def expand(s):
inferable_pair = [tpl[0:2] for tpl in accepted_state]
uncertain_pair = [tpl[0:2] for tpl in unknown_state]
infered = []
sorted_entits = sorted(s.events + s.timex3s, key=lambda x: x.span[0].begin)
l = len(sorted_entits)
for r in range(2, l):
for fi in range(0, l - r):
from_enty = sorted_entits[fi]
ti = fi + r
to_enty = sorted_entits[ti]
tl3 = create_tlink_between_enty('dummy_id', from_enty, to_enty, src='expand')
tl3.votes = {'BEFORE': 0, 'OVERLAP': 0, 'AFTER': 0}
for d in range(1, r):
mi = fi + d
mid_enty = sorted_entits[mi]
tl1 = search_tlink_between_enty(from_enty, mid_enty, s.candids_within)
tl2 = search_tlink_between_enty(mid_enty, to_enty, s.candids_within)
if tl1 and tl2:
if (tl1.pred, tl2.pred) in inferable_pair[0:3]:
tl3.votes['BEFORE'] += 1
if (tl1.pred, tl2.pred) in inferable_pair[3:6]:
tl3.votes['AFTER'] += 1
if (tl1.pred, tl2.pred) in uncertain_pair:
tl3.votes['OVERLAP'] += 1
else:
print 'some bridge tlink is not ready'
if sum(tl3.votes.values()) > 0:
tl3.pred = get_kth_large_key(tl3.votes)
infered.append(tl3)
s.candids_within.append(tl3)
def verify(candid, gold):
matched = []
unmatched = []
unwanted = []
missed = []
for c in candid:
g = search_tlink_between_enty(c.from_enty, c.to_enty, gold)
if g:
if c.pred == g.type:
matched.append(c)
else:
c.real = g.type
unmatched.append(c)
else:
unwanted.append(c)
for g in gold:
c = search_tlink_between_enty(g.from_enty, g.to_enty, candid)
if not c:
missed.append(g)
return matched, unmatched, unwanted, missed
def get_conflict_info2(entits, tlinks, sort_key, stat_func):
sorted_entits = sorted(entits, key=sort_key)
accepted = []
conflict = []
unknown = []
incomplete = []
for tpl in combinations(sorted_entits, 3):
t1 = search_tlink_between_enty(tpl[0], tpl[1], tlinks)
t2 = search_tlink_between_enty(tpl[1], tpl[2], tlinks)
t3 = search_tlink_between_enty(tpl[0], tpl[2], tlinks)
if t1 and t2 and t3:
if (stat_func(t1), stat_func(t2), stat_func(t3)) in accepted_state:
accepted.append((t1, t2, t3))
if (stat_func(t1), stat_func(t2), stat_func(t3)) in conflict_state:
conflict.append((t1, t2, t3))
if (stat_func(t1), stat_func(t2), stat_func(t3)) in unknown_state:
unknown.append((t1, t2, t3))
else:
incomplete.append((t1, t2, t3))
#t3.votes = {'BEFORE': 0, 'OVERLAP': 0, 'AFTER': 0}
#t3.vote_probs = {'BEFORE': 0, 'OVERLAP': 0, 'AFTER': 0}
return accepted, conflict, unknown, incomplete
def get_conflict_info(entits, tlinks, sort_key, stat_func):
has_conflict = False
num_conflict = 0
sorted_entits = sorted(entits, key=sort_key)
for t3 in tlinks:
t3.votes = {'BEFORE': 0, 'OVERLAP': 0, 'AFTER': 0}
t3.vote_probs = {'BEFORE': 0, 'OVERLAP': 0, 'AFTER': 0}
bridge_entits = sorted_entits[:]
if t3.from_enty == t3.to_enty:
bridge_entits.remove(t3.from_enty)
print 'from_enty and to_enty are the same'
else:
bridge_entits.remove(t3.from_enty)
bridge_entits.remove(t3.to_enty)
#if len(bridge_entits) > 0:
# weight = 1.0 / len(bridge_entits)
for enty in bridge_entits:
if sorted_entits.index(enty) < sorted_entits.index(t3.from_enty):
t1 = search_tlink_between_enty(enty, t3.from_enty, tlinks)
t2 = search_tlink_between_enty(enty, t3.to_enty, tlinks)
if t1 and t2:
state = (opposite_label(stat_func(t1)), stat_func(t2), stat_func(t3))
else:
continue
elif sorted_entits.index(enty) > sorted_entits.index(t3.to_enty):
t1 = search_tlink_between_enty(t3.from_enty, enty, tlinks)
t2 = search_tlink_between_enty(t3.to_enty, enty, tlinks)
if t1 and t2:
state = (stat_func(t1), opposite_label(stat_func(t2)), stat_func(t3))
else:
continue
else:
#same_char = False
t1 = search_tlink_between_enty(t3.from_enty, enty, tlinks)
'''
if t1:
t1_stat = stat_func(t1)
else: # to handel two phrases starting at the same character
t1 = search_tlink_between_enty(enty, t3.from_enty, tlinks)
if t1:
t1_stat = opposite_label(stat_func(t1))
same_char = True
else:
continue
'''
t2 = search_tlink_between_enty(enty, t3.to_enty, tlinks)
'''
if t2:
t2_stat = stat_func(t2)
else: # to handel two phrases starting at the same character
t2 = search_tlink_between_enty(t3.to_enty, enty, tlinks)
if t2:
t2_stat = opposite_label(stat_func(t2))
same_char = True
else:
continue
'''
#if same_char:
# print 'same character', t3.from_enty, enty, t3.to_enty
#state = (t1_stat, t2_stat, stat_func(t3))
state = (stat_func(t1), stat_func(t2), stat_func(t3))
#print state
#prior = prior2
if state in accepted_state:
t3.votes[state[2]] += 1
#t3.vote_probs[state[2]] += t1.probs[state[0]] * prior[state[0]] * t2.probs[state[1]] * prior[state[1]]
#t3.vote_probs[state[2]] += t1.probs[state[0]] * t2.probs[state[1]]
if state in conflict_state[0:6]:
t3.votes['BEFORE'] += 1
#t3.vote_probs['BEFORE'] += t1.probs[state[0]] * prior[state[0]] * t2.probs[state[1]] * prior[state[1]]
#t3.vote_probs['BEFORE'] += t1.probs[state[0]] * t2.probs[state[1]]
if state in conflict_state[6:12]:
t3.votes['AFTER'] += 1
#t3.vote_probs['AFTER'] += t1.probs[state[0]] * prior[state[0]] * t2.probs[state[1]] * prior[state[1]]
#t3.vote_probs['AFTER'] += t1.probs[state[0]] * t2.probs[state[1]]
#if state in conflict_state[6:8]:
# t3.votes['OVERLAP'] += 1
#t3.vote_probs['OVERLAP'] += t1.probs[state[0]] * prior[state[0]] * t2.probs[state[1]] * prior[state[1]]
#t3.vote_probs['OVERLAP'] += t1.probs[state[0]] * t2.probs[state[1]]
#for key in ['BEFORE', 'OVERLAP', 'AFTER']:
# if t3.votes[key] > 0:
# t3.vote_probs[key] = t3.vote_probs[key] / t3.votes[key]
if sum(t3.votes.values()) > t3.votes[stat_func(t3)]:
num_conflict += 1
if num_conflict > 0:
has_conflict = True
return has_conflict, num_conflict
def resolve_conflict_within(s):
sorted_entits = sorted(s.events + s.timex3s, key=lambda x: x.span[0].begin)
l = len(sorted_entits)
if l <= 2:
return
for i in range(l - 2):
for j in range(i + 2, l):
a = sorted_entits[i]
b = sorted_entits[j - 1]
c = sorted_entits[j]
t1 = search_tlink_between_enty(a, b, s.candids_within)
t2 = search_tlink_between_enty(b, c, s.candids_within)
t3 = search_tlink_between_enty(a, c, s.candids_within)
if t1.pred == 'BEFORE' and t2.pred == 'BEFORE' and t3.pred == 'OVERLAP':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'BEFORE' and t2.pred == 'BEFORE' and t3.pred == 'AFTER':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'BEFORE' and t2.pred == 'OVERLAP' and t3.pred == 'OVERLAP':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'BEFORE' and t2.pred == 'OVERLAP' and t3.pred == 'AFTER':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'BEFORE' and t3.pred == 'OVERLAP':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'BEFORE' and t3.pred == 'AFTER':
#t3.pred = 'BEFORE'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'OVERLAP' and t3.pred == 'BEFORE':
#t3.pred = 'OVERLAP'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'OVERLAP' and t3.pred == 'AFTER':
#t3.pred = 'OVERLAP'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'AFTER' and t3.pred == 'BEFORE':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'OVERLAP' and t2.pred == 'AFTER' and t3.pred == 'OVERLAP':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'AFTER' and t2.pred == 'OVERLAP' and t3.pred == 'BEFORE':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'AFTER' and t2.pred == 'OVERLAP' and t3.pred == 'OVERLAP':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'AFTER' and t2.pred == 'AFTER' and t3.pred == 'BEFORE':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
continue
if t1.pred == 'AFTER' and t2.pred == 'AFTER' and t3.pred == 'OVERLAP':
#t3.pred = 'AFTER'
t3.pred = get_kth_large_key(t3.probs, 2)
'''
for tl in s.candids_within:
try:
tl.pred = tl.temp
except AttributeError:
continue
'''
return