class Relativizers(object):
def __init__(self,model=None):
self.features=RelFeatures()
self.model=model
def learn(self,tree,outfile):
for rel in tree.rels:
features=self.features.create(rel.gov,rel.dep,tree)
types=is_dep(rel.gov,rel.dep,tree)
if len(types)<2: continue # post-processed dependency, do not use in training
for dtype in types:
if dtype==u"rel": continue
klass=dtype.split(u"&",1)[1]
writeData(outfile,klass,features)
break
def post_process_rel(self,g,d,t,tree):
# move dependency if iccomp and it does not have this dtype already
for dep in tree.deps:
if dep.gov==g and dep.dtype==u"iccomp":
deps=set()
for dep2 in tree.childs[dep.dep]: # dependents of iccomp
deps.add(dep2.dtype) # collect all dependents of iccomp
if t not in deps:
g=dep.dep
break
return g,d,t
def predict(self,tree):
if self.model is None:
print >> sys.stderr, u"no model found"
sys.exit(1)
for rel in tree.rels:
features=self.features.create(rel.gov,rel.dep,tree)
klass=self.model.predict_one(features)
klass_str=u"rel&"+self.model.number2klass[klass]
g,d,t=self.post_process_rel(rel.gov,rel.dep,klass_str,tree)
dependency=Dep(g,d,t,flag=u"REL")
#print dependency
tree.add_dep(dependency)
def __init__(self,model=None):
self.features=RelFeatures()
self.model=model
