add datasets script
Browse files- KorQuADv1.py +103 -0
KorQuADv1.py
ADDED
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
import pandas as pd
|
3 |
+
|
4 |
+
_DATA_URLS = {
|
5 |
+
"train": "https://korquad.github.io/dataset/KorQuAD_v1.0_train.json",
|
6 |
+
"dev": "https://korquad.github.io/dataset/KorQuAD_v1.0_dev.json",
|
7 |
+
}
|
8 |
+
_VERSION = "0.0.0"
|
9 |
+
_LICENSE = "CC BY-ND 2.0 KR"
|
10 |
+
_CITATION = """\
|
11 |
+
@misc{lim2019korquad1.0,
|
12 |
+
title={KorQuAD1.0: Korean QA Dataset for Machine Reading Comprehension},
|
13 |
+
author={Seungyoung Lim, Myungji Kim, Jooyoul Lee},
|
14 |
+
year={2019},
|
15 |
+
eprint={1909.07005},
|
16 |
+
archivePrefix={arXiv},
|
17 |
+
primaryClass={cs.CL}
|
18 |
+
}
|
19 |
+
"""
|
20 |
+
_HOMEPAGE = "https://korquad.github.io/KorQuad%201.0/"
|
21 |
+
_DESCRIPTION = """\
|
22 |
+
KorQuAD 1.0 (Korean Question Answering Dataset v1.0)
|
23 |
+
KorQuAD 1.0 is a dataset created for Korean Machine Reading Comprehension.
|
24 |
+
The answers to all your questions are made up of some subareas in the corresponding Wikipedia article paragraphs.
|
25 |
+
It is structured in the same way as the Stanford Question Answering Dataset (SQuAD) v1.0.
|
26 |
+
"""
|
27 |
+
|
28 |
+
|
29 |
+
class KorQuADV1Config(datasets.BuilderConfig):
|
30 |
+
def __init__(self, data_url, features, **kwargs):
|
31 |
+
super().__init__(version=datasets.Version(_VERSION, ""), **kwargs)
|
32 |
+
self.features = features
|
33 |
+
self.data_url = data_url
|
34 |
+
|
35 |
+
|
36 |
+
class KorQuADV1(datasets.GeneratorBasedBuilder):
|
37 |
+
DEFAULT_CONFIG_NAME = "korquad"
|
38 |
+
BUILDER_CONFIGS = [
|
39 |
+
KorQuADV1Config(
|
40 |
+
name="korquad",
|
41 |
+
data_url=_DATA_URLS,
|
42 |
+
features=datasets.Features(
|
43 |
+
{
|
44 |
+
"answers": datasets.Sequence(
|
45 |
+
feature={
|
46 |
+
"text": datasets.Value(dtype="string"),
|
47 |
+
"answer_start": datasets.Value(dtype="int32"),
|
48 |
+
},
|
49 |
+
),
|
50 |
+
"context": datasets.Value(dtype="string"),
|
51 |
+
"guid": datasets.Value(dtype="string"),
|
52 |
+
"question": datasets.Value(dtype="string"),
|
53 |
+
"title": datasets.Value(dtype="string"),
|
54 |
+
}
|
55 |
+
)
|
56 |
+
),
|
57 |
+
]
|
58 |
+
|
59 |
+
def _info(self):
|
60 |
+
return datasets.DatasetInfo(
|
61 |
+
features=self.config.features,
|
62 |
+
description=_DESCRIPTION,
|
63 |
+
homepage=_HOMEPAGE,
|
64 |
+
citation=_CITATION,
|
65 |
+
license=_LICENSE,
|
66 |
+
)
|
67 |
+
|
68 |
+
def _split_generators(self, dl_manager):
|
69 |
+
data_path = dl_manager.download(self.config.data_url)
|
70 |
+
return [
|
71 |
+
datasets.SplitGenerator(
|
72 |
+
name=datasets.Split.TRAIN,
|
73 |
+
# These kwargs will be passed to _generate_examples
|
74 |
+
gen_kwargs={"data_file": data_path["train"]}
|
75 |
+
),
|
76 |
+
datasets.SplitGenerator(
|
77 |
+
name=datasets.Split.VALIDATION,
|
78 |
+
# These kwargs will be passed to _generate_examples
|
79 |
+
gen_kwargs={"data_file": data_path["dev"]}
|
80 |
+
),
|
81 |
+
]
|
82 |
+
|
83 |
+
def _generate_examples(self, data_file: str):
|
84 |
+
idx = 0
|
85 |
+
korquad = pd.read_json(data_file)
|
86 |
+
for example in korquad["data"].tolist():
|
87 |
+
paragraphs = example["paragraphs"]
|
88 |
+
title = example["title"]
|
89 |
+
for paragraph in paragraphs:
|
90 |
+
qas = paragraph["qas"]
|
91 |
+
context = paragraph["context"]
|
92 |
+
for qa in qas:
|
93 |
+
text = [answers["text"] for answers in qa["answers"]]
|
94 |
+
answer_start = [answers["answer_start"] for answers in qa["answers"]]
|
95 |
+
features = {
|
96 |
+
"guid": qa["id"],
|
97 |
+
"question": qa["question"],
|
98 |
+
"answers": {"text": text, "answer_start": answer_start},
|
99 |
+
"context": context,
|
100 |
+
"title": title,
|
101 |
+
}
|
102 |
+
yield idx, features
|
103 |
+
idx += 1
|