holylovenia commited on
Commit
a5b8b50
1 Parent(s): 22d4069

Upload vimmrc.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. vimmrc.py +198 -0
vimmrc.py ADDED
@@ -0,0 +1,198 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ import json
17
+ from pathlib import Path
18
+ from typing import Dict, List, Tuple
19
+
20
+ import datasets
21
+
22
+ from seacrowd.utils.configs import SEACrowdConfig
23
+ from seacrowd.utils.constants import (SCHEMA_TO_FEATURES, TASK_TO_SCHEMA,
24
+ Licenses, Tasks)
25
+
26
+ _CITATION = """
27
+ @ARTICLE{vimmrc,
28
+ author={Nguyen, Kiet Van and Tran, Khiem Vinh and Luu, Son T. and Nguyen, Anh Gia-Tuan and Nguyen, Ngan Luu-Thuy},
29
+ journal={IEEE Access},
30
+ title={Enhancing Lexical-Based Approach With External Knowledge for Vietnamese Multiple-Choice Machine Reading Comprehension},
31
+ year={2020},
32
+ volume={8},
33
+ pages={201404-201417},
34
+ doi={10.1109/ACCESS.2020.3035701}}
35
+ """
36
+
37
+ _DATASETNAME = "vimmrc"
38
+
39
+ _DESCRIPTION = """
40
+ ViMMRC, a challenging machine comprehension corpus with multiple-choice questions,
41
+ intended for research on the machine comprehension of Vietnamese text. This corpus
42
+ includes 2,783 multiple-choice questions and answers based on a set of 417 Vietnamese
43
+ texts used for teaching reading comprehension for 1st to 5th graders.
44
+ """
45
+
46
+ _HOMEPAGE = "https://sites.google.com/uit.edu.vn/kietnv/datasets#h.1qeaynfs79d1"
47
+
48
+ _LANGUAGES = ["vie"]
49
+
50
+ _LICENSE = f"{Licenses.UNKNOWN.value} | The corpus is freely available at our website for research purposes."
51
+
52
+ _LOCAL = False
53
+
54
+ _URL = "https://drive.google.com/file/d/14Rq-YANUv8qyi4Ze8ReEAEu_uxgcV_Yk/view" # ~2mb
55
+
56
+ _SUPPORTED_TASKS = [Tasks.COMMONSENSE_REASONING]
57
+ _SEACROWD_SCHEMA = f"seacrowd_{TASK_TO_SCHEMA[_SUPPORTED_TASKS[0]].lower()}" # qa
58
+
59
+ _SOURCE_VERSION = "1.0.0"
60
+
61
+ _SEACROWD_VERSION = "2024.06.20"
62
+
63
+
64
+ class ViMMRCDataset(datasets.GeneratorBasedBuilder):
65
+ """A Vietnamese machine comprehension corpus with multiple-choice questions"""
66
+
67
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
68
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
69
+
70
+ BUILDER_CONFIGS = [
71
+ SEACrowdConfig(
72
+ name=f"{_DATASETNAME}_source",
73
+ version=SOURCE_VERSION,
74
+ description=f"{_DATASETNAME} source schema",
75
+ schema="source",
76
+ subset_id=_DATASETNAME,
77
+ ),
78
+ SEACrowdConfig(
79
+ name=f"{_DATASETNAME}_{_SEACROWD_SCHEMA}",
80
+ version=SEACROWD_VERSION,
81
+ description=f"{_DATASETNAME} SEACrowd schema",
82
+ schema=_SEACROWD_SCHEMA,
83
+ subset_id=_DATASETNAME,
84
+ ),
85
+ ]
86
+
87
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
88
+
89
+ def _info(self) -> datasets.DatasetInfo:
90
+ if self.config.schema == "source":
91
+ features = datasets.Features(
92
+ {
93
+ "file_path": datasets.Value("string"),
94
+ "article": datasets.Value("string"),
95
+ "question": datasets.Value("string"),
96
+ "choices": datasets.Sequence(datasets.Value("string")),
97
+ "answer": datasets.Value("string"),
98
+ }
99
+ )
100
+ elif self.config.schema == _SEACROWD_SCHEMA:
101
+ features = SCHEMA_TO_FEATURES[TASK_TO_SCHEMA[_SUPPORTED_TASKS[0]]] # qa_features
102
+
103
+ return datasets.DatasetInfo(
104
+ description=_DESCRIPTION,
105
+ features=features,
106
+ homepage=_HOMEPAGE,
107
+ license=_LICENSE,
108
+ citation=_CITATION,
109
+ )
110
+
111
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
112
+ """Returns SplitGenerators."""
113
+ # check if gdown is installed
114
+ try:
115
+ import gdown
116
+ except ImportError as err:
117
+ raise ImportError("Please install `gdown` to enable reliable data download from google drive.") from err
118
+
119
+ # download data from gdrive
120
+ output_dir = Path.cwd() / "data" / "vimmrc"
121
+ output_dir.mkdir(parents=True, exist_ok=True)
122
+ output_file = output_dir / "vimmrc.zip"
123
+ if not output_file.exists():
124
+ gdown.download(_URL, str(output_file), fuzzy=True)
125
+ else:
126
+ print(f"File already downloaded: {str(output_file)}")
127
+
128
+ # extract data
129
+ data_dir = Path(dl_manager.extract(output_file)) / "ViMMRC"
130
+
131
+ return [
132
+ datasets.SplitGenerator(
133
+ name=datasets.Split.TRAIN,
134
+ gen_kwargs={
135
+ "data_dir": data_dir / "train",
136
+ },
137
+ ),
138
+ datasets.SplitGenerator(
139
+ name=datasets.Split.VALIDATION,
140
+ gen_kwargs={
141
+ "data_dir": data_dir / "dev",
142
+ },
143
+ ),
144
+ datasets.SplitGenerator(
145
+ name=datasets.Split.TEST,
146
+ gen_kwargs={
147
+ "data_dir": data_dir / "test",
148
+ },
149
+ ),
150
+ ]
151
+
152
+ def _generate_examples(self, data_dir: Path) -> Tuple[int, Dict]:
153
+ """Yields examples as (key, example) tuples."""
154
+ # a data_dir consists of several json files
155
+ json_files = sorted(list(data_dir.glob("*.json")))
156
+
157
+ key = 0
158
+ for json_file in json_files:
159
+ with open(json_file, "r", encoding="utf-8") as file:
160
+ # load per json file
161
+ data = json.load(file)
162
+ assert len(data["questions"]) == len(data["options"]) == len(data["answers"]), f"Mismatched data length on {str(json_file)}"
163
+
164
+ for idx, question in enumerate(data["questions"]):
165
+
166
+ # get answer based on the answer key
167
+ if data["answers"][idx] == "A":
168
+ answer = data["options"][idx][0]
169
+ elif data["answers"][idx] == "B":
170
+ answer = data["options"][idx][1]
171
+ elif data["answers"][idx] == "C":
172
+ answer = data["options"][idx][2]
173
+ elif data["answers"][idx] == "D":
174
+ answer = data["options"][idx][3]
175
+
176
+ if self.config.schema == "source":
177
+ yield key, {
178
+ "file_path": str(json_file),
179
+ "article": data["article"],
180
+ "question": question,
181
+ "choices": data["options"][idx],
182
+ "answer": answer,
183
+ }
184
+ key += 1
185
+
186
+ elif self.config.schema == _SEACROWD_SCHEMA:
187
+ yield key, {
188
+ "id": key,
189
+ "question_id": None,
190
+ "document_id": str(json_file),
191
+ "question": question,
192
+ "type": "multiple_choice",
193
+ "choices": data["options"][idx],
194
+ "context": data["article"],
195
+ "answer": [answer],
196
+ "meta": None,
197
+ }
198
+ key += 1