def init(self, force=False):
base_path = util.path_dataset(self)
idxs = [self.index, self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
train_qrels = os.path.join(base_path, 'train.qrels.txt')
valid_qrels = os.path.join(base_path, 'valid.qrels.txt')
test_qrels = os.path.join(base_path, 'test.qrels.txt')
if (force or not os.path.exists(train_qrels)
or not os.path.exists(valid_qrels)) and self._confirm_dua():
source_stream = util.download_stream(**_FILES['qrels_2013'],
encoding='utf8')
source_stream2 = util.download_stream(**_FILES['qrels_2014'],
encoding='utf8')
with util.finialized_file(train_qrels, 'wt') as tf, \
util.finialized_file(valid_qrels, 'wt') as vf, \
util.finialized_file(test_qrels, 'wt') as Tf:
for line in source_stream:
cols = line.strip().split()
if int(cols[0]) in VALIDATION_QIDS:
vf.write(' '.join(cols) + '\n')
elif int(cols[0]) in TEST_QIDS:
Tf.write(' '.join(cols) + '\n')
else:
tf.write(' '.join(cols) + '\n')
for line in source_stream2:
cols = line.strip().split()
if cols[0] in VALIDATION_QIDS:
vf.write(' '.join(cols) + '\n')
elif int(cols[0]) in TEST_QIDS:
Tf.write(' '.join(cols) + '\n')
else:
tf.write(' '.join(cols) + '\n')
all_queries = os.path.join(base_path, 'topics.txt')
if (force or not os.path.exists(all_queries)) and self._confirm_dua():
source_stream = util.download_stream(**_FILES['queries_2013'],
encoding='utf8')
source_stream2 = util.download_stream(**_FILES['queries_2014'],
encoding='utf8')
train, valid = [], []
for _id, _query in trec.parse_query_mbformat(source_stream):
nid = _id.replace('MB', '').strip()
train.append([nid, _query])
for _id, _query in trec.parse_query_mbformat(source_stream2):
nid = _id.replace('MB', '').strip()
train.append([nid, _query])
plaintext.write_tsv(all_queries, train)
def init(self, force=False):
idxs = [self.index, self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
train_qrels = os.path.join(util.path_dataset(self), 'train.qrels.txt')
valid_qrels = os.path.join(util.path_dataset(self), 'valid.qrels.txt')
if (force or not os.path.exists(train_qrels)
or not os.path.exists(valid_qrels)) and self._confirm_dua():
source_stream = util.download_stream(
'https://ciir.cs.umass.edu/downloads/Antique/antique-train.qrel',
encoding='utf8')
with util.finialized_file(train_qrels, 'wt') as tf, \
util.finialized_file(valid_qrels, 'wt') as vf:
for line in source_stream:
cols = line.strip().split()
if cols[0] in VALIDATION_QIDS:
vf.write(' '.join(cols) + '\n')
else:
tf.write(' '.join(cols) + '\n')
train_queries = os.path.join(util.path_dataset(self),
'train.queries.txt')
valid_queries = os.path.join(util.path_dataset(self),
'valid.queries.txt')
if (force or not os.path.exists(train_queries)
or not os.path.exists(valid_queries)) and self._confirm_dua():
source_stream = util.download_stream(
'https://ciir.cs.umass.edu/downloads/Antique/antique-train-queries.txt',
encoding='utf8')
train, valid = [], []
for cols in plaintext.read_tsv(source_stream):
if cols[0] in VALIDATION_QIDS:
valid.append(cols)
else:
train.append(cols)
plaintext.write_tsv(train_queries, train)
plaintext.write_tsv(valid_queries, valid)
test_qrels = os.path.join(util.path_dataset(self), 'test.qrels.txt')
if (force or not os.path.exists(test_qrels)) and self._confirm_dua():
util.download(
'https://ciir.cs.umass.edu/downloads/Antique/antique-test.qrel',
test_qrels)
test_queries = os.path.join(util.path_dataset(self),
'test.queries.txt')
if (force or not os.path.exists(test_queries)) and self._confirm_dua():
util.download(
'https://ciir.cs.umass.edu/downloads/Antique/antique-test-queries.txt',
test_queries)
def init(self, force=False):
base_path = util.path_dataset(self)
idxs = [self.index, self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
qrels_file = os.path.join(base_path, 'qrels.robust2004.txt')
if (force or not os.path.exists(qrels_file)) and self._confirm_dua():
util.download(**_FILES['qrels'], file_name=qrels_file)
for fold in FOLDS:
fold_qrels_file = os.path.join(base_path, f'{fold}.qrels')
if (force or not os.path.exists(fold_qrels_file)):
all_qrels = trec.read_qrels_dict(qrels_file)
fold_qrels = {
qid: dids
for qid, dids in all_qrels.items() if qid in FOLDS[fold]
}
trec.write_qrels_dict(fold_qrels_file, fold_qrels)
query_file = os.path.join(base_path, 'topics.txt')
if (force or not os.path.exists(query_file)) and self._confirm_dua():
query_file_stream = util.download_stream(**_FILES['queries'],
encoding='utf8')
with util.finialized_file(query_file, 'wt') as f:
plaintext.write_tsv(f,
trec.parse_query_format(query_file_stream))
def _init_qrels(self, subset, qrels_files, force=False, expected_md5=None):
qrelsf = os.path.join(util.path_dataset(self), f'{subset}.qrels')
if (force or not os.path.exists(qrelsf)) and self._confirm_dua():
qrels = itertools.chain(*(trec.read_qrels(
util.download_stream(f, 'utf8', expected_md5=expected_md5))
for f in qrels_files))
trec.write_qrels(qrelsf, qrels)
def _init_topics(self,
subset,
topic_files,
qid_prefix=None,
encoding=None,
xml_prefix=None,
force=False,
expected_md5=None):
topicf = os.path.join(util.path_dataset(self), f'{subset}.topics')
if (force or not os.path.exists(topicf)) and self._confirm_dua():
topics = []
for topic_file in topic_files:
topic_file_stream = util.download_stream(
topic_file, encoding, expected_md5=expected_md5)
for t, qid, text in trec.parse_query_format(
topic_file_stream, xml_prefix):
if qid_prefix is not None:
qid = qid.replace(qid_prefix, '')
topics.append((t, qid, text))
plaintext.write_tsv(topicf, topics)
def _init_iter_collection(self):
files = {
'2020-04-10': {
'comm_use_subset':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-04-10/comm_use_subset.tar.gz',
"253cecb4fee2582a611fb77a4d537dc5"),
'noncomm_use_subset':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-04-10/noncomm_use_subset.tar.gz',
"734b462133b3c00da578a909f945f4ae"),
'custom_license':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-04-10/custom_license.tar.gz',
"2f1c9864348025987523b86d6236c40b"),
'biorxiv_medrxiv':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-04-10/biorxiv_medrxiv.tar.gz',
"c12acdec8b3ad31918d752ba3db36121"),
},
'2020-05-01': {
'comm_use_subset':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/comm_use_subset.tar.gz',
"af4202340182209881d3d8cba2d58a24"),
'noncomm_use_subset':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/noncomm_use_subset.tar.gz',
"9cc25b9e8674197446e7cbd4381f643b"),
'custom_license':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/custom_license.tar.gz',
"1cb6936a7300a31344cd8a5ecc9ca778"),
'biorxiv_medrxiv':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/biorxiv_medrxiv.tar.gz',
"9d6c6dc5d64b01e528086f6652b3ccb7"),
'arxiv':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/arxiv.tar.gz',
"f10890174d6f864f306800d4b02233bc"),
}
}
metadata = {
'2020-04-10':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-04-10/metadata.csv',
"42a21f386be86c24647a41bedde34046"),
'2020-05-01':
('https://ai2-semanticscholar-cord-19.s3-us-west-2.amazonaws.com/2020-05-01/metadata.csv',
"b1d2e409026494e0c8034278bacd1248"),
}
meta_url, meta_md5 = metadata[self.config['date']]
fulltexts = {}
with contextlib.ExitStack() as stack:
for fid, (file, md5) in files[self.config['date']].items():
fulltexts[fid] = stack.enter_context(
util.download_tmp(file, tarf=True, expected_md5=md5))
meta = pd.read_csv(
util.download_stream(meta_url, expected_md5=meta_md5))
for _, row in meta.iterrows():
did = str(row['cord_uid'])
title = str(row['title'])
doi = str(row['doi'])
abstract = str(row['abstract'])
date = str(row['publish_time'])
body = ''
heads = ''
if row['has_pmc_xml_parse']:
path = os.path.join(row['full_text_file'], 'pmc_json',
row['pmcid'] + '.xml.json')
data = json.load(
fulltexts[row['full_text_file']].extractfile(path))
if 'body_text' in data:
body = '\n'.join(b['text'] for b in data['body_text'])
heads = '\n'.join(
set(b['section'] for b in data['body_text']))
elif row['has_pdf_parse']:
path = os.path.join(
row['full_text_file'], 'pdf_json',
row['sha'].split(';')[0].strip() + '.json')
data = json.load(
fulltexts[row['full_text_file']].extractfile(path))
if 'body_text' in data:
body = '\n'.join(b['text'] for b in data['body_text'])
heads = '\n'.join(
set(b['section'] for b in data['body_text']))
contents = f'{title}\n\n{abstract}\n\n{body}\n\n{heads}'
doc = indices.RawDoc(did,
text=contents,
title=title,
abstract=abstract,
title_abs=f'{title}\n\n{abstract}',
body=body,
doi=doi,
date=date)
yield doc
def init(self, force=False):
idxs = [self.index_stem, self.doc_store]
self._init_indices_parallel(idxs, self._init_iter_collection(), force)
base_path = util.path_dataset(self)
needs_queries = []
if force or not os.path.exists(
os.path.join(base_path, 'train.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'train.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.train.tsv' and qid not in MINI_DEV)))
if force or not os.path.exists(
os.path.join(base_path, 'minidev.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'minidev.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.train.tsv' and qid in MINI_DEV)))
if force or not os.path.exists(
os.path.join(base_path, 'dev.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'dev.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.dev.tsv')))
if force or not os.path.exists(
os.path.join(base_path, 'eval.queries.tsv')):
needs_queries.append(lambda it: plaintext.write_tsv(
os.path.join(base_path, 'eval.queries.tsv'),
((qid, txt) for file, qid, txt in it
if file == 'queries.eval.tsv')))
if needs_queries and self._confirm_dua():
with util.download_tmp(_SOURCES['queries']) as f, \
tarfile.open(fileobj=f) as tarf, \
contextlib.ExitStack() as ctxt:
def _extr_subf(subf):
for qid, txt in plaintext.read_tsv(
io.TextIOWrapper(tarf.extractfile(subf))):
yield subf, qid, txt
query_iter = [
_extr_subf('queries.train.tsv'),
_extr_subf('queries.dev.tsv'),
_extr_subf('queries.eval.tsv')
]
query_iter = tqdm(itertools.chain(*query_iter), desc='queries')
query_iters = util.blocking_tee(query_iter, len(needs_queries))
for fn, it in zip(needs_queries, query_iters):
ctxt.enter_context(
util.CtxtThread(functools.partial(fn, it)))
file = os.path.join(base_path, 'train.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['train-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
if qid not in MINI_DEV:
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'minidev.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['train-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
if qid in MINI_DEV:
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'dev.qrels')
if (force or not os.path.exists(file)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['dev-qrels'], 'utf8')
with util.finialized_file(file, 'wt') as out:
for qid, _, did, score in plaintext.read_tsv(stream):
trec.write_qrels(out, [(qid, did, score)])
file = os.path.join(base_path, 'train.mspairs.gz')
if not os.path.exists(file) and os.path.exists(
os.path.join(base_path, 'qidpidtriples.train.full')):
# legacy
os.rename(os.path.join(base_path, 'qidpidtriples.train.full'),
file)
if (force or not os.path.exists(file)) and self._confirm_dua():
util.download(_SOURCES['qidpidtriples.train.full'], file)
if not self.config['init_skip_msrun']:
for file_name, subf in [('dev.msrun', 'top1000.dev'),
('eval.msrun', 'top1000.eval'),
('train.msrun', 'top1000.train.txt')]:
file = os.path.join(base_path, file_name)
if (force or not os.path.exists(file)) and self._confirm_dua():
run = {}
with util.download_tmp(_SOURCES[file_name]) as f, \
tarfile.open(fileobj=f) as tarf:
for qid, did, _, _ in tqdm(
plaintext.read_tsv(
io.TextIOWrapper(tarf.extractfile(subf)))):
if qid not in run:
run[qid] = {}
run[qid][did] = 0.
if file_name == 'train.msrun':
minidev = {
qid: dids
for qid, dids in run.items() if qid in MINI_DEV
}
with self.logger.duration('writing minidev.msrun'):
trec.write_run_dict(
os.path.join(base_path, 'minidev.msrun'),
minidev)
run = {
qid: dids
for qid, dids in run.items() if qid not in MINI_DEV
}
with self.logger.duration(f'writing {file_name}'):
trec.write_run_dict(file, run)
query_path = os.path.join(base_path, 'trec2019.queries.tsv')
if (force or not os.path.exists(query_path)) and self._confirm_dua():
stream = util.download_stream(_SOURCES['trec2019.queries'], 'utf8')
plaintext.write_tsv(query_path, plaintext.read_tsv(stream))
msrun_path = os.path.join(base_path, 'trec2019.msrun')
if (force or not os.path.exists(msrun_path)) and self._confirm_dua():
run = {}
with util.download_stream(_SOURCES['trec2019.msrun'],
'utf8') as stream:
for qid, did, _, _ in plaintext.read_tsv(stream):
if qid not in run:
run[qid] = {}
run[qid][did] = 0.
with util.finialized_file(msrun_path, 'wt') as f:
trec.write_run_dict(f, run)
qrels_path = os.path.join(base_path, 'trec2019.qrels')
if not os.path.exists(qrels_path) and self._confirm_dua():
util.download(_SOURCES['trec2019.qrels'], qrels_path)
qrels_path = os.path.join(base_path, 'judgedtrec2019.qrels')
if not os.path.exists(qrels_path):
os.symlink('trec2019.qrels', qrels_path)
query_path = os.path.join(base_path, 'judgedtrec2019.queries.tsv')
judged_qids = util.Lazy(
lambda: trec.read_qrels_dict(qrels_path).keys())
if (force or not os.path.exists(query_path)):
with util.finialized_file(query_path, 'wt') as f:
for qid, qtext in plaintext.read_tsv(
os.path.join(base_path, 'trec2019.queries.tsv')):
if qid in judged_qids():
plaintext.write_tsv(f, [(qid, qtext)])
msrun_path = os.path.join(base_path, 'judgedtrec2019.msrun')
if (force or not os.path.exists(msrun_path)) and self._confirm_dua():
with util.finialized_file(msrun_path, 'wt') as f:
for qid, dids in trec.read_run_dict(
os.path.join(base_path, 'trec2019.msrun')).items():
if qid in judged_qids():
trec.write_run_dict(f, {qid: dids})
# A subset of dev that only contains queries that have relevance judgments
judgeddev_path = os.path.join(base_path, 'judgeddev')
judged_qids = util.Lazy(lambda: trec.read_qrels_dict(
os.path.join(base_path, 'dev.qrels')).keys())
if not os.path.exists(f'{judgeddev_path}.qrels'):
os.symlink('dev.qrels', f'{judgeddev_path}.qrels')
if not os.path.exists(f'{judgeddev_path}.queries.tsv'):
with util.finialized_file(f'{judgeddev_path}.queries.tsv',
'wt') as f:
for qid, qtext in plaintext.read_tsv(
os.path.join(base_path, 'dev.queries.tsv')):
if qid in judged_qids():
plaintext.write_tsv(f, [(qid, qtext)])
if self.config['init_skip_msrun']:
if not os.path.exists(f'{judgeddev_path}.msrun'):
with util.finialized_file(f'{judgeddev_path}.msrun',
'wt') as f:
for qid, dids in trec.read_run_dict(
os.path.join(base_path, 'dev.msrun')).items():
if qid in judged_qids():
trec.write_run_dict(f, {qid: dids})
if not self.config['init_skip_train10']:
file = os.path.join(base_path, 'train10.queries.tsv')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout:
for qid, qtext in self.logger.pbar(
plaintext.read_tsv(
os.path.join(base_path, 'train.queries.tsv')),
desc='filtering queries for train10'):
if int(qid) % 10 == 0:
plaintext.write_tsv(fout, [(qid, qtext)])
file = os.path.join(base_path, 'train10.qrels')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.qrels'), 'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering qrels for train10'):
qid = line.split()[0]
if int(qid) % 10 == 0:
fout.write(line)
if not self.config['init_skip_msrun']:
file = os.path.join(base_path, 'train10.msrun')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.msrun'),
'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering msrun for train10'):
qid = line.split()[0]
if int(qid) % 10 == 0:
fout.write(line)
file = os.path.join(base_path, 'train10.mspairs.gz')
if not os.path.exists(file):
with gzip.open(file, 'wt') as fout, gzip.open(
os.path.join(base_path, 'train.mspairs.gz'),
'rt') as fin:
for qid, did1, did2 in self.logger.pbar(
plaintext.read_tsv(fin),
desc='filtering mspairs for train10'):
if int(qid) % 10 == 0:
plaintext.write_tsv(fout, [(qid, did1, did2)])
if not self.config['init_skip_train_med']:
med_qids = util.Lazy(
lambda: {
qid.strip()
for qid in util.download_stream(
'https://raw.githubusercontent.com/Georgetown-IR-Lab/covid-neural-ir/master/med-msmarco-train.txt',
'utf8',
expected_md5="dc5199de7d4a872c361f89f08b1163ef")
})
file = os.path.join(base_path, 'train_med.queries.tsv')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout:
for qid, qtext in self.logger.pbar(
plaintext.read_tsv(
os.path.join(base_path, 'train.queries.tsv')),
desc='filtering queries for train_med'):
if qid in med_qids():
plaintext.write_tsv(fout, [(qid, qtext)])
file = os.path.join(base_path, 'train_med.qrels')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.qrels'), 'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering qrels for train_med'):
qid = line.split()[0]
if qid in med_qids():
fout.write(line)
if not self.config['init_skip_msrun']:
file = os.path.join(base_path, 'train_med.msrun')
if not os.path.exists(file):
with util.finialized_file(file, 'wt') as fout, open(
os.path.join(base_path, 'train.msrun'),
'rt') as fin:
for line in self.logger.pbar(
fin, desc='filtering msrun for train_med'):
qid = line.split()[0]
if qid in med_qids():
fout.write(line)
file = os.path.join(base_path, 'train_med.mspairs.gz')
if not os.path.exists(file):
with gzip.open(file, 'wt') as fout, gzip.open(
os.path.join(base_path, 'train.mspairs.gz'),
'rt') as fin:
for qid, did1, did2 in self.logger.pbar(
plaintext.read_tsv(fin),
desc='filtering mspairs for train_med'):
if qid in med_qids():
plaintext.write_tsv(fout, [(qid, did1, did2)])
def _init_iter_collection(self):
strm = util.download_stream(
'https://ciir.cs.umass.edu/downloads/Antique/antique-collection.txt',
'utf8')
for did, text in plaintext.read_tsv(strm):
yield indices.RawDoc(did, text)