def partition_CAR(args):
placeholder_out = '' # REPLACE WITH ARGS
cbor_file = 'paragraphCorpus/dedup.articles-paragraphs.cbor'
# Will fill files until they have 2e9 Bytes
i = 0 # Starting index for files
num_bytes = 3e9 # Just for initialization
size_threshold = 2e9 # Number of bytes to signal creation of new output .tsv file
for para in read_data.iter_paragraphs(
open(args.ctx_files_dir + cbor_file, 'rb')):
if num_bytes >= size_threshold:
i += 1 # Will start at 1
print(i)
print(num_bytes)
if i >= 2:
out_file.close(
) # Close the file that surpassed size_threshold
else:
pass
out_file = open(args.ctx_files_dir +
'CAR_collection_{}.tsv'.format(i),
'wt',
encoding='utf-8') # Needed to add encoding
tsv_writer = csv.writer(out_file, delimiter='\t')
# Write to file
tsv_writer.writerow(
['CAR' + '_' + para.para_id,
' '.join(para.get_text().split())]) # ["CAR_PID", "passage"]
num_bytes = os.path.getsize(args.ctx_files_dir +
'CAR_collection_{}.tsv'.format(i))
return None
def iterate_paragraphs(
self,
paragraph_cbor_file,
func: Callable[[Paragraph, List[Union[ParaLink, ParaText]]], Any],
max_paras: Optional[int] = None) -> List[Tuple[str, Any]]:
"""
:param paragraph_cbor_file: Location of the paragraphCorpus.cbor file
"""
processed_paragraphs = 0
unique_paragraphs_seen = 0
total = len(self.paragraphs_to_consider)
result = []
with open(paragraph_cbor_file, 'rb') as f:
for p in iter_paragraphs(f):
processed_paragraphs += 1
if processed_paragraphs % 100000 == 0:
print("(Searching paragraph cbor): {}".format(
processed_paragraphs))
if max_paras and processed_paragraphs >= max_paras:
break
if p.para_id in self.paragraphs_to_consider:
for para in self.paragraphs_to_consider[p.para_id]:
result.append((p.para_id, func(para, p.bodies)))
unique_paragraphs_seen += 1
if unique_paragraphs_seen == total:
break
return result
def get_paragraphs(paragraphs_file):
with open(paragraphs_file, 'rb') as f:
for p in iter_paragraphs(f):
texts = [
elem.text if isinstance(elem, ParaText) else elem.anchor_text
for elem in p.bodies
]
yield p.para_id + '|__|' + (' '.join(texts))
def get_mapping(file_path: str, aspect_to_entity_dict: Dict[str, str]):
with open(file_path, 'rb') as cbor:
for para in tqdm.tqdm(read_data.iter_paragraphs(cbor), total=10000):
for body in para.bodies:
if isinstance(
body,
read_data.ParaLink) and body.link_section is not None:
aspect_to_entity_dict[body.link_section] = body.pageid
def parse(self, f):
for paragraph in iter_paragraphs(f):
for body in paragraph.bodies:
if isinstance(body, ParaLink):
# print(body)
if body.link_section is not None:
print("Link Section: {}".format(body.link_section))
print("Link Text: {}".format(body.get_text()))
print("Link Anchor: {}".format(body.anchor_text))
print("Link Page: {}".format(body.page))
def create_database(corpus: str, save: str):
with open(corpus, 'rb') as cbor:
id_to_name_dict: Dict[str, str] = dict(
(body.pageid, body.page)
for para in tqdm.tqdm(read_data.iter_paragraphs(cbor), total=total)
for body in para.bodies if isinstance(body, read_data.ParaLink))
write_to_file(id_to_name_dict, save)
print('File written to: {}'.format(save))
def quick_ids(self):
counter = 0
myset = set()
with open(self.cbor_loc, 'rb') as f:
for paragraph in iter_paragraphs(f):
counter += 1
if counter > 10:
break
myset.add(paragraph.para_id)
return myset
def explore(self):
progress = 0
offset = 0
with open(self.cbor_loc, 'rb') as f:
for paragraph in iter_paragraphs(f):
wee = self.get_entities(paragraph)
self.extract_from_text(paragraph, wee)
progress += 1
if progress % 10 == 0:
break
def print_paragraphs(path=WikiParagrahs.file_path_list[0], limit=1):
print('*** reading {} paragraph from file: {} ***'.format(limit, path))
with open(path, 'rb') as f:
counter = 1
for p in iter_paragraphs(f):
print()
print('*** PRINTING PARAGRAPH {} ***'.format(counter))
print(
'----------------------- PARAGRAPH ID -----------------------'
)
print(p.para_id)
print(
'----------------------- RAW PARAGRAPH -----------------------'
)
print(p)
# Print just the text
texts = [
elem.text if isinstance(elem, ParaText) else elem.anchor_text
for elem in p.bodies
]
print(
'----------------------- TEXT -----------------------------')
print(' '.join(texts))
print(
'----------------------- ENTITIES -----------------------------'
)
entities = [
elem.page for elem in p.bodies if isinstance(elem, ParaLink)
]
print(entities)
print(
'----------------------- MIXED -----------------------------')
mixed = [(elem.anchor_text,
elem.page) if isinstance(elem, ParaLink) else
(elem.text, None) for elem in p.bodies]
print(mixed)
if counter >= limit:
break
counter += 1
def process_paragraphs(self):
if not self.freq_dict:
para_dict = {}
raw_data = {}
para_text = {}
with open(self.paragraph_file, 'rb') as f:
for p in iter_paragraphs(f):
# entities = [elem.page
# for elem in p.bodies
# if isinstance(elem, ParaLink)]# how to retrieve entities from paragraph; p@5 get a bit higher
# para_dict[p.para_id] = self.preprocess_text(p.get_text(), ret="freq")
raw_data[p.para_id] = self.preprocess_text(p.get_text(),
ret="raw")
para_text[p.para_id] = p.get_text()
self.freq_dict = para_dict
self.raw_data = raw_data
self.para_text = para_text
def retrieve_paragraph_mappings(self, cbor_loc):
"""
:param cbor_loc: Location of the paragraphCorpus.cbor file
"""
counter = 0
seen = 0
total = len(self.paragraphs_to_retrieve)
with open(cbor_loc, 'rb') as f:
for p in iter_paragraphs(f):
counter += 1
if counter % 100000 == 0:
print("(Searching paragraph cbor): {}".format(counter))
if p.para_id in self.paragraphs_to_retrieve:
for p_to_be_updated in self.paragraphs_to_retrieve[
p.para_id]:
self.update_paragraph(p_to_be_updated, p.bodies)
seen += 1
if seen == total:
break
def retrieve_text_matching_ids(self, ids: Set[str]):
jsons = OrderedDict() # type: OrderedDict[str, str]
out = open(self.json_dump_name + ".jsonl", 'w')
counter = 0
with open(self.cbor_loc, 'rb') as f:
for paragraph in iter_paragraphs(f):
counter += 1
if paragraph.para_id in ids:
jsons[paragraph.para_id] = self.create_json(paragraph)
# stop once we've retrieved all of the paragraphs
ids.remove(paragraph.para_id)
if not ids:
break
for _, json in jsons.items():
out.write(json + "\n")
out.close()
def build_d(read_path=WikiParagrahs.file_path_list[0],
write_path=write_d_path,
paragraph_limit=1):
print('*** reading {} paragraph from file: {} ***'.format(
paragraph_limit, read_path))
with open(read_path, 'rb') as f_read:
with open(write_path, 'w') as f_write:
counter = 1
for p in iter_paragraphs(f_read):
if counter % 10000 == 0:
print('{} / {} of paragraphs processed'.format(
counter, paragraph_limit))
f_write.write(p.para_id + '\t' + p.get_text() + '\n')
if counter >= paragraph_limit:
break
counter += 1
def dump_cbor(self):
out = open(self.json_dump_name + ".jsonl", 'w')
# index_dir = self.json_dump_name + "_index"
pmap_out = open(self.json_dump_name + "_pmap.txt", 'w')
# os.mkdir(index_dir)
progress = 0
offset = 0
with open(self.cbor_loc, 'rb') as f:
for paragraph in iter_paragraphs(f):
to_json = self.create_json(paragraph) + "\n"
out.write(to_json)
pmap_out.write("{} {} {}\n".format(paragraph.para_id, offset,
offset + len(to_json) - 1))
offset += len(to_json)
progress += 1
if progress % 10000 == 0:
print(progress)
out.close()
pmap_out.close()
def create_paragraph_map(self, paragraph_path):
for i, page in enumerate(
read_data.iter_paragraphs(open(paragraph_path, 'rb'))):
self.paragraph_map[page.para_id] = page.get_text()
def create_para_id_list(paragraph_cbor_file: str) -> List[str]:
with open(paragraph_cbor_file, 'rb') as f:
return [p.para_id for p in iter_paragraphs(f)]
