Datasets:
Tasks:
Question Answering
Formats:
parquet
Sub-tasks:
multiple-choice-qa
Languages:
English
Size:
10K - 100K
ArXiv:
License:
Delete loading script
Browse files
codah.py
DELETED
|
@@ -1,141 +0,0 @@
|
|
| 1 |
-
# coding=utf-8
|
| 2 |
-
# Copyright 2020 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 |
-
"""The COmmonsense Dataset Adversarially-authored by Humans (CODAH)"""
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
import csv
|
| 19 |
-
|
| 20 |
-
import datasets
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
_CITATION = """\
|
| 24 |
-
@inproceedings{chen2019codah,
|
| 25 |
-
title={CODAH: An Adversarially-Authored Question Answering Dataset for Common Sense},
|
| 26 |
-
author={Chen, Michael and D'Arcy, Mike and Liu, Alisa and Fernandez, Jared and Downey, Doug},
|
| 27 |
-
booktitle={Proceedings of the 3rd Workshop on Evaluating Vector Space Representations for NLP},
|
| 28 |
-
pages={63--69},
|
| 29 |
-
year={2019}
|
| 30 |
-
}
|
| 31 |
-
"""
|
| 32 |
-
|
| 33 |
-
_DESCRIPTION = """\
|
| 34 |
-
The COmmonsense Dataset Adversarially-authored by Humans (CODAH) is an evaluation set for commonsense \
|
| 35 |
-
question-answering in the sentence completion style of SWAG. As opposed to other automatically \
|
| 36 |
-
generated NLI datasets, CODAH is adversarially constructed by humans who can view feedback \
|
| 37 |
-
from a pre-trained model and use this information to design challenging commonsense questions. \
|
| 38 |
-
Our experimental results show that CODAH questions present a complementary extension to the SWAG dataset, testing additional modes of common sense.
|
| 39 |
-
"""
|
| 40 |
-
|
| 41 |
-
_URL = "https://raw.githubusercontent.com/Websail-NU/CODAH/master/data/"
|
| 42 |
-
_FULL_DATA_URL = _URL + "full_data.tsv"
|
| 43 |
-
|
| 44 |
-
QUESTION_CATEGORIES_MAPPING = {
|
| 45 |
-
"i": "Idioms",
|
| 46 |
-
"r": "Reference",
|
| 47 |
-
"p": "Polysemy",
|
| 48 |
-
"n": "Negation",
|
| 49 |
-
"q": "Quantitative",
|
| 50 |
-
"o": "Others",
|
| 51 |
-
}
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
class CodahConfig(datasets.BuilderConfig):
|
| 55 |
-
"""BuilderConfig for CODAH."""
|
| 56 |
-
|
| 57 |
-
def __init__(self, fold=None, **kwargs):
|
| 58 |
-
"""BuilderConfig for CODAH.
|
| 59 |
-
|
| 60 |
-
Args:
|
| 61 |
-
fold: `string`, official cross validation fold.
|
| 62 |
-
**kwargs: keyword arguments forwarded to super.
|
| 63 |
-
"""
|
| 64 |
-
super(CodahConfig, self).__init__(**kwargs)
|
| 65 |
-
self.fold = fold
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
class Codah(datasets.GeneratorBasedBuilder):
|
| 69 |
-
"""The COmmonsense Dataset Adversarially-authored by Humans (CODAH)"""
|
| 70 |
-
|
| 71 |
-
VERSION = datasets.Version("1.0.0")
|
| 72 |
-
BUILDER_CONFIGS = [
|
| 73 |
-
CodahConfig(name="codah", version=datasets.Version("1.0.0"), description="Full CODAH dataset", fold=None),
|
| 74 |
-
CodahConfig(
|
| 75 |
-
name="fold_0", version=datasets.Version("1.0.0"), description="Official CV split (fold_0)", fold="fold_0"
|
| 76 |
-
),
|
| 77 |
-
CodahConfig(
|
| 78 |
-
name="fold_1", version=datasets.Version("1.0.0"), description="Official CV split (fold_1)", fold="fold_1"
|
| 79 |
-
),
|
| 80 |
-
CodahConfig(
|
| 81 |
-
name="fold_2", version=datasets.Version("1.0.0"), description="Official CV split (fold_2)", fold="fold_2"
|
| 82 |
-
),
|
| 83 |
-
CodahConfig(
|
| 84 |
-
name="fold_3", version=datasets.Version("1.0.0"), description="Official CV split (fold_3)", fold="fold_3"
|
| 85 |
-
),
|
| 86 |
-
CodahConfig(
|
| 87 |
-
name="fold_4", version=datasets.Version("1.0.0"), description="Official CV split (fold_4)", fold="fold_4"
|
| 88 |
-
),
|
| 89 |
-
]
|
| 90 |
-
|
| 91 |
-
def _info(self):
|
| 92 |
-
return datasets.DatasetInfo(
|
| 93 |
-
description=_DESCRIPTION,
|
| 94 |
-
features=datasets.Features(
|
| 95 |
-
{
|
| 96 |
-
"id": datasets.Value("int32"),
|
| 97 |
-
"question_category": datasets.features.ClassLabel(
|
| 98 |
-
names=["Idioms", "Reference", "Polysemy", "Negation", "Quantitative", "Others"]
|
| 99 |
-
),
|
| 100 |
-
"question_propmt": datasets.Value("string"),
|
| 101 |
-
"candidate_answers": datasets.features.Sequence(datasets.Value("string")),
|
| 102 |
-
"correct_answer_idx": datasets.Value("int32"),
|
| 103 |
-
}
|
| 104 |
-
),
|
| 105 |
-
supervised_keys=None,
|
| 106 |
-
homepage="https://github.com/Websail-NU/CODAH",
|
| 107 |
-
citation=_CITATION,
|
| 108 |
-
)
|
| 109 |
-
|
| 110 |
-
def _split_generators(self, dl_manager):
|
| 111 |
-
if self.config.name == "codah":
|
| 112 |
-
data_file = dl_manager.download(_FULL_DATA_URL)
|
| 113 |
-
return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_file": data_file})]
|
| 114 |
-
|
| 115 |
-
base_url = f"{_URL}cv_split/{self.config.fold}/"
|
| 116 |
-
_urls = {
|
| 117 |
-
"train": base_url + "train.tsv",
|
| 118 |
-
"dev": base_url + "dev.tsv",
|
| 119 |
-
"test": base_url + "test.tsv",
|
| 120 |
-
}
|
| 121 |
-
downloaded_files = dl_manager.download_and_extract(_urls)
|
| 122 |
-
|
| 123 |
-
return [
|
| 124 |
-
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_file": downloaded_files["train"]}),
|
| 125 |
-
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"data_file": downloaded_files["dev"]}),
|
| 126 |
-
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"data_file": downloaded_files["test"]}),
|
| 127 |
-
]
|
| 128 |
-
|
| 129 |
-
def _generate_examples(self, data_file):
|
| 130 |
-
with open(data_file, encoding="utf-8") as f:
|
| 131 |
-
rows = csv.reader(f, delimiter="\t")
|
| 132 |
-
for i, row in enumerate(rows):
|
| 133 |
-
question_category = QUESTION_CATEGORIES_MAPPING[row[0]] if row[0] != "" else -1
|
| 134 |
-
example = {
|
| 135 |
-
"id": i,
|
| 136 |
-
"question_category": question_category,
|
| 137 |
-
"question_propmt": row[1],
|
| 138 |
-
"candidate_answers": row[2:-1],
|
| 139 |
-
"correct_answer_idx": int(row[-1]),
|
| 140 |
-
}
|
| 141 |
-
yield i, example
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|