Commit
·
0629b4d
1
Parent(s):
824b3b4
feat: dataloader
Browse files
logiqa.py
ADDED
@@ -0,0 +1,106 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""LogiQA: A Challenge Dataset for Machine Reading Comprehension with Logical Reasoning"""
|
2 |
+
|
3 |
+
import re
|
4 |
+
import datasets
|
5 |
+
|
6 |
+
logger = datasets.logging.get_logger(__name__)
|
7 |
+
|
8 |
+
|
9 |
+
_HOMEPAGE = "https://github.com/lgw863/LogiQA-dataset"
|
10 |
+
|
11 |
+
_DESCRIPTION = """\
|
12 |
+
LogiQA is constructed from the logical comprehension problems from \
|
13 |
+
publically available questions of the National Civil Servants Examination \
|
14 |
+
of China, which are designed to test the civil servant candidates’ critical \
|
15 |
+
thinking and problem solving. This dataset includes the English versions only; \
|
16 |
+
the Chinese versions are available via the homepage/original source."""
|
17 |
+
|
18 |
+
_CITATION = """\
|
19 |
+
@article{liu2020logiqa,
|
20 |
+
title={Logiqa: A challenge dataset for machine reading comprehension with logical reasoning},
|
21 |
+
author={Liu, Jian and Cui, Leyang and Liu, Hanmeng and Huang, Dandan and Wang, Yile and Zhang, Yue},
|
22 |
+
journal={arXiv preprint arXiv:2007.08124},
|
23 |
+
year={2020}
|
24 |
+
}
|
25 |
+
"""
|
26 |
+
|
27 |
+
_URLS = {
|
28 |
+
"en_train": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Train.txt",
|
29 |
+
"en_test": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Test.txt",
|
30 |
+
"en_eval": "https://raw.githubusercontent.com/lgw863/LogiQA-dataset/master/Eval.txt",
|
31 |
+
}
|
32 |
+
|
33 |
+
def _process_answer(answer):
|
34 |
+
if not any(answer.startswith(x) for x in "ABCD"):
|
35 |
+
return answer
|
36 |
+
else:
|
37 |
+
return answer[3:]
|
38 |
+
|
39 |
+
def _process_sentences(text):
|
40 |
+
text = text.replace("\n", "")
|
41 |
+
sents = text.split(".")
|
42 |
+
text = ""
|
43 |
+
for sent in sents:
|
44 |
+
if len(sent) == 0:
|
45 |
+
continue
|
46 |
+
if len(text) == 0:
|
47 |
+
text += sent
|
48 |
+
elif sent[0].isnumeric():
|
49 |
+
text += "."+sent
|
50 |
+
else:
|
51 |
+
text += ". "+sent
|
52 |
+
text = text.replace(" ", " ")
|
53 |
+
text = text.replace("\\'", "'")
|
54 |
+
while text.endswith(" "):
|
55 |
+
text = text[:-1]
|
56 |
+
if re.match('^[A-Z][\w\s]+[?.!]$', text) is None:
|
57 |
+
text += "."
|
58 |
+
text = text.replace("?.", "?")
|
59 |
+
text = text.replace("!.", "!")
|
60 |
+
text = text.replace("..", ".")
|
61 |
+
return text
|
62 |
+
|
63 |
+
class LogiQA(datasets.GeneratorBasedBuilder):
|
64 |
+
def _info(self):
|
65 |
+
features = datasets.Features(
|
66 |
+
{
|
67 |
+
"context": datasets.Value("string"),
|
68 |
+
"query": datasets.Value("string"),
|
69 |
+
"options": datasets.features.Sequence(datasets.Value("string")),
|
70 |
+
"correct_option": datasets.Value("int32")
|
71 |
+
}
|
72 |
+
)
|
73 |
+
return datasets.DatasetInfo(
|
74 |
+
description=_DESCRIPTION,
|
75 |
+
features=features,
|
76 |
+
homepage=_HOMEPAGE,
|
77 |
+
citation=_CITATION,
|
78 |
+
)
|
79 |
+
|
80 |
+
def _split_generators(self, dl_manager):
|
81 |
+
downloaded_files = dl_manager.download_and_extract(_URLS)
|
82 |
+
return [
|
83 |
+
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["en_train"]}),
|
84 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["en_eval"]}),
|
85 |
+
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["en_test"]}),
|
86 |
+
]
|
87 |
+
|
88 |
+
def _generate_examples(self, filepath):
|
89 |
+
logger.info("generating examples from = %s", filepath)
|
90 |
+
with open(filepath, encoding="utf-8") as f:
|
91 |
+
logiqa = f.readlines()
|
92 |
+
logiqa = [_process_sentences(s) for s in logiqa]
|
93 |
+
|
94 |
+
for key in range(int(len(logiqa)/8)):
|
95 |
+
row = 8*key
|
96 |
+
correct_answer = logiqa[row+1].replace(".","")
|
97 |
+
context = logiqa[row+2]
|
98 |
+
query = logiqa[row+3]
|
99 |
+
answers = logiqa[row+4:row+8]
|
100 |
+
|
101 |
+
yield key, {
|
102 |
+
"context": context,
|
103 |
+
"query": query,
|
104 |
+
"options": [_process_answer(answers[i]) for i in range(4)],
|
105 |
+
"correct_option": "abcd".index(correct_answer)
|
106 |
+
}
|