File size: 14,711 Bytes
ff4cf6a 8b6b3f1 ff4cf6a 8b6b3f1 ff4cf6a 8b6b3f1 ff4cf6a 9100988 8b6b3f1 ff4cf6a 8b6b3f1 620fe76 8b6b3f1 ff4cf6a 8b6b3f1 620fe76 87d4f1f 8b6b3f1 ff4cf6a 8b6b3f1 620fe76 8b6b3f1 ff4cf6a 8b6b3f1 6fad1a6 8b6b3f1 ff4cf6a 6fad1a6 eee9e35 ff4cf6a 6fad1a6 7a380ae 6fad1a6 ff4cf6a 8b6b3f1 ff4cf6a 8b6b3f1 ff4cf6a 87c9ef8 d618589 8b6b3f1 ff4cf6a 8b6b3f1 87c9ef8 8b6b3f1 620fe76 8b6b3f1 325e028 1258ac1 325e028 8b6b3f1 325e028 8b6b3f1 ff4cf6a 8b6b3f1 ff4cf6a 8b6b3f1 6fad1a6 8b6b3f1 6fad1a6 8b6b3f1 620fe76 8b6b3f1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 |
"""
The Harvard USPTO Patent Dataset (HUPD) is a large-scale, well-structured, and multi-purpose corpus
of English-language patent applications filed to the United States Patent and Trademark Office (USPTO)
between 2004 and 2018. With more than 4.5 million patent documents, HUPD is two to three times larger
than comparable corpora. Unlike other NLP patent datasets, HUPD contains the inventor-submitted versions
of patent applications, not the final versions of granted patents, allowing us to study patentability at
the time of filing using NLP methods for the first time.
"""
from __future__ import absolute_import, division, print_function
import os
import datetime
import pandas as pd
import numpy as np
from pathlib import Path
try:
import ujson as json
except:
import json
import datasets
_CITATION = """\
@InProceedings{suzgun2021:hupd,
title = {The Harvard USPTO Patent Dataset},
authors={Mirac Suzgun and Suproteem Sarkar and Luke Melas-Kyriazi and Scott Kominers and Stuart Shieber},
year={2021}
}
"""
_DESCRIPTION = """
The Harvard USPTO Patent Dataset (HUPD) is a large-scale, well-structured, and multi-purpose corpus
of English-language patent applications filed to the United States Patent and Trademark Office (USPTO)
between 2004 and 2018. With more than 4.5 million patent documents, HUPD is two to three times larger
than comparable corpora. Unlike other NLP patent datasets, HUPD contains the inventor-submitted versions
of patent applications, not the final versions of granted patents, allowing us to study patentability at
the time of filing using NLP methods for the first time.
"""
RANDOM_STATE = 1729
_FEATURES = [
"patent_number",
"decision",
"title",
"abstract",
"claims",
"background",
"summary",
"description",
"cpc_label",
"ipc_label",
"filing_date",
"patent_issue_date",
"date_published",
"examiner_id"
]
def str_to_date(s):
"""A helper function to convert strings to dates"""
return datetime.datetime.strptime(s, '%Y-%m-%d')
class PatentsConfig(datasets.BuilderConfig):
"""BuilderConfig for Patents"""
def __init__(
self,
metadata_url: str,
data_url: str,
data_dir: str,
ipcr_label: str = None,
cpc_label: str = None,
train_filing_start_date: str = None,
train_filing_end_date: str = None,
val_filing_start_date: str = None,
val_filing_end_date: str = None,
query_string: str = None,
val_set_balancer=False,
uniform_split=False,
force_extract=False,
**kwargs
):
"""
If train_filing_end_date is None, then a random train-val split will be used. If it is
specified, then the specified date range will be used for the split. If train_filing_end_date
if specified and val_filing_start_date is not specifed, then val_filing_start_date defaults to
train_filing_end_date.
Args:
metadata_url: `string`, url from which to download the metadata file
data_url: `string`, url from which to download the json files
data_dir: `string`, folder (in cache) in which downloaded json files are stored
ipcr_label: International Patent Classification code
cpc_label: Cooperative Patent Classification code
train_filing_start_date: Start date for patents in train set (and val set if random split is used)
train_filing_end_date: End date for patents in train set
val_filing_start_date: Start date for patents in val set
val_filing_end_date: End date for patents in val set (and train set if random split is used)
force_extract: Extract only the relevant years if this parameter is used.
**kwargs: keyword arguments forwarded to super
"""
super().__init__(**kwargs)
self.metadata_url = metadata_url
self.data_url = data_url
self.data_dir = data_dir
self.ipcr_label = ipcr_label
self.cpc_label = cpc_label
self.train_filing_start_date = train_filing_start_date
self.train_filing_end_date = train_filing_end_date
self.val_filing_start_date = val_filing_start_date
self.val_filing_end_date = val_filing_end_date
self.query_string = query_string
self.val_set_balancer = val_set_balancer
self.uniform_split = uniform_split
self.force_extract = force_extract
class Patents(datasets.GeneratorBasedBuilder):
_DESCRIPTION
VERSION = datasets.Version("1.0.2")
# This is an example of a dataset with multiple configurations.
# If you don't want/need to define several sub-sets in your dataset,
# just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
BUILDER_CONFIG_CLASS = PatentsConfig
BUILDER_CONFIGS = [
PatentsConfig(
name="sample",
description="Patent data from January 2016, for debugging",
metadata_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/hupd_metadata_jan16_2022-02-22.feather",
data_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/data/sample-jan-2016.tar.gz",
data_dir="sample", # this will unpack to data/sample/2016
),
PatentsConfig(
name="all",
description="Patent data from all years (2004-2018)",
metadata_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/hupd_metadata_2022-02-22.feather",
data_url="https://huggingface.co/datasets/HUPD/hupd/resolve/main/data/all-years.tar",
data_dir="data", # this will unpack to data/{year}
),
]
def _info(self):
return datasets.DatasetInfo(
# This is the description that will appear on the datasets page.
description=_DESCRIPTION,
# This defines the different columns of the dataset and their types
features=datasets.Features(
{k: datasets.Value("string") for k in _FEATURES}
),
# If there's a common (input, target) tuple from the features,
# specify them here. They'll be used if as_supervised=True in
# builder.as_dataset.
supervised_keys=("claims", "decision"),
homepage="https://github.com/suzgunmirac/hupd",
citation=_CITATION,
)
def _split_generators(self, dl_manager: datasets.DownloadManager):
"""Returns SplitGenerators."""
print(f'Loading dataset with config: {self.config}')
# Download metadata
# NOTE: Metadata is stored as a Pandas DataFrame in Apache Feather format
metadata_url = self.config.metadata_url
metadata_file = dl_manager.download_and_extract(self.config.metadata_url)
print(f'Using metadata file: {metadata_file}')
# Download data
# NOTE: The extracted path contains a subfolder, data_dir. This directory holds
# a large number of json files (one json file per patent application).
download_dir = dl_manager.download_and_extract(self.config.data_url)
json_dir = os.path.join(download_dir, self.config.data_dir)
# Load metadata file
print(f'Reading metadata file: {metadata_file}')
if metadata_url.endswith('.feather'):
df = pd.read_feather(metadata_file)
elif metadata_url.endswith('.csv'):
df = pd.read_csv(metadata_file)
elif metadata_url.endswith('.tsv'):
df = pd.read_csv(metadata_file, delimiter='\t')
elif metadata_url.endswith('.pickle'):
df = pd.read_pickle(metadata_file)
else:
raise ValueError(f'Metadata file invalid: {metadata_url}')
# Filter based on ICPR / CPC label
if self.config.ipcr_label:
print(f'Filtering by IPCR label: {self.config.ipcr_label}')
df = df[df['main_ipcr_label'].str.startswith(self.config.ipcr_label)]
elif self.config.cpc_label:
print(f'Filtering by CPC label: {self.config.cpc_label}')
df = df[df['main_cpc_label'].str.startswith(self.config.cpc_label)]
# Filter metadata based on arbitrary query string
if self.config.query_string:
df = df.query(self.config.query_string)
if self.config.force_extract:
if self.config.name == 'all':
if self.config.train_filing_start_date and self.config.val_filing_end_date:
if self.config.train_filing_end_date and self.config.val_filing_start_date:
training_year_range = set(range(int(self.config.train_filing_start_date[:4]), int(self.config.train_filing_end_date[:4]) + 1))
validation_year_range = set(range(int(self.config.val_filing_start_date[:4]), int(self.config.val_filing_end_date[:4]) + 1))
full_year_range = training_year_range.union(validation_year_range)
else:
full_year_range = set(range(int(self.config.train_filing_start_date[:4]), int(self.config.val_filing_end_date[:4]) + 1))
else:
full_year_range = set(range(2004, 2019))
import tarfile
for year in full_year_range:
tar_file_path = f'{json_dir}/{year}.tar.gz'
print(f'Extracting {tar_file_path}')
# open file
tar_file = tarfile.open(tar_file_path)
# extracting file
tar_file.extractall(f'{json_dir}')
tar_file.close()
# Train-validation split (either uniform or by date)
if self.config.uniform_split:
# Assumes that training_start_data < val_end_date
if self.config.train_filing_start_date:
df = df[df['filing_date'] >= self.config.train_filing_start_date]
if self.config.val_filing_end_date:
df = df[df['filing_date'] <= self.config.val_filing_end_date]
df = df.sample(frac=1.0, random_state=RANDOM_STATE)
num_train_samples = int(len(df) * 0.85)
train_df = df.iloc[0:num_train_samples]
val_df = df.iloc[num_train_samples:-1]
else:
# Check
if not (self.config.train_filing_start_date and self.config.train_filing_end_date and
self.config.val_filing_start_date and self.config.train_filing_end_date):
raise ValueError("Please either use uniform_split or specify your exact \
training and validation split dates.")
# Does not assume that training_start_data < val_end_date
print(f'Filtering train dataset by filing start date: {self.config.train_filing_start_date}')
print(f'Filtering train dataset by filing end date: {self.config.train_filing_end_date}')
print(f'Filtering val dataset by filing start date: {self.config.val_filing_start_date}')
print(f'Filtering val dataset by filing end date: {self.config.val_filing_end_date}')
train_df = df[
(df['filing_date'] >= self.config.train_filing_start_date) &
(df['filing_date'] < self.config.train_filing_end_date)
]
val_df = df[
(df['filing_date'] >= self.config.val_filing_start_date) &
(df['filing_date'] < self.config.val_filing_end_date)
]
# TODO: We can probably make this step faster
if self.config.val_set_balancer:
rejected_df = val_df[val_df.status == 'REJECTED']
num_rejected = len(rejected_df)
accepted_df = val_df[val_df.status == 'ACCEPTED']
num_accepted = len(accepted_df)
if num_rejected < num_accepted:
accepted_df = accepted_df.sample(frac=1.0, random_state=RANDOM_STATE) # shuffle(accepted_df)
accepted_df = accepted_df[:num_rejected]
else:
rejected_df = rejected_df.sample(frac=1.0, random_state=RANDOM_STATE) # shuffle(rejected_df)
rejected_df = rejected_df[:num_accepted]
val_df = pd.concat([rejected_df, accepted_df])
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
gen_kwargs=dict( # these kwargs are passed to _generate_examples
df=train_df,
json_dir=json_dir,
split='train',
),
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
gen_kwargs=dict(
df=val_df,
json_dir=json_dir,
split='val',
),
),
]
def _generate_examples(self, df, json_dir, split):
""" Yields examples by loading JSON files containing patent applications. """
# NOTE: df.itertuples() is way faster than df.iterrows()
for id_, x in enumerate(df.itertuples()):
# JSON files are named by application number (unique)
application_year = str(x.filing_date.year)
application_number = x.application_number
filepath = os.path.join(json_dir, application_year, application_number + '.json')
try:
with open(filepath, 'r') as f:
patent = json.load(f)
except Exception as e:
print('------------')
print(f'ERROR WITH {filepath}\n')
print(repr(e))
print()
yield id_, {k: "error" for k in _FEATURES}
# Most up-to-date-decision in meta dataframe
decision = x.decision
yield id_, {
"patent_number": application_number,
"decision": patent["decision"], # decision,
"title": patent["title"],
"abstract": patent["abstract"],
"claims": patent["claims"],
"description": patent["full_description"],
"background": patent["background"],
"summary": patent["summary"],
"cpc_label": patent["main_cpc_label"],
'filing_date': patent['filing_date'],
'patent_issue_date': patent['patent_issue_date'],
'date_published': patent['date_published'],
'examiner_id': patent['examiner_id'],
"ipc_label": patent["main_ipcr_label"],
}
|