search_qa / search_qa.py
albertvillanova's picture
Host data files (#4)
06907e4
raw
history blame
7.49 kB
# coding=utf-8
# Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# Lint as: python3
"""SearchQA dataset."""
import itertools
import json
import datasets
_CITATION = r"""
@article{DBLP:journals/corr/DunnSHGCC17,
author = {Matthew Dunn and
Levent Sagun and
Mike Higgins and
V. Ugur G{\"{u}}ney and
Volkan Cirik and
Kyunghyun Cho},
title = {SearchQA: {A} New Q{\&}A Dataset Augmented with Context from a
Search Engine},
journal = {CoRR},
volume = {abs/1704.05179},
year = {2017},
url = {http://arxiv.org/abs/1704.05179},
archivePrefix = {arXiv},
eprint = {1704.05179},
timestamp = {Mon, 13 Aug 2018 16:47:09 +0200},
biburl = {https://dblp.org/rec/journals/corr/DunnSHGCC17.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
"""
# pylint: disable=line-too-long
_DESCRIPTION = """
We publicly release a new large-scale dataset, called SearchQA, for machine comprehension, or question-answering. Unlike recently released datasets, such as DeepMind
CNN/DailyMail and SQuAD, the proposed SearchQA was constructed to reflect a full pipeline of general question-answering. That is, we start not from an existing article
and generate a question-answer pair, but start from an existing question-answer pair, crawled from J! Archive, and augment it with text snippets retrieved by Google.
Following this approach, we built SearchQA, which consists of more than 140k question-answer pairs with each pair having 49.6 snippets on average. Each question-answer-context
tuple of the SearchQA comes with additional meta-data such as the snippet's URL, which we believe will be valuable resources for future research. We conduct human evaluation
as well as test two baseline methods, one simple word selection and the other deep learning based, on the SearchQA. We show that there is a meaningful gap between the human
and machine performances. This suggests that the proposed dataset could well serve as a benchmark for question-answering.
"""
_DL_URLS = {
"raw_jeopardy": [
"data/raw_jeopardy/000000-029999.zip",
"data/raw_jeopardy/030000-49999.zip",
"data/raw_jeopardy/050000-059999.zip",
"data/raw_jeopardy/060000-089999.zip",
"data/raw_jeopardy/090000-119999.zip",
"data/raw_jeopardy/120000-149999.zip",
"data/raw_jeopardy/150000-179999.zip",
"data/raw_jeopardy/180000-216929.zip",
],
"train_test_val": {
"train": "data/train_test_val/train.zip",
"test": "data/train_test_val/test.zip",
"validation": "data/train_test_val/val.zip",
},
}
# pylint: enable=line-too-long
class SearchQaConfig(datasets.BuilderConfig):
"""BuilderConfig for SearchQA."""
def __init__(self, data_url, **kwargs):
"""BuilderConfig for SearchQA
Args:
**kwargs: keyword arguments forwarded to super.
"""
super(SearchQaConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
self.data_url = data_url
class SearchQa(datasets.GeneratorBasedBuilder):
"""Search QA Dataset."""
BUILDER_CONFIGS = [SearchQaConfig(name=name, description="", data_url=_DL_URLS[name]) for name in _DL_URLS.keys()]
def _info(self):
return datasets.DatasetInfo(
description=_DESCRIPTION + "\n" + self.config.description,
features=datasets.Features(
{
"category": datasets.Value("string"),
"air_date": datasets.Value("string"),
"question": datasets.Value("string"),
"value": datasets.Value("string"),
"answer": datasets.Value("string"),
"round": datasets.Value("string"),
"show_number": datasets.Value("int32"),
"search_results": datasets.features.Sequence(
{
"urls": datasets.Value("string"),
"snippets": datasets.Value("string"),
"titles": datasets.Value("string"),
"related_links": datasets.Value("string"),
}
)
# These are the features of your dataset like images, labels ...
}
),
homepage="https://github.com/nyu-dl/dl4ir-searchQA",
citation=_CITATION,
)
def _split_generators(self, dl_manager):
"""Returns SplitGenerators."""
data_dirs = dl_manager.download_and_extract(_DL_URLS[self.config.name])
if self.config.name == "raw_jeopardy":
filepaths = itertools.chain.from_iterable(dl_manager.iter_files(data_dir) for data_dir in data_dirs)
return [
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": filepaths}),
]
elif self.config.name == "train_test_val":
return [
datasets.SplitGenerator(name=split, gen_kwargs={"filepaths": dl_manager.iter_files(data_dirs[split])})
for split in (datasets.Split.TRAIN, datasets.Split.TEST, datasets.Split.VALIDATION)
]
def _generate_examples(self, filepaths):
"""Yields examples."""
for i, filepath in enumerate(filepaths):
with open(filepath, encoding="utf-8") as f:
data = json.load(f)
category = data["category"]
air_date = data["air_date"]
question = data["question"]
value = data["value"]
answer = data["answer"]
round_ = data["round"]
show_number = int(data["show_number"])
search_results = data["search_results"]
urls = [result["url"] for result in search_results]
snippets = [result["snippet"] for result in search_results]
titles = [result["title"] for result in search_results]
related_links = [
result["related_links"] if result["related_links"] else "" for result in search_results
]
yield i, {
"category": category,
"air_date": air_date,
"question": question,
"value": value,
"answer": answer,
"round": round_,
"category": category,
"show_number": show_number,
"search_results": {
"urls": urls,
"snippets": snippets,
"titles": titles,
"related_links": related_links,
},
}