[0]
answer2 = answers[1]
answer = answer1
if answer1 == ''.join(answer2.split()) and not any(map(lambda c:'\u4e00' <= c <= '\u9fa5',answer1)) and not any(map(lambda c:'\u4e00' <= c <= '\u9fa5',answer1)):
answer = answer2
if debug and question not in question_set:
print('答案是英文 并且只差空格，取有空格的\t', question)
print('\t', answer1, '###', answer2)
question_set.add(question)
elif answer1 in answer2 and '哪些' in question and any(map(lambda x: some_cases(x), answer2)):
answer = answer2
if len(answers) >= 3:
answer3 = answers[2]
if answer2 in answer3 and '哪些' in question and any(map(lambda x: some_cases(x), answer3)):
answer = answer3
if debug and question not in question_set:
print('question有哪些，要长答案\t', question)
print('\t', answers[:3], ' \tfinal: ', answer)
question_set.add(question)
elif which and answer1 in answer2 and '哪几' in question and any(map(lambda x: some_cases(x), answer2)):
answer = answer2
if len(answers) >= 3:
answer3 = answers[2]
if answer2 in answer3 and '哪几' in question and any(map(lambda x: some_cases(x), answer3)):
answer = answer3
if debug and question not in question_set:
print('question有 哪几 ，要长答案\t', question)
print('\t', answers[:3], ' \tfinal: ', answer)
question_set.add(question)
elif len(set(answer1) - set('，。,.！!?？、=-')) == 0:
answer = answer2
if debug and question not in question_set:
print('只有标点的不要\t', question)
print('\t', answer1, '###', answer2)
question_set.add(question)
else:
if question not in question_set and debug:
print('未处理\t', question)
print('\t', answer1, '###', answer2)
#
return answer
def my_num_contain_feature(text_list, question):
pred_answers_lst = sorted(text_list.items(), key=lambda d: d[1], reverse=True)
pred_answers_text_lst = [item[0] for item in pred_answers_lst]
ret2_answer = Num_adj(pred_answers_lst, question)
if ret2_answer is not None:
final_answer = ret2_answer
# print('不能忽略数字的修饰词\t', question)
# print('\ttop2: ', pred_answers_text_lst[:2], ' final: ', final_answer)
else:
ret3_answer = multi_sub_answer(pred_answers_lst, question)
if ret3_answer is not None:
final_answer = ret3_answer
# print('选长答案\t', question)
# print('\ttop3: ', pred_answers_text_lst[:3], ' final: ', final_answer)
else:
final_answer = keep_answer(pred_answers_text_lst, question, False)
assert isinstance(final_answer, str)
return final_answer