def runSetExpan(seedEidsWithConfidence, numToExpand):
expandedEidsWithConfidence = set_expan.setExpan(
seedEidsWithConfidence=seedEidsWithConfidence,
negativeSeedEids=negativeSeedEids,
eid2patterns=eid2patterns,
pattern2eids=pattern2eids,
eidAndPattern2strength=eidAndPattern2strength,
eid2types=eid2types,
type2eids=type2eids,
eidAndType2strength=eidAndType2strength,
eid2ename=eid2ename,
eid2embed=eid2embed,
source_weights={
"sg": 1.0,
"tp": 5.0,
"eb": 5.0
},
max_expand_eids=numToExpand,
use_embed=True,
use_type=True,
FLAGS_VERBOSE=False,
FLAGS_DEBUG=False)
print(expandedEidsWithConfidence)
for ele in expandedEidsWithConfidence:
print("eid=", ele[0], "ename=", eid2ename[ele[0]], "confidence_score=",
ele[1])
return expandedEidsWithConfidence
def runsetexpan(setname):
for st in good_gold_set:
if st == setname:
fin = open('../data/eval/results/{}/setexpan/{}.log'.format(dataset, st), 'w+')
entities = good_gold_set[st]
for numseeds, qs in queries[st].items():
fin.write("[{} seeds]\n".format(numseeds))
start = time.time()
ap1 = []
idx = 0
for q in qs:
idx += 1
print("Running \"{}\" [{} seeds] exp {} ...".format(st, numseeds, idx))
samples = [ename2eid[x] for x in list(q)]
entities_to_retrieve = [i for i in entities if i not in samples]
truth = [eid2ename[int(x)] for x in entities_to_retrieve] # names
seedEidsWithConfidence = [(int(ele), 0.0) for ele in samples]
negativeSeedEids = set()
try:
expandedEidsWithConfidence = set_expan.setExpan(
seedEidsWithConfidence=seedEidsWithConfidence,
negativeSeedEids=negativeSeedEids,
eid2patterns=eid2patterns,
pattern2eids=pattern2eids,
eidAndPattern2strength=eidAndPattern2strength,
eid2types=eid2types,
type2eids=type2eids,
eidAndType2strength=eidAndType2strength,
eid2ename=eid2ename,
FLAGS_VERBOSE=False,
FLAGS_DEBUG=False
)
answers = []
for ele in expandedEidsWithConfidence:
answers.append(eid2ename[ele[0]])
ap1.append(ap(answers, truth))
except:
print(setname + " from " + str(current_process().name))
ap1.append(0.0)
end = time.time()
fin.write("{}\n".format(str(ap1)))
fin.write("{}\n".format((end-start)/idx))
fin.close()
def runsetexpan(setname):
print("running eval for {}".format(setname))
for st in good_gold_set:
if st == setname:
fin = open('../data/eval/results/{}/{}.log'.format(dataset, st),
'w+')
entities = good_gold_set[st]
ap2 = []
var1 = []
var2 = []
times = []
for numseeds, qs in queries[st].items():
if numseeds != 3:
continue
start = time.time()
ap1 = []
idx = 0
for q in qs:
idx += 1
print("Running \"{}\" [{} seeds] iteration {} ...".format(
st, numseeds, idx))
samples = [ename2eid[x] for x in list(q)]
entities_to_retrieve = [
i for i in entities if i not in samples
]
truth = [eid2ename[int(x)]
for x in entities_to_retrieve] # names
seedEidsWithConfidence = [(int(ele), 0.0)
for ele in samples]
fin.write("input: {}\n".format(
str([eid2ename[int(x)] for x in samples])))
negativeSeedEids = set()
expandedEidsWithConfidence = set_expan.setExpan(
seedEidsWithConfidence=seedEidsWithConfidence,
negativeSeedEids=negativeSeedEids,
eid2patterns=eid2patterns,
pattern2eids=pattern2eids,
eidAndPattern2strength=eidAndPattern2strength,
eid2types=eid2types,
type2eids=type2eids,
eidAndType2strength=eidAndType2strength,
eid2ename=eid2ename,
FLAGS_VERBOSE=False,
FLAGS_DEBUG=False)
answers = []
for ele in expandedEidsWithConfidence:
answers.append(eid2ename[ele[0]])
fin.write("output: {}\n".format(str(answers)))
fin.write('----------------------------------------\n')
ap1.append(ap(answers, truth))
end = time.time()
fin.write(
"[{} seeds]: MAP: {}, variance: {}, std.variance: {}, avg. time: {:.2f} secs\n\n"
.format(numseeds,
sum(ap1) / len(ap1), np.var(ap1), np.std(ap1),
(end - start) / idx))
fin.write("{}\n".format(str(ap1)))
ap2.append(sum(ap1) / len(ap1))
var1.append(np.var(ap1))
var2.append(np.std(ap1))
times.append((end - start) / idx)
fin.write(str(ap2) + '\n')
fin.write(str(times) + '\n')
fin.close()
print('SEED_SAMPLE #' + str(idx))
userInput = random.sample(seeds, 10)
seedEidsWithConfidence = [(ename2eid[ele], 0.0)
for ele in userInput]
negativeSeedEids = set()
for ele in eid2ename:
if not eid2ename[ele].startswith('PHRASE_'):
negativeSeedEids.add(ele)
expandedEidsWithConfidence = set_expan.setExpan(
seedEidsWithConfidence=seedEidsWithConfidence,
negativeSeedEids=negativeSeedEids,
eid2patterns=eid2patterns,
pattern2eids=pattern2eids,
eidAndPattern2strength=eidAndPattern2strength,
eid2types=eid2types,
type2eids=type2eids,
eidAndType2strength=eidAndType2strength,
eid2ename=eid2ename,
FLAGS_VERBOSE=True,
FLAGS_DEBUG=True)
for idx, ele in enumerate(expandedEidsWithConfidence):
if ele[0] not in MRR:
MRR[ele[0]] = 0.0
MRR[ele[0]] += 1 / (idx + 1)
MRR_sorted = sorted(MRR.items(),
key=lambda kv: kv[1],
reverse=True)
with open(folder + enttype + '/seed' + str(seedSetNo) + '.txt',
if args.debug:
print("[Width Expansion] Expand: {}, restrictions: {}".format(
targetNode, negativeSeedEids))
# at least grow one node
max_expand_eids = max(
len(negativeSeedEids) + 1,
level2max_expand_eids[targetNode.level])
newOrderedChildrenEidsWithConfidence = setExpan(
seedEidsWithConfidence,
negativeSeedEids,
eid2patterns,
pattern2eids,
eidAndPattern2strength,
eid2types,
type2eids,
eidAndType2strength,
eid2ename,
eid2embed,
source_weights=level2source_weights[targetNode.level],
max_expand_eids=max_expand_eids,
use_embed=True,
use_type=True)
newOrderedChildren = []
for ele in newOrderedChildrenEidsWithConfidence:
newChildEid = ele[0]
confidence_score = ele[1]
confidence_score += targetNode.confidence_score
synonym_FLAG = False # Check synonmy
for sibling in targetNode.children:
if isSynonym(args, newChildEid, sibling.eid):
# seedEidsWithConfidence_dblp_method_class = [(9012, 0.0), (652, 0.0), (13169, 0.0), (15819, 0.0), (14650, 0.0),
# (15554, 0.0), (3334, 0.0), (14542, 0.0), (3044, 0.0), (9847, 0.0),
# (9826, 0.0), (9619, 0.0), (11985, 0.0), (6825, 0.0), (2767, 0.0),
# (9152, 0.0), (12024, 0.0), (10105, 0.0), (3848, 0.0), (4810, 0.0),
# (11944, 0.0), (6319, 0.0), (126, 0.0)]
negativeSeedEids = set()
expandedEidsWithConfidence = set_expan.setExpan(
seedEidsWithConfidence=seedEidsWithConfidence,
negativeSeedEids=negativeSeedEids,
eid2patterns=eid2patterns,
pattern2eids=pattern2eids,
eidAndPattern2strength=eidAndPattern2strength,
eid2types=eid2types,
type2eids=type2eids,
eidAndType2strength=eidAndType2strength,
eid2ename=eid2ename,
eid2embed=eid2embed,
source_weights={
"sg": 1.0,
"tp": 5.0,
"eb": 5.0
},
use_embed=True,
use_type=True,
FLAGS_VERBOSE=True,
FLAGS_DEBUG=True)
print("=== In test case ===")
for ele in expandedEidsWithConfidence:
print("eid=", ele[0], "ename=", eid2ename[ele[0]], "confidence_score=",
ele[1])
with open("./test_setexpan.txt", "w") as fout: