def single_relation(query):
if len(query.r) != 1:
return False
r = query.r[-1]
if inverted(r):
return False
return r
def single_relation(query):
if len(query.r) != 1:
return False
r = query.r[-1]
if inverted(r):
return False
return r
def t(word):
try:
if inverted(word):
return invert(d[invert(word)])
else:
return d[word]
except KeyError:
return word
def t(word):
try:
if inverted(word):
return invert(d[invert(word)])
else:
return d[word]
except KeyError:
return word
def summarize_neighborhood(graph,
seed=None,
max_depth=2,
nbr_samples=20,
save_path=None):
if seed is None:
seed = random.choice(graph.neighbors.keys())
print 'seed:', seed
triples = set()
explored = set()
queue = deque()
queue.append((seed, 0))
while len(queue) != 0:
entity, depth = queue.popleft()
if depth >= max_depth:
continue
# loop through each available relation
for r in graph.neighbors[entity]:
# sample neighbors
nbrs = graph.neighbors[entity][r]
sampled_nbrs = util.sample_if_large(nbrs,
nbr_samples,
replace=False)
num_missed = len(nbrs) - len(sampled_nbrs)
edge_crossed = lambda target: (entity, r, target) if not inverted(
r) else (target, invert(r), entity)
# document edges crossed, and add nbrs to queue
for nbr in sampled_nbrs:
triples.add(edge_crossed(nbr))
if nbr not in explored:
queue.append((nbr, depth + 1))
# add "summary entity" for all entities we missed
if num_missed > 0:
triples.add(
edge_crossed('{}_{}_{}'.format(entity, r, num_missed)))
if save_path is not None:
with open(save_path, 'w') as f:
for tr in triples:
f.write('\t'.join(tr) + '\n')
return list(triples)
def find_inverses(graph):
inverse = {}
# heuristic for finding inverse relation
for r in graph.relation_args.keys():
ct = Counter()
# count co-occurring relations
for k, s in enumerate(graph.relation_args[r]['s']):
t = graph.neighbors[s][r][0] # pick an arbitrary target
ct.update(graph.neighbors[t].keys()) # look at the target's relations
if k >= 100:
break
# inverse = top co-occurring relation that isn't inverse
for r2, count in ct.most_common():
if not inverted(r2):
inverse[r] = r2
print r, r2, count
break
def find_inverses(graph):
inverse = {}
# heuristic for finding inverse relation
for r in graph.relation_args.keys():
ct = Counter()
# count co-occurring relations
for k, s in enumerate(graph.relation_args[r]['s']):
t = graph.neighbors[s][r][0] # pick an arbitrary target
ct.update(
graph.neighbors[t].keys()) # look at the target's relations
if k >= 100:
break
# inverse = top co-occurring relation that isn't inverse
for r2, count in ct.most_common():
if not inverted(r2):
inverse[r] = r2
print r, r2, count
break
def summarize_neighborhood(graph, seed=None, max_depth=2, nbr_samples=20, save_path=None):
if seed is None:
seed = random.choice(graph.neighbors.keys())
print 'seed:', seed
triples = set()
explored = set()
queue = deque()
queue.append((seed, 0))
while len(queue) != 0:
entity, depth = queue.popleft()
if depth >= max_depth:
continue
# loop through each available relation
for r in graph.neighbors[entity]:
# sample neighbors
nbrs = graph.neighbors[entity][r]
sampled_nbrs = util.sample_if_large(nbrs, nbr_samples, replace=False)
num_missed = len(nbrs) - len(sampled_nbrs)
edge_crossed = lambda target: (entity, r, target) if not inverted(r) else (target, invert(r), entity)
# document edges crossed, and add nbrs to queue
for nbr in sampled_nbrs:
triples.add(edge_crossed(nbr))
if nbr not in explored:
queue.append((nbr, depth + 1))
# add "summary entity" for all entities we missed
if num_missed > 0:
triples.add(edge_crossed('{}_{}_{}'.format(entity, r, num_missed)))
if save_path is not None:
with open(save_path, 'w') as f:
for tr in triples:
f.write('\t'.join(tr) + '\n')
return list(triples)