blbooks / blbooks.py
davanstrien's picture
davanstrien HF staff
Add British Library books dataset (#3603)
3c0072d
raw
history blame
8.95 kB
# coding=utf-8
# Copyright 2022 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.
import gzip
import json
from datetime import datetime
from functools import lru_cache
from typing import Dict, List
import datasets
from datasets.tasks import LanguageModeling
_CITATION = """\
@misc{BritishLibraryBooks2021,
author = {British Library Labs},
title = {Digitised Books. c. 1510 - c. 1900. JSONL (OCR derived text + metadata)},
year = {2021},
publisher = {British Library},
howpublished={https://doi.org/10.23636/r7w6-zy15}
"""
_DESCRIPTION = """\
A dataset comprising of text created by OCR from the 49,455 digitised books, equating to 65,227 volumes (25+ million pages), published between c. 1510 - c. 1900.
The books cover a wide range of subject areas including philosophy, history, poetry and literature.
"""
_BASE_URL = "https://bl.iro.bl.uk/downloads/"
_DATA_URLS = {
"1510_1699": _BASE_URL + "61f58234-b370-422f-8591-8f98e46c2757?locale=en",
"1700_1799": _BASE_URL + "78b4a8ec-395e-4383-831c-809faff85ad7?locale=en",
"1800_1809": _BASE_URL + "91ae15cb-e08f-4abf-8396-e4742d9d4e37?locale=en",
"1810_1819": _BASE_URL + "6d1a6e17-f28d-45b9-8f7a-a03cf3a96491?locale=en",
"1820_1829": _BASE_URL + "ec764dbd-1ed4-4fc2-8668-b4df5c8ec451?locale=en",
"1830_1839": _BASE_URL + "eab68022-0418-4df7-a401-78972514ed20?locale=en",
"1840_1849": _BASE_URL + "d16d88b0-aa3f-4dfe-b728-c58d168d7b4d?locale=en",
"1850_1859": _BASE_URL + "a6a44ea8-8d33-4880-8b17-f89c90e3d89a?locale=en",
"1860_1869": _BASE_URL + "2e17f00f-52e6-4259-962c-b88ad60dec23?locale=en",
"1870_1879": _BASE_URL + "899c3719-030c-4517-abd3-b28fdc85eed4?locale=en",
"1880_1889": _BASE_URL + "ec3b8545-775b-47bd-885d-ce895263709e?locale=en",
"1890_1899": _BASE_URL + "54ed2842-089a-439a-b751-2179b3ffba28?locale=en",
}
_ALL = list(_DATA_URLS.values())
_1800_1899 = [
_DATA_URLS.get(subset)
for subset in [
"1800_1809",
"1810_1819",
"1820_1829",
"1830_1839",
"1840_1849",
"1850_1859",
"1860_1869",
"1870_1879",
"1880_1889",
"1890_1899",
]
]
_1700_1799 = [_DATA_URLS.get(subset) for subset in ["1700_1799"]]
_1510_1699 = [_DATA_URLS.get(subset) for subset in ["1510_1699"]]
URL = "https://doi.org/10.23636/r7w6-zy15"
features = datasets.Features(
{
"record_id": datasets.Value("string"),
"date": datasets.Value("timestamp[s]"),
"raw_date": datasets.Value("string"),
"title": datasets.Value("string"),
"place": datasets.Value("string"),
"empty_pg": datasets.Value("bool"),
"text": datasets.Value("string"),
"pg": datasets.Value("int32"),
"mean_wc_ocr": datasets.Value("float32"),
"std_wc_ocr": datasets.Value("float64"),
"name": datasets.Value("string"),
"all_names": datasets.Value("string"),
"Publisher": datasets.Value("string"),
"Country of publication 1": datasets.Value("string"),
"all Countries of publication": datasets.Value("string"),
"Physical description": datasets.Value("string"),
"Language_1": datasets.Value("string"),
"Language_2": datasets.Value("string"),
"Language_3": datasets.Value("string"),
"Language_4": datasets.Value("string"),
"multi_language": datasets.Value("bool"),
}
)
class BritishLibraryBooksConfig(datasets.BuilderConfig):
"""BuilderConfig for BritishLibraryBooks."""
def __init__(self, data_urls, citation, url, skip_empty=False, **kwargs):
"""BuilderConfig for BritishLibraryBooks.
Args:
data_url: `string`, url to download the zip file from.
citation: `string`, citation for the data set.
url: `string`, url for information about the data set.
skip_empty: `bool`, whether to skip empty pages.
**kwargs: keyword arguments forwarded to super.
"""
super(BritishLibraryBooksConfig, self).__init__(version=datasets.Version("1.0.2"), **kwargs)
self.url: str = url
self.data_urls: List[str] = data_urls
self.citation: str = citation
self.skip_empty: bool = skip_empty
class BritishLibraryBooks(datasets.GeneratorBasedBuilder):
"""The BritishLibraryBooks dataset."""
BUILDER_CONFIGS = [
BritishLibraryBooksConfig(
name="1500_1899",
description="All periods of" + _DESCRIPTION,
data_urls=_ALL,
citation=_CITATION,
url=URL,
skip_empty=True,
),
BritishLibraryBooksConfig(
name="1800_1899",
description="A subset covering texts published during the 1800-1899 of" + _DESCRIPTION,
data_urls=_1800_1899,
citation=_CITATION,
url=URL,
skip_empty=True,
),
BritishLibraryBooksConfig(
name="1700_1799",
description="Subset covering 1700-1799 of" + _DESCRIPTION,
data_urls=_1700_1799,
citation=_CITATION,
url=URL,
skip_empty=True,
),
BritishLibraryBooksConfig(
name="1510_1699",
description="Subset covering 1510-1699 of " + _DESCRIPTION,
data_urls=_1510_1699,
citation=_CITATION,
url=URL,
skip_empty=True,
),
]
DEFAULT_CONFIG_NAME = "1500_1899"
def _info(self):
return datasets.DatasetInfo(
description=_DESCRIPTION,
features=features,
supervised_keys=None,
homepage="https://www.bl.uk/collection-guides/digitised-printed-books",
citation=_CITATION,
task_templates=[LanguageModeling(text_column="text")],
)
def _split_generators(self, dl_manager: datasets.DownloadManager):
urls_to_download = self.config.data_urls
downloaded_archives = dl_manager.download(urls_to_download)
downloaded_archives = [dl_manager.iter_archive(archive) for archive in downloaded_archives]
return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_dirs": downloaded_archives})]
@lru_cache(maxsize=512)
def _parse_date(self, date):
if date is not None:
date = datetime.strptime(str(date), "%Y")
return date
def _parse_data(self, data: Dict) -> Dict:
mean_wc_ocr = data["mean_wc_ocr"]
mean_wc_ocr = float(mean_wc_ocr) if mean_wc_ocr else None
std_wc_ocr = data["std_wc_ocr"]
std_wc_ocr = float(data["std_wc_ocr"]) if std_wc_ocr else None
date = data["date"]
if date is not None:
date = datetime.strptime(str(date), "%Y")
return {
"record_id": data["record_id"],
"date": date,
"raw_date": data["raw_date"],
"title": data["title"],
"place": data["place"],
"text": data["text"],
"pg": int(data["pg"]),
"mean_wc_ocr": data["mean_wc_ocr"],
"std_wc_ocr": std_wc_ocr,
"name": data["Name"],
"all_names": data["All names"],
"Publisher": data["Publisher"],
"Country of publication 1": data["Country of publication 1"],
"all Countries of publication": data["All Countries of publication"],
"Physical description": data["Physical description"],
"Language_1": data["Language_1"],
"Language_2": data["Language_2"],
"Language_3": data["Language_3"],
"Language_4": data["Language_4"],
"multi_language": data["multi_language"],
}
def _generate_examples(self, data_dirs):
skip_empty = self.config.skip_empty
id_ = 0
for data_dir in data_dirs:
for path, file in data_dir:
if not path.endswith(".gz"):
continue
with gzip.open(file) as json_l:
for row in json_l:
data = json.loads(row)
empty_pg = data["empty_pg"]
if skip_empty and empty_pg:
continue
parsed_data = self._parse_data(data)
yield id_, {**parsed_data, **{"empty_pg": empty_pg}}
id_ += 1