-
Notifications
You must be signed in to change notification settings - Fork 25
/
data_utils.py
214 lines (189 loc) · 7.92 KB
/
data_utils.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
import os
import jsonlines
import re
from nltk.tokenize import sent_tokenize
from rouge_score import rouge_scorer
class MyDataset:
def __init__(self, split, args, eval_only=False, traindata_obj=None):
#self.counter = 0
if hasattr(args, 'start_pos'):
self.start_pos = args.start_pos
if hasattr(args, 'end_pos'):
self.end_pos = args.end_pos
if hasattr(args, 'model_name'):
self.model_name = args.model_name
self.dataset_name = args.dataset_name
self.dir_path = args.dataset_dir
self.split = split # train / test
self.load() # load dataset -> load data in self.data
# load answers -> self.choice_ref / self.ref
if args.dataset_name == 'MedQA':
self.build_choice_ref_MedQA()
elif args.dataset_name == 'MedMCQA' or 'MMLU' in args.dataset_name:
self.build_choice_ref_MedMCQA()
elif args.dataset_name == 'PubMedQA':
self.build_choice_ref_MedMCQA()
elif args.dataset_name == 'MedicationQA':
self.build_ref()
def load(self): # load dataset -> self.data
filename = os.path.join(self.dir_path, self.split + '.jsonl')
self.data = []
with open(filename) as f:
for item in jsonlines.Reader(f):
self.data.append(item)
def get_by_idx(self, idx):
return self.data[idx]
def __len__(self):
return len(self.data)
def build_ref(self):
self.ref = []
for i in range(len(self)):
item = self.get_by_idx(i)
self.ref.append({'answers': {'text': item['answer']}, 'id': i})
def build_choice_ref_MedQA(self):
self.choice_ref = []
for i in range(len(self)):
item = self.get_by_idx(i)
self.choice_ref.append({
'answers': {'text': item['answer'],'choice': item['answer_idx']},
'options': item['options'],
'type': item['meta_info'],
'id': i})
def build_choice_ref_MedMCQA(self):
self.choice_ref = []
for i in range(len(self)):
item = self.get_by_idx(i)
self.choice_ref.append({
'answers': {'text': item['answer'],
'choice': item['answer_idx']},
'options': item['options'],
'id': i})
def compute_rougescore(self, preds):
sum_score = 0.0
scorer = rouge_scorer.RougeScorer(['rouge1', 'rouge2', 'rougeL'], use_stemmer=True)
for i, answer in enumerate(preds):
correct_answer = self.ref[i]['answers']['text']
# correct_answer = correct_answer.replace('\n', ' ')
score = scorer.score(answer, correct_answer)
sum_score += score['rouge1'].fmeasure
# print(f'id: {i}, answer: {answer}, correct answer: {correct_answer}, rouge1 score: {score["rouge1"].fmeasure}')
# print(score)
# break
return sum_score / len(preds)
def compute_accuracy(self, preds):
if 'PubMedQA' in self.dir_path:
correct_num = 0.0
all_num = 0.0
for i, answer in enumerate(preds):
all_num += 1
correct_choice = self.choice_ref[i]['answers']['choice']
correct_answer = self.choice_ref[i]['answers']['text']
if answer == correct_choice or correct_answer in answer:
correct_num += 1
# print(f"id: {i}, choice: {answer}, correct choice: {correct_choice}")
print(f"correct_num: {correct_num}, all_num: {all_num}")
return correct_num / all_num
elif 'MedQA' in self.dir_path:
correct_num = {'step1': 0.0, 'step2&3': 0.0, 'all': 0.0}
all_num = {'step1': 0.0, 'step2&3': 0.0, 'all': 0.0}
for i, answer in enumerate(preds):
# choice = answer[:3]
answer = answer.strip()
all_num['all'] += 1
correct_choice = self.choice_ref[i]['answers']['choice']
correct_answer = self.choice_ref[i]['answers']['text']
type = self.choice_ref[i]['type']
all_num[type] += 1
if answer == correct_choice or (correct_choice in answer and answer != 'ERROR') or correct_answer in answer:
correct_num[type] += 1
correct_num['all'] += 1
# print(f"id: {i}, choice: {answer}, correct choice: {correct_choice}")
print(f"correct_num: {correct_num}, all_num: {all_num}")
return [correct_num[key] / all_num[key] for key in ['step1', 'step2&3', 'all']]
elif 'MedMCQA' in self.dir_path or 'MMLU' in self.dir_path:
correct_num = 0.0
all_num = 0.0
for i, answer in enumerate(preds):
# choice = answer[:3]
all_num += 1
correct_choice = self.choice_ref[i]['answers']['choice']
correct_answer = self.choice_ref[i]['answers']['text']
if answer == correct_choice or correct_answer in answer:
correct_num += 1
# print(f"id: {i}, choice: {answer}, correct choice: {correct_choice}")
print(f"correct_num: {correct_num}, all_num: {all_num}")
return correct_num / all_num
def remove_incomplete_sentence(text):
sentences = sent_tokenize(text)
if len(sentences) > 1 and sentences[-1][-1] != '.':
return ' '.join(sentences[:-1]) + '.' #remove the last sentence
else:
return text
def cleansing_analysis(analyses, domains, type):
analysis = {}
for i, item in enumerate(analyses):
if item == "ERROR.":
item = f"There is no analysis for this {type}."
item = remove_incomplete_sentence(item)
if "as an ai language model" in item.lower():
end_index = item.lower().find("as an ai language model")+len("as an ai language model")
item= item[end_index:].strip().strip(',').strip()
analysis[domains[i]] = item
return analysis
def cleansing_syn_report(question, options, raw_synthesized_report):
tmp = raw_synthesized_report.split("Total Analysis:")
total_analysis_text = tmp[1].strip()
if "Key Knowledge" in tmp:
key_knowledge_text = tmp[0].split("Key Knowledge:")[-1].strip()
final_syn_repo = f"Question: {question} \n" \
f"Options: {options} \n" \
f"Key Knowledge: {key_knowledge_text} \n" \
f"Total Analysis: {total_analysis_text} \n"
else:
final_syn_repo = f"Question: {question} \n" \
f"Options: {options} \n" \
f"Total Analysis: {total_analysis_text} \n"
return final_syn_repo
def cleansing_final_output(output):
try:
ans = output.split(":")[-1]
ans = re.findall(r'A|B|C|D|E', ans)
if len(ans) == 0:
ans = ""
else:
ans = ans[0]
except:
ans = re.findall(r'A|B|C|D|E', ans)
if len(ans) == 0:
ans = ""
else:
ans = ans[-1]
return ans, output
def cleansing_voting(output):
output = output.lower()
ans = re.findall(r'yes|no', output)
if len(ans) == 0:
ans = "yes"
else:
ans = ans[0]
return ans
def transform_dict2text(analyses, type, content):
if type == "question":
report = ""
i = 0
for _domain, _analysis in analyses.items():
report += f"Report{i} \n" \
f"Question: {content} \n" \
f"Domain: {_domain} \n" \
f"Analysis: {_analysis} \n\n"
i += 1
elif type == "options":
report = ""
i = 0
for _domain, _analysis in analyses.items():
report += f"Report{i}: \n" \
f"Options: {content} \n" \
f"Domain: {_domain} \n" \
f"Analysis: {_analysis} \n\n"
i += 1
return report