MotzWanted
commited on
Commit
·
6622076
1
Parent(s):
6e3fb43
Create medwiki file
Browse files- dataset.py +121 -0
dataset.py
ADDED
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import re
|
3 |
+
from pathlib import Path
|
4 |
+
|
5 |
+
import datasets
|
6 |
+
|
7 |
+
|
8 |
+
logger = datasets.logging.get_logger(__name__)
|
9 |
+
|
10 |
+
TXT_PATTERN = r"^.*\.txt$"
|
11 |
+
|
12 |
+
_CITATION = """\
|
13 |
+
@misc{https://doi.org/10.48550/arxiv.2210.06345,
|
14 |
+
doi = {10.48550/ARXIV.2210.06345},
|
15 |
+
|
16 |
+
url = {https://arxiv.org/abs/2210.06345},
|
17 |
+
|
18 |
+
author = {Liévin, Valentin and Motzfeldt, Andreas Geert and Jensen, Ida Riis and Winther, Ole},
|
19 |
+
|
20 |
+
keywords = {Computation and Language (cs.CL), Information Retrieval (cs.IR), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences, I.2.7; H.3.3; I.2.1},
|
21 |
+
|
22 |
+
title = {Variational Open-Domain Question Answering},
|
23 |
+
|
24 |
+
publisher = {arXiv},
|
25 |
+
|
26 |
+
year = {2022},
|
27 |
+
|
28 |
+
copyright = {arXiv.org perpetual, non-exclusive license}
|
29 |
+
}
|
30 |
+
"""
|
31 |
+
|
32 |
+
_VERSION = "0.0.1"
|
33 |
+
_HOMEPAGE = "https://github.com/VodLM"
|
34 |
+
|
35 |
+
_DESCRIPTION = """\
|
36 |
+
The MedWiki corpus, which is distributed under the MIT license,
|
37 |
+
consists of a subset of 4.5% of the English Wikipedia articles
|
38 |
+
and has been specifically curated for the MedMCQA and USMLE datasets.
|
39 |
+
The collection was created by utilizing the Wikipedia API to search
|
40 |
+
for articles related to each answer option present in the MedMCQA and
|
41 |
+
USMLE datasets. The top ten Wikipedia articles for each answer option
|
42 |
+
were selected and included in the final corpus. This subset covers a
|
43 |
+
wide range of topics in the field of medicine that could be relevant
|
44 |
+
for answering questions in this domain.
|
45 |
+
questions.
|
46 |
+
"""
|
47 |
+
|
48 |
+
_URL = "https://f001.backblazeb2.com/file/FindZebraData/fz-openqa/datasets/medwiki_v6.zip"
|
49 |
+
|
50 |
+
|
51 |
+
class MedWikipediaCorpusConfig(datasets.BuilderConfig):
|
52 |
+
"""BuilderConfig for the MedQa English Corpus objecxt."""
|
53 |
+
|
54 |
+
def __init__(self, **kwargs):
|
55 |
+
"""BuilderConfig for the Corpus object.
|
56 |
+
Args:
|
57 |
+
**kwargs: keyword arguments forwarded to super.
|
58 |
+
"""
|
59 |
+
super(MedWikipediaCorpusConfig, self).__init__(**kwargs)
|
60 |
+
|
61 |
+
class MedWikipediaCorpusGenerator(datasets.GeneratorBasedBuilder):
|
62 |
+
"""MedWikipediaCorpus Dataset. Version 0.0.1"""
|
63 |
+
|
64 |
+
BUILDER_CONFIGS = [
|
65 |
+
MedWikipediaCorpusConfig(
|
66 |
+
name="plain_text",
|
67 |
+
version=datasets.Version(_VERSION, ""),
|
68 |
+
description="Plain text",
|
69 |
+
),
|
70 |
+
]
|
71 |
+
|
72 |
+
def _info(self):
|
73 |
+
return datasets.DatasetInfo(
|
74 |
+
description=_DESCRIPTION,
|
75 |
+
features=datasets.Features(
|
76 |
+
{
|
77 |
+
"document.idx": datasets.Value("int32"),
|
78 |
+
"document.text": datasets.Value("string"),
|
79 |
+
"document.title": datasets.Value("string"),
|
80 |
+
}
|
81 |
+
),
|
82 |
+
supervised_keys=None,
|
83 |
+
homepage=_HOMEPAGE,
|
84 |
+
citation=_CITATION,
|
85 |
+
)
|
86 |
+
|
87 |
+
def _split_generators(self, dl_manager):
|
88 |
+
"""Returns SplitGenerators."""
|
89 |
+
downloaded_file = dl_manager.download_and_extract(_URL)
|
90 |
+
if not Path(downloaded_file).is_dir():
|
91 |
+
raise Exception(
|
92 |
+
f"Could not download the dataset Content of `downloaded_file`:"
|
93 |
+
f"{open(downloaded_file, 'r').read()}"
|
94 |
+
)
|
95 |
+
return [
|
96 |
+
datasets.SplitGenerator(
|
97 |
+
name=datasets.Split.TRAIN,
|
98 |
+
gen_kwargs={"output_dir": Path(downloaded_file)},
|
99 |
+
)
|
100 |
+
]
|
101 |
+
|
102 |
+
def _generate_examples(self, output_dir: str):
|
103 |
+
logger.info("generating examples from = %s", output_dir)
|
104 |
+
paths = [
|
105 |
+
p
|
106 |
+
for p in Path(output_dir).iterdir()
|
107 |
+
if p.is_dir() and (p.name.startswith("med_x_wiki") or p.name.startswith("wikipedia"))
|
108 |
+
]
|
109 |
+
assert len(paths) == 1, f"Found {len(paths)} directories in {output_dir}: {paths}"
|
110 |
+
path = paths[0]
|
111 |
+
|
112 |
+
# list files
|
113 |
+
data_files = [os.path.join(path, p) for p in os.listdir(path) if re.findall(TXT_PATTERN, p)]
|
114 |
+
|
115 |
+
# iterate and yield documents
|
116 |
+
for i, fn in enumerate(data_files):
|
117 |
+
with open(fn, "r") as f:
|
118 |
+
# the first line is the title
|
119 |
+
title = f.readline()
|
120 |
+
text = f.read()
|
121 |
+
yield i, {"document.text": text, "document.idx": i, "document.title": title}
|