def build_search_index():
start = time()
try:
search_idx = db.nodes.indexes.get('search_idx')
with db.transaction:
search_idx.delete()
except ValueError:
pass
search_idx = db.nodes.indexes.create('search_idx',type='fulltext')
# Loop through papers
for batch_count, batch in enumerate(group_generator(PAPER.type.incoming, 1000)):
print "Building search index. Processing paper %d. Elapsed time %d sec," % ( batch_count*1000, time()-start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
search_idx['title'][paper_node['title']]=paper_node
for batch_count, batch in enumerate(group_generator(AUTHOR.type.incoming, 1000)):
print "Building search index. Processing paper %d. Elapsed time %d sec," % ( batch_count*1000, time()-start)
with db.transaction:
for author_rel in batch:
author_node = author_rel.startNode
name = author_node['name'].replace(',',' ')
search_idx['author'][name]=author_node
def build_search_index():
start = time()
try:
search_idx = db.nodes.indexes.get('search_idx')
with db.transaction:
search_idx.delete()
except ValueError:
pass
search_idx = db.nodes.indexes.create('search_idx', type='fulltext')
# Loop through papers
for batch_count, batch in enumerate(
group_generator(PAPER.type.incoming, 1000)):
print "Building search index. Processing paper %d. Elapsed time %d sec," % (
batch_count * 1000, time() - start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
search_idx['title'][paper_node['title']] = paper_node
for batch_count, batch in enumerate(
group_generator(AUTHOR.type.incoming, 1000)):
print "Building search index. Processing paper %d. Elapsed time %d sec," % (
batch_count * 1000, time() - start)
with db.transaction:
for author_rel in batch:
author_node = author_rel.startNode
name = author_node['name'].replace(',', ' ')
search_idx['author'][name] = author_node
def write_cite_rank(iterations=4):
start = time()
# damping factor
d = 0.85
# initial value
cr_0 = 1.
for iteration in range(iterations):
for batch_count, batch in enumerate(
group_generator(PAPER.type.incoming, 1000)):
print "Calculating CiteRank. Iteration %d. Processed %d papers in %d sec." % (
iteration, batch_count * 1000, time() - start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
cr = (1 - d)
for cite_rel in paper_node.ref.incoming:
cite_node = cite_rel.startNode
try:
cr_node = cite_node['c_cite_rank']
except KeyError:
cr_node = cr_0
cr += d * cr_node / cite_node['c_reference_count']
paper_node['c_cite_rank'] = cr
def unmatched_reference_fill_db(match_file=match_file, db=db):
"""
Findes lines in match_file, where the target cannot be found, and adds
the corresponding reference strings to 'unknown_references' list in the source.
"""
in_iter = open(match_file)
# line format examples at the end of the file
start = time()
rel_count = 0
last_id = ''
ref_buffer = []
for batch_count, batch in enumerate(group_generator(in_iter, 10000)):
sys.stderr.write(
'Processing reference %d x 10000. Elapsed time %d sec. Relations created %d. \n'
% (batch_count, time() - start, rel_count))
with db.transaction:
for line in batch:
try:
source_id, ref_string, target_id = line.rstrip().split('|')
except ValueError:
print 'Skipped line: not not enough separators "|". ', line[:
20]
continue
if not target_id == '':
# Lookup target_id in the index
target_node = None
for target_node in source_idx['id']['arxiv:' + target_id]:
break
if target_node:
# skip if target exists
last_id = source_id
continue
# target node does not exist here #
if source_id == last_id: # on same node?
ref_buffer.append(ref_string)
continue
# new source_id #
# lookup source node
for last_node in source_idx['id']['arxiv:' + last_id]:
break
else:
print "Skipped line: source id not forund. ", last_id
last_id = source_id
continue
# last node exists #
# write old references to last node
last_node['unknown_references'] += ref_buffer
last_id = source_id
ref_buffer = [ref_string]
def json_export(json_dir=json_dir, pkl_dir=pkl_dir):
if not os.path.exists(json_dir):
os.makedirs(json_dir)
for i, batch in enumerate(
group_generator(get_meta_from_pkl(pkl_dir), 10000)):
fh = open(json_dir + 'META_%03d.json' % i, 'w')
json.dump(batch, fh)
fh.close()
def unmatched_reference_fill_db(match_file = match_file , db=db):
"""
Findes lines in match_file, where the target cannot be found, and adds
the corresponding reference strings to 'unknown_references' list in the source.
"""
in_iter = open(match_file)
# line format examples at the end of the file
start = time()
rel_count = 0
last_id = ''
ref_buffer = []
for batch_count, batch in enumerate(group_generator(in_iter,10000)):
sys.stderr.write('Processing reference %d x 10000. Elapsed time %d sec. Relations created %d. \n' % (batch_count, time() - start, rel_count))
with db.transaction:
for line in batch:
try:
source_id, ref_string, target_id = line.rstrip().split('|')
except ValueError:
print 'Skipped line: not not enough separators "|". ', line[:20]
continue
if not target_id == '':
# Lookup target_id in the index
target_node = None
for target_node in source_idx['id']['arxiv:' + target_id]:
break
if target_node:
# skip if target exists
last_id = source_id
continue
# target node does not exist here #
if source_id == last_id: # on same node?
ref_buffer.append(ref_string)
continue
# new source_id #
# lookup source node
for last_node in source_idx['id']['arxiv:' + last_id]:
break
else:
print "Skipped line: source id not forund. ", last_id
last_id = source_id
continue
# last node exists #
# write old references to last node
last_node['unknown_references'] += ref_buffer
last_id = source_id
ref_buffer = [ref_string]
def reference_fill_db(match_file=match_file, db=db):
"""
Reads references from match_file and creates corresponding links in db
"""
in_iter = open(match_file)
# line format examples at the end of the file
start = time()
rel_count = 0
for batch_count, batch in enumerate(group_generator(in_iter, 10000)):
sys.stderr.write(
'Processing reference %d x 10000. Elapsed time %d sec. Relations created %d. \n'
% (batch_count, time() - start, rel_count))
with db.transaction:
for line in batch:
try:
source_id, ref_string, target_id = line.rstrip().split('|')
except ValueError:
print 'Skipped line: not not enough separators "|". ', line[:
20]
continue
# .single property is broken! Have to loop through results (which should be a single one)
# equivalent to: source_node = source_idx['id']['arxiv:' + source_id].single
for source_node in source_idx['id']['arxiv:' + source_id]:
break
else:
print "Skipped line: source id not forund. ", line[:20]
continue
if not target_id == '':
# Lookup target_id in the index
target_node = None
for target_node in source_idx['id']['arxiv:' + target_id]:
break
if target_node:
# create reference relation
source_node.ref(target_node,
ref_string=ref_string,
label='Reference')
rel_count += 1
continue
def write_caches():
start = time()
for i, batch in enumerate(group_generator(PAPER.type.incoming, 1000)):
print "Filling citaion and author buffers. %d papers processed in %d sec." % (i*1000, time()-start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
ref_count = 0
for ref in paper_node.ref.outgoing:
ref_count += 1
cite_count = 0
for ref in paper_node.ref.incoming:
cite_count += 1
paper_node['c_reference_count'] = ref_count
paper_node['c_citation_count'] = cite_count
paper_node['c_authors'] = ' and '.join([ a_rel.endNode['name'] for a_rel in paper_node.author ])
def write_caches():
start = time()
for i, batch in enumerate(group_generator(PAPER.type.incoming, 1000)):
print "Filling citaion and author buffers. %d papers processed in %d sec." % (
i * 1000, time() - start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
ref_count = 0
for ref in paper_node.ref.outgoing:
ref_count += 1
cite_count = 0
for ref in paper_node.ref.incoming:
cite_count += 1
paper_node['c_reference_count'] = ref_count
paper_node['c_citation_count'] = cite_count
paper_node['c_authors'] = ' and '.join(
[a_rel.endNode['name'] for a_rel in paper_node.author])
def reference_fill_db(match_file = match_file , db=db):
"""
Reads references from match_file and creates corresponding links in db
"""
in_iter = open(match_file)
# line format examples at the end of the file
start = time()
rel_count = 0
for batch_count, batch in enumerate(group_generator(in_iter,10000)):
sys.stderr.write('Processing reference %d x 10000. Elapsed time %d sec. Relations created %d. \n' % (batch_count, time() - start, rel_count))
with db.transaction:
for line in batch:
try:
source_id, ref_string, target_id = line.rstrip().split('|')
except ValueError:
print 'Skipped line: not not enough separators "|". ', line[:20]
continue
# .single property is broken! Have to loop through results (which should be a single one)
# equivalent to: source_node = source_idx['id']['arxiv:' + source_id].single
for source_node in source_idx['id']['arxiv:' + source_id]:
break
else:
print "Skipped line: source id not forund. ", line[:20]
continue
if not target_id == '':
# Lookup target_id in the index
target_node = None
for target_node in source_idx['id']['arxiv:' + target_id]:
break
if target_node:
# create reference relation
source_node.ref(target_node, ref_string=ref_string, label='Reference')
rel_count += 1
continue
def meta_fill_db(db=db, limit=-1):
#
# Create Paper Nodes
#
start = time()
chunk_size = 1000
for batch_count, batch in enumerate(
group_generator(get_json_from_dir(meta_json_dir, limit=limit),
chunk_size)):
print 'Processing metadata record %d. Time elapsed: %d sec.' % (
batch_count * chunk_size, time() - start)
with db.transaction:
for rec_id, meta_dict in batch:
# create a new node
paper_node = db.node(
label='paper_node arxiv:' + rec_id,
title=meta_dict['title'][0],
abstract=meta_dict['description'][0],
unknown_references=[''],
date=meta_dict['date'][0],
source_url=meta_dict['identifier']
[0], # Check if really works?
source_id='arxiv:' + rec_id,
# arxiv_meta_dict = [ x for k,v in meta_dict.items() for x in (k, "|".join(v)) ],
)
# add a relation paper_node --[type]--> PAPER
paper_node.type(PAPER)
# register in source_id index
source_idx['id'][paper_node['source_id']] = paper_node
for author_name in meta_dict['creator']:
# create an author name node
author_node = add_get_author(author_name)
# create a relation paper_node --[author]--> author_node
paper_node.author(author_node)
print 'closing transaction'
def fill_meta_table(db_file, pkl_dir=pkl_dir, max_batch=-1):
"""
Creates a table meta in db_file and inserts records from pkl_files in pkl_dir
Schema: arxiv_id, author, title, abstract, info, subject, year
"""
# Initialize db
con = lite.connect(db_file)
# Create/Reset table
with con:
cur = con.cursor()
cur.execute("DROP TABLE IF EXISTS meta")
cur.execute(
"CREATE TABLE meta(arxiv_id TEXT, author TEXT, title TEXT, abstract TXT, info TEXT, subject TEXT, year INT)"
)
def prepare_meta_row(rec_id, meta_dict):
authors = ' and '.join(meta_dict['creator'])
title = meta_dict['title'][0]
abstract = meta_dict['description'][0]
info = ', '.join(meta_dict['identifier'][1:])
date = meta_dict['date'][0]
subject = ', '.join(meta_dict['subject'])
year = date[0:4]
return map(cleanup_rec,
[rec_id, authors, title, abstract, info, subject, year])
rows = (prepare_meta_row(rec_id, meta_dict)
for rec_id, meta_dict in get_meta_from_pkl(pkl_dir))
# Write rows, 10.000 per transaction
for batch_count, batch in enumerate(group_generator(rows, 10000)):
if batch_count == max_batch: break
if DEBUG: print "Writing meta row ", batch_count * 10000
with con:
cur.executemany("INSERT INTO meta VALUES(?, ?, ?, ?, ?, ?, ?)",
batch)
def meta_fill_db(db=db,limit = -1):
#
# Create Paper Nodes
#
start = time()
chunk_size = 1000
for batch_count, batch in enumerate(group_generator(
get_json_from_dir(meta_json_dir, limit = limit),
chunk_size)):
print 'Processing metadata record %d. Time elapsed: %d sec.' % (batch_count * chunk_size, time() - start)
with db.transaction:
for rec_id, meta_dict in batch:
# create a new node
paper_node = db.node(
label = 'paper_node arxiv:'+rec_id,
title = meta_dict['title'][0],
abstract = meta_dict['description'][0],
unknown_references = [''],
date = meta_dict['date'][0],
source_url = meta_dict['identifier'][0], # Check if really works?
source_id = 'arxiv:'+rec_id,
# arxiv_meta_dict = [ x for k,v in meta_dict.items() for x in (k, "|".join(v)) ],
)
# add a relation paper_node --[type]--> PAPER
paper_node.type(PAPER)
# register in source_id index
source_idx['id'][paper_node['source_id']] = paper_node
for author_name in meta_dict['creator']:
# create an author name node
author_node = add_get_author(author_name)
# create a relation paper_node --[author]--> author_node
paper_node.author(author_node)
print 'closing transaction'
def fill_author_table(db_file, pkl_dir=pkl_dir, max_batch=-1):
"""
Creates a lookup table with schema: author, year, arxiv_id, title
All elements from each 'creator' value in meta_dict give rows of the ayit table.
"""
# Initialize db
con = lite.connect(db_file)
# Write meta table
with con:
cur = con.cursor()
cur.execute("DROP TABLE IF EXISTS ayit_lookup")
cur.execute(
"CREATE TABLE ayit_lookup(author TEXT, year INT, arxiv_id TEXT, title TEXT)"
)
def prepare_ayit_row(rec_id, meta_dict):
title = meta_dict['title'][0]
date = meta_dict['date'][0]
year = date[0:4]
for author in meta_dict['creator']:
author = author.split(',')[0]
yield map(cleanup_rec, [author, year, rec_id, title])
# take union/chain of all generators returned by prepare_ayit_row
rows = (row for rec_id, meta_dict in get_meta_from_pkl(pkl_dir)
for row in prepare_ayit_row(rec_id, meta_dict))
# Write rows, 10.000 per transaction
for batch_count, batch in enumerate(group_generator(rows, 10000)):
if batch_count == max_batch: break
if DEBUG: print "Writing author info", batch_count * 10000
with con:
cur.executemany("INSERT INTO ayit_lookup VALUES(?, ?, ?, ?)",
batch)
def write_cite_rank(iterations=4):
start = time()
# damping factor
d = 0.85
# initial value
cr_0 = 1.
for iteration in range(iterations):
for batch_count, batch in enumerate(group_generator(PAPER.type.incoming, 1000)):
print "Calculating CiteRank. Iteration %d. Processed %d papers in %d sec." % (iteration, batch_count*1000, time()-start)
with db.transaction:
for paper_rel in batch:
paper_node = paper_rel.startNode
cr = (1 - d)
for cite_rel in paper_node.ref.incoming:
cite_node = cite_rel.startNode
try:
cr_node = cite_node['c_cite_rank']
except KeyError:
cr_node = cr_0
cr += d * cr_node / cite_node['c_reference_count']
paper_node['c_cite_rank'] = cr
from MetaRead import get_meta_from_pkl
import sys
sys.path.append('../tools')
from shared import group_generator
import json
for i, batch in enumerate(group_generator(get_meta_from_pkl('../DATA/META/PKL/'),10000)):
fh = open('../DATA/META/JSON/META_%03d.json' % i, 'w')
json.dump(batch,fh)
fh.close()
from MetaRead import get_meta_from_pkl
import sys
sys.path.append('../tools')
from shared import group_generator
import json
for i, batch in enumerate(
group_generator(get_meta_from_pkl('../DATA/META/PKL/'), 10000)):
fh = open('../DATA/META/JSON/META_%03d.json' % i, 'w')
json.dump(batch, fh)
fh.close()
