def linkTemplates(self, sentence):
""" link value template to best matching mention template in the same sentence.
It is assumed that mention clustering has not occurred yet.
"""
templates = sentence.templates
onList = templates.getList('on')
erList = templates.getList('eventrate')
omList = []
unmatchedER = []
unmatchedON = []
for er in erList:
for on in onList:
if on.shouldBelongToSameOutcomeMeasurement(er):
om = OutcomeMeasurement(on)
om.addEventRate(er)
omList.append(om)
# print '&&&&&&& Associating:', on.value, er.value, on.outcomeMeasurement, er.outcomeMeasurement
# print er.outcomeNumber, on.textEventrate
break
if er.outcomeNumber == None:
unmatchedER.append(er)
# om = OutcomeMeasurement(er)
# omList.append(om)
# create outcome measurement templates lone on templates
for on in onList:
if on.textEventrate == None:
unmatchedON.append(on)
# om = OutcomeMeasurement(on)
# omList.append(om)
for er in unmatchedER:
if er.outcomeNumber == None:
# eventrate still not matched, create outcome measurement just for it
om = OutcomeMeasurement(er)
omList.append(om)
for on in unmatchedON:
if on.textEventrate == None:
# outcome number still not matched, create outcome measurement just for it
om = OutcomeMeasurement(on)
omList.append(om)
# for om in omList:
# om.display()
sentence.templates.addOutcomeMeasurementList(omList)
def computeTrueStats(self, abstract):
""" compute summaries statistics using annotations """
self.outcomeNumbers = []
self.eventRates = []
self.stats = []
self.trueStats = True
self.abstract = abstract
self.groupsById = {}
self.outcomesById = {}
self.timesById = {}
omHash = {}
for s in abstract.sentences:
# find all of the annotated templates in the sentence
templates = s.annotatedTemplates
gList = templates.getList('group')
oList = templates.getList('outcome')
gsList = templates.getList('gs')
onList = templates.getList('on')
erList = templates.getList('eventrate')
tList = templates.getList('time')
# print abstract.id
# for er in erList:
# print er.value,
# print
for t in tList:
# times.append(t)
if t.getAnnotatedId() in self.timesById:
self.timesById[t.getAnnotatedId()].merge(t)
else:
self.timesById[t.getAnnotatedId()] = t
for g in gList:
# groups.append(g)
if g.getAnnotatedId() in self.groupsById:
self.groupsById[g.getAnnotatedId()].merge(g)
else:
self.groupsById[g.getAnnotatedId()] = g
for outcome in oList:
# outcomes.append(outcome)
if outcome.getAnnotatedId() != None and len(outcome.getAnnotatedId()) > 0:
if outcome.getAnnotatedId() in self.outcomesById:
self.outcomesById[outcome.getAnnotatedId()].merge(outcome)
else:
self.outcomesById[outcome.getAnnotatedId()] = outcome
else:
print abstract.id, outcome.name, 'does not have an ID.',
print 'Not using it for summary stats.'
# for gs in gsList:
# self.groupSizes.append(gs)
# link groups and their sizes
for gs in gsList:
gid = gs.token.getAnnotationAttribute('gs', 'group')
if gid in self.groupsById:
g = self.groupsById[gid]
gs.group = g
g.addSize(gs)
tid = gs.token.getAnnotationAttribute('gs', 'time')
if tid in self.timesById:
t = self.timesById[tid]
gs.time = t
# for gid,g in self.groupsById.items():
# print 'Group id:', gid, ', name = ', g.name, ', size =', g.getSize()
# link all relevant information needed for each outcome measurement
for on in onList:
gid = on.token.getAnnotationAttribute('on', 'group')
oid = on.token.getAnnotationAttribute('on', 'outcome')
tid = on.token.getAnnotationAttribute('on', 'time')
csID = on.token.getAnnotationAttribute('on', 'compareSet')
# print 'on:',on.value, csID
if oid in self.outcomesById:
oTemplate = self.outcomesById[oid]
gTemplate = self.groupsById.get(gid, None)
tTemplate = self.timesById.get(tid, None)
if oid not in omHash:
omHash[oid] = []
om = OutcomeMeasurement(on)
om.addGroup(gTemplate)
om.addOutcome(oTemplate)
om.addTime(tTemplate)
omHash[oid].append(om)
else:
print abstract.id, '??? Outcome number', on.value,
print 'does not have a matching outcome with id =', oid
# print '-->',
# om.write(sys.stdout)
for er in erList:
gid = er.token.getAnnotationAttribute('eventrate', 'group')
oid = er.token.getAnnotationAttribute('eventrate', 'outcome')
tid = er.token.getAnnotationAttribute('eventrate', 'time')
csID = er.token.getAnnotationAttribute('eventrate', 'compareSet')
# print abstract.id+': er: ',er.value, csID
if oid in self.outcomesById:
oTemplate = self.outcomesById[oid]
gTemplate = self.groupsById.get(gid, None)
tTemplate = self.timesById.get(tid, None)
# print abstract.id+': er: ', er.value, gTemplate, tTemplate, csID
if oid not in omHash:
omHash[oid] = []
matchFound = False
for om in omHash[oid]:
if om.getGroup() == gTemplate and om.getTime() == tTemplate and om.getCompareSetID() == csID:
om.addEventRate(er)
# print 'adding', er.value
# om.write(sys.stdout)
matchFound = True
break
# else:
# print om.getGroup(), om.getTime(), om.getCompareSetID
if matchFound == False:
# event rate not added to existing outcome measurement, create new measurement
om = OutcomeMeasurement(er)
om.addGroup(gTemplate)
om.addOutcome(oTemplate)
om.addTime(tTemplate)
omHash[oid].append(om)
else:
print 'Event rate missing outcome annotation in abstract ',
print abstract.id, ':', s.toString()
er.write(sys.stdout)
for oid in omHash.keys():
omList = omHash[oid]
for i in range(0, len(omList)):
om1 = omList[i]
csID1 = om1.getCompareSetID()
# print abstract.id, csID1,':',
for j in range(i+1, len(omList)):
om2 = omList[j]
csID2 = om2.getCompareSetID()
# print csID2,
if csID1 == csID2 and om1.isComplete() and om2.isComplete() \
and om1.getGroup() != om2.getGroup() and om1.getTime() == om2.getTime():
ssTemplate = SummaryStat(om1, om2, useAnnotated=True)
self.stats.append(ssTemplate)
om1.used = True
om2.used = True
# print
if om1.used == False:
self.unmatchedMeasurements.append(om1)
for om in self.unmatchedMeasurements:
if om.getOutcome() != None:
om.getOutcome().unusedNumbers.append(om)
def linkTemplates(self, sentence):
""" link group size and group templates using Hungarian matching algorithm """
# print 'linking all templates'
templates = sentence.templates
onList = templates.getList('on')
erList = templates.getList('eventrate')
abstract = sentence.abstract
if abstract not in self.incompleteMatches:
self.incompleteMatches[abstract] = []
omHash = {}
for er in erList:
if er.group != None and er.outcome != None:
# remember the mention matched with the value
# save this information in a feature vector to be retrieved in linkQuantityAndMention()
fv = FeatureVector(-1, -1, None)
fv.mTemplate = er.group
fv.qTemplate = er
er.addMatchFeatures(fv)
fv = FeatureVector(-1, -1, None)
fv.mTemplate = er.outcome
fv.qTemplate = er
er.addMatchFeatures(fv)
groupEntity = er.group.rootMention()
outcomeEntity = er.outcome.rootMention()
er.group = None
er.outcome = None
om = OutcomeMeasurement(er)
if (groupEntity, outcomeEntity) not in omHash:
omHash[(groupEntity, outcomeEntity)] = om
elif omHash[(groupEntity, outcomeEntity)] != None:
# there is already an outcome measurement for this group, outcome
# check if this one is closer
# closer if it distance to closest mention is less
# if the same, use total distance
# if that is the same, use value that occurs earlier in sentence
currentOM = omHash[(groupEntity, outcomeEntity)]
current = currentOM.getTextEventRate()
closest = self.closestValue(er, current)
if closest == None:
# both same distance, discard both
omHash[(groupEntity, outcomeEntity)] = None
elif closest == er:
omHash[(groupEntity, outcomeEntity)] = om
self.incompleteMatches[abstract].append(
OutcomeMeasurementAssociation(
groupEntity, outcomeEntity, currentOM, 0))
else:
self.incompleteMatches[abstract].append(
OutcomeMeasurementAssociation(
groupEntity, outcomeEntity, om, 0))
for on in onList:
if on.group != None and on.outcome != None:
fv = FeatureVector(-1, -1, None)
fv.mTemplate = on.group
fv.qTemplate = on
on.addMatchFeatures(fv)
fv = FeatureVector(-1, -1, None)
fv.mTemplate = on.outcome
fv.qTemplate = on
on.addMatchFeatures(fv)
groupEntity = on.group.rootMention()
outcomeEntity = on.outcome.rootMention()
on.group = None
on.outcome = None
# check if this ON is useful, can we compute an event rate with it?
gs = on.getGroupSize()
if gs > 0:
# we can compute an event rate
om = OutcomeMeasurement(on)
if (groupEntity, outcomeEntity) not in omHash:
omHash[(groupEntity, outcomeEntity)] = om
elif omHash[(groupEntity, outcomeEntity)] != None:
# there is already a outcome measurement
currentOM = omHash[(groupEntity, outcomeEntity)]
currentON = currentOM.getOutcomeNumber()
currentER = currentOM.getTextEventRate()
# check if this on should be merged with an event rate
if currentON == None and currentER != None and on.equivalentEventRates(
currentER.eventRate()):
currentOM.addOutcomeNumber(on)
else:
# on not compatible with existing value
# is it closer?
if currentON != None and currentER != None:
closestVal = self.closestValue(
currentON, currentER)
if closestVal == None:
# if both the same distance, just use the ER
closestVal = currentER
elif currentON != None:
closestVal = currentON
else:
closestVal = currentER
closestVal = self.closestValue(closestVal, on)
if closestVal == None:
# both same distance, discard both
omHash[(groupEntity, outcomeEntity)] = None
elif closestVal == on:
omHash[(groupEntity, outcomeEntity)] = om
self.incompleteMatches[abstract].append(
OutcomeMeasurementAssociation(
groupEntity, outcomeEntity, currentOM,
0))
else:
self.incompleteMatches[abstract].append(
OutcomeMeasurementAssociation(
groupEntity, outcomeEntity, om, 0))
omList = []
for (group, outcome), om in omHash.items():
if om != None:
self.linkOutcomeMeasurementAssociations(
om, group, outcome, 0.5)
omList.append(om)
sentence.templates.addOutcomeMeasurementList(omList)