def __init__(self,model=None):
self.features=JumpFeatures()
self.model=model
self.resttypes=set([u"cc",u"conj",u"punct",u"ellipsis"])
class ConjPropagation(object):
def __init__(self,model=None):
self.features=JumpFeatures()
self.model=model
self.resttypes=set([u"cc",u"conj",u"punct",u"ellipsis"])
def can_jump(self,dep,tree):
""" Check whether dep can jump. """
if dep.dtype in self.resttypes: return False
for conj in tree.conjs:
if dep.dep==conj.gov or dep.gov==conj.gov:
return True
return False
def possible_jumps(self,g,d,tree):
""" g,d is one candidate for propagation, returns a set of 'propagated' (gov,dep) tuples """
candidates=set()
for conj in tree.conjs:
if conj.gov==d: # dependent can move
candidates.add((g,conj.dep))
elif conj.gov==g: # governor can move
candidates.add((conj.dep,d))
return candidates
# Recursive search for all possible rec jumps
def gather_all_jumps(self,g,d,tree):
recs=self.possible_jumps(g,d,tree)
if len(recs)>0:
new_set=set()
for gov,dep in recs:
new_set|=self.gather_all_jumps(gov,dep,tree)
recs|=new_set
return recs
else:
return recs
def learn(self,tree,outfile):
for dep in tree.deps:
if dep.flag!=u"CC" and self.can_jump(dep,tree):
if dep.dtype==u"rel":
continue
new_deps=self.gather_all_jumps(dep.gov,dep.dep,tree)
for g,d in new_deps:
types=is_dep(g,d,tree)
if not types:
klass=u"no"
elif len(types)==2 and u"rel" in types:
for t in types:
if t!=u"rel":
klass=t
break
else:
assert len(types)<2
klass=types[0]
features=self.features.create(dep,g,d,tree)
writeData(outfile,klass,features)
def predict(self,tree):
""" Jump one tree. """
if self.model is None:
print >> sys.stderr, u"no model found"
sys.exit(1)
new=[]
for dep in tree.deps:
if self.can_jump(dep,tree):
if dep.dtype==u"rel":
continue # this is just the rel, we want the secondary function
new_deps=self.gather_all_jumps(dep.gov,dep.dep,tree)
for g,d in new_deps:
features=self.features.create(dep,g,d,tree) #...should return (name,value) tuples
klass=self.model.predict_one(features)
klass_str=self.model.number2klass[klass]
if klass_str==u"no": continue
if u"&" in klass_str: # this is merged rel
#print >> sys.stderr, klass_str
dependency=Dep(g,d,u"rel",flag=u"CC") # add also rel
new.append(dependency)
dependency=Dep(g,d,klass_str,flag=u"CC")
new.append(dependency)
for dep in new:
tree.add_dep(dep)
