-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathclean_dataset.py
232 lines (219 loc) · 10.5 KB
/
clean_dataset.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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
import json
from tot.tasks import get_task
import random
from load_data import *
import re
# task = get_task('fever')
# path = 'feverous_13b_2.json'
# file_name = 'feverous_13b_2_data.json'
# final_sentence = 'step 3, so the final answer is: '
# thought_number = 3
task = get_task('bamboogle')
path = 'bamboogle_7b.json'
file_name = 'bamboogle_7b_data.json'
final_sentence = 'step 3, so the final answer is: '
thought_number = 3
final_thought = str(thought_number-1)
with open(path, 'r') as f:
instances = json.load(f)
Corpus = {}
for instance in instances:
sample = list(instance.keys())[0]
try:
correct_predict = instance[sample]['correct'][0].split(final_sentence)[1]
except:
# correct_predict = instance[sample]['correct'][1].split('final_sentence')[1]
# print(instance[sample]['correct'][0])
continue
if ('fever' in path) or (('vitaminc' in path)):
if ('suport' in correct_predict) or ('support' in correct_predict):
correct_predict = 'supports'
if ('refu' in correct_predict) or ('reject' in correct_predict):
correct_predict = 'refutes'
if ('not enough' in correct_predict) or ('no enough' in correct_predict):
correct_predict = 'not enough information'
if correct_predict.replace('.','').strip() not in ['supports', 'refutes', 'refuted', 'not enough info', 'not enough information']:
# print(correct_predict)
continue
for j in range(len(instance[sample][final_thought]['candiate'])):
try:
if (final_sentence) not in instance[sample][final_thought]['candiate'][j]:
continue
pre = instance[sample][final_thought]['candiate'][j].split(final_sentence)[1]
except:
print(instance[sample][final_thought]['candiate'][j])
print(pre)
exit()
if 'fever' in path:
if correct_predict.replace('.','').strip() in ['supports']:
if 'support' in pre:
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
elif correct_predict.replace('.','').strip() in ['refutes', 'refuted']:
if 'refute' in pre:
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
elif correct_predict.replace('.','').strip() in ['not enough info', 'not enough information']:
if 'not enough' in pre:
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
else:
if (correct_predict.replace('.','').strip() in pre.replace('.','').strip() )&(instance[sample][final_thought]['candiate'][j] not in instance[sample]['correct']):
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
if (pre.replace('.','').strip() in correct_predict.replace('.','').strip() )&(instance[sample][final_thought]['candiate'][j] not in instance[sample]['correct']):
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
if ('yes' in pre.lower()) & ('yes' in correct_predict.lower()) & (instance[sample][final_thought]['candiate'][j] not in instance[sample]['correct']):
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
if ('no' in pre.lower()) & ('no' in correct_predict.lower()) & (instance[sample][final_thought]['candiate'][j] not in instance[sample]['correct']):
instance[sample]['correct'].append(instance[sample][final_thought]['candiate'][j])
if 'correct' in instance[sample]:
if isinstance(instance[sample]['correct'], str):
instance[sample]['correct'] = [instance[sample]['correct']]
Corpus[sample] = {}
for cor in instance[sample]['correct']:
correct_list = cor.lower().replace(' * ','*').replace(' + ','+').replace(' +','+').replace('+ ','+').replace(' = ','=').replace('= ','=').replace(' - ','-').replace(' -','-').replace('- ','-').replace(' x ','x').replace(' / ','/').replace('/ ','/').replace('. so the final answer is', '. step 3, so the final answer is').split('step')
correct_list = correct_list[1:]
# print(1111)
if len(correct_list) < thought_number:
continue
if final_sentence.split(', ')[1] not in correct_list[int(final_thought)]:
continue
for thought_idx in range(thought_number):
if str(thought_idx) not in Corpus[sample]:
Corpus[sample][str(thought_idx)] = {}
if 'pos' not in Corpus[sample][str(thought_idx)]:
Corpus[sample][str(thought_idx)]['pos'] = []
Corpus[sample][str(thought_idx)]['neg'] = []
Corpus[sample][str(thought_idx)]['prompt'] = []
choice_list = instance[sample][str(thought_idx)]['candiate']
if thought_idx == 0:
pos = 'step' + correct_list[thought_idx]
pos = pos.strip().lower()
if pos in Corpus[sample][str(thought_idx)]['pos']:
continue
Corpus[sample][str(thought_idx)]['pos'].append(pos)
neg_template = []
for choice in choice_list:
choice = choice.strip().replace(' * ','*').replace(' + ','+').replace('+ ','+').replace(' +','+').replace(' ？','？').lower().replace(' = ','=').replace('= ','=').replace(' x ','x').replace(' - ','-').replace(' -','-').replace('- ','-').replace('answer: ','').replace('..','.').replace(' ',' ').replace('\"','\'').replace(' / ','/').replace('/ ','/').replace(' ？','？')
# choice = choice.replace('\'','')
if choice != pos:
neg_template.append(choice)
Corpus[sample][str(thought_idx)]['neg'].append(list(set(neg_template)))
Corpus[sample][str(thought_idx)]['prompt'].append('')
else:
# print(correct_list)
pos = 'step' + correct_list[thought_idx]
pos = pos.lower().strip()
neg_template = []
neg_correct_part = 'step'+'step'.join(correct_list[:thought_idx])
neg_correct_part = neg_correct_part.strip()
if (pos in Corpus[sample][str(thought_idx)]['pos'])&(neg_correct_part in Corpus[sample][str(thought_idx)]['prompt']):
continue
Corpus[sample][str(thought_idx)]['pos'].append(pos)
for choice in choice_list:
choice = choice.strip().lower().replace(' * ','*').replace(' + ','+').replace('+ ','+').replace(' +','+').replace(' ？','？').replace(' = ','=').replace('= ','=').replace(' - ','-').replace(' -','-').replace('- ','-').replace('answer: ','').replace(' ',' ').replace('..','.').replace('\"','\'').replace(' / ','/').replace('/ ','/').replace(' ？','？').replace('therefore, the final answer is','so the final answer is').replace('. so the final answer is', '. step 3, so the final answer is')
if (thought_idx == 1)&('step 3' in choice):
choice = choice.split('step 3')[0]
if len(choice.split('step 3'))>2:
choice = 'step 3'.join(choice.split('step 3')[:-1])
if neg_correct_part in choice:
choice = choice.replace(neg_correct_part,'').strip()
if choice.replace('answer: ','').strip().startswith('step') == False:
choice = 'step'+ 'step'.join(choice.split('step')[1:])
if choice != pos:
neg_template.append(choice)
Corpus[sample][str(thought_idx)]['neg'].append(list(set(neg_template)))
Corpus[sample][str(thought_idx)]['prompt'].append(neg_correct_part)
for thought_idx in range(thought_number):
neg_samples = Corpus[sample][str(thought_idx)]['neg']
for j in range(len(neg_samples)):
neg_sample = neg_samples[j]
filtered_neg = []
for i in range(len(neg_sample)):
neg_s = neg_sample[i].replace('answer: ','').replace(' + ','+').replace('+ ','+').replace(' +','+').replace(' / ','/').replace('/ ','/').replace(' ',' ').replace('..','.').replace('= ','=').replace(' - ','-').replace(' -','-').replace('- ','-').replace(' x ','x').replace('\"','\'').replace(' ？','？').replace('therefore, the final answer is','so the final answer is').replace('. so the final answer is', '. step 3, so the final answer is')
# print(neg_s)
# print(Corpus[sample][str(thought_idx)]['pos'])
if neg_s not in Corpus[sample][str(thought_idx)]['pos']:
filtered_neg.append(neg_s)
neg_samples[j] = filtered_neg
Corpus[sample][str(thought_idx)]['neg'] = neg_samples
if len(Corpus[sample]) == 0:
del Corpus[sample]
paired_data = []
for instance in Corpus:
for thought_idx in range(thought_number):
for i, pos in enumerate(Corpus[instance][str(thought_idx)]['pos']):
if task == 'math':
propmt = create_demo_text() + "Q: " + instance + "\nA: " + Corpus[instance][str(thought_idx)]['prompt'][i]
else:
propmt = task.cot_prompt_wrap(instance, Corpus[instance][str(thought_idx)]['prompt'][i])
if len(Corpus[instance][str(thought_idx)]['neg'][i])==0:
continue
for neg in Corpus[instance][str(thought_idx)]['neg'][i]:
if len(neg) <=3:
continue
if ('fever' in path) or (('vitaminc' in path)):
if final_sentence in pos:
if ('suport' in pos) or ('support' in pos):
correct_predict = 'supports'
if ('refu' in pos) or ('reject' in pos):
correct_predict = 'refutes'
if ('not enough' in pos) or ('no enough' in pos):
correct_predict = 'not enough information'
pos = final_sentence + correct_predict + '.'
if ('math' in path) or ('svamp' in path):
if 'step 4' in pos:
if '(arabic numerals) is ' in pos:
# print(111)
pos_ = pos.split('(arabic numerals) is ')[1]
pos_ = pos_.replace(',','')
numbers = re.findall(r'\d+', pos_)
if len(numbers) == 0:
# print(instance)
# print(pos_)
# exit()
continue
else:
continue
# if random.randint(0, 1) == 0:
# pos,neg = neg,pos
pos = pos.replace(' ',' ').strip().replace('..','.').replace(' .','.').replace('..','.').replace('..','.')
neg = neg.replace('answer: ','').replace('..','.').strip().replace(' .','.').replace('..','.').replace('..','.')
if ('=' in pos):
if ('-' not in pos) & ('+' not in pos) & ('/' not in pos) & ('*' not in pos):
continue
if ('=' in neg):
if ('-' not in neg) & ('+' not in neg) & ('/' not in neg) & ('*' not in neg):
continue
if '841 34.' in pos:
continue
if pos[-1] != '.':
pos_split = pos.split('.')
if len(pos_split)<=1:
continue
else:
pos = '.'.join(pos_split[:-1]) + '.'
if neg[-1] != '.':
neg_split = neg.split('.')
if len(neg_split)<=1:
continue
else:
neg = '.'.join(neg_split[:-1]) + '.'
if '[tex]' in neg:
continue
if {
"prompt":
propmt
,
"chosen": pos, # rated better than k
"rejected": neg, # rated worse than j
} not in paired_data:
paired_data.append(
{
"prompt":
propmt
,
"chosen": pos, # rated better than k
"rejected": neg, # rated worse than j
}
)
print(len(paired_data))
with open(file_name,'w') as f:
f.write(json.dumps(paired_data,ensure_ascii=False,indent=4))