def _combine_probs(phrasebook, attach, relate):
"""for all EDU pairs, retrieve probability of the best relation
on that pair, given the probability of an attachment
:type attach: DataAndModel
:type relate: DataAndModel
"""
# !! instances set must correspond to same edu pair in the same order !!
distrib = []
edu_pair = mk_edu_pairs(phrasebook, attach.data.domain)
attach_instances = sorted(attach.data, key=lambda x: x.get_metas())
relate_instances = sorted(relate.data, key=lambda x: x.get_metas())
inst_pairs = zip(attach_instances, relate_instances)
for i, (att, rel) in enumerate(inst_pairs):
if not _instance_check(phrasebook, att, rel):
print(
"mismatch of attachment/relation instance, "
"instance number",
i,
_instance_help(phrasebook, att),
_instance_help(phrasebook, rel),
file=sys.stderr)
prob, best = _combine_single_prob(attach, relate, att, rel)
edu1, edu2 = edu_pair(att)
distrib.append((edu1, edu2, prob, best))
return distrib
def _combine_probs(phrasebook, attach, relate):
"""for all EDU pairs, retrieve probability of the best relation
on that pair, given the probability of an attachment
:type attach: DataAndModel
:type relate: DataAndModel
"""
# !! instances set must correspond to same edu pair in the same order !!
distrib = []
edu_pair = mk_edu_pairs(phrasebook, attach.data.domain)
attach_instances = sorted(attach.data, key=lambda x: x.get_metas())
relate_instances = sorted(relate.data, key=lambda x: x.get_metas())
inst_pairs = zip(attach_instances, relate_instances)
for i, (att, rel) in enumerate(inst_pairs):
if not _instance_check(phrasebook, att, rel):
print("mismatch of attachment/relation instance, "
"instance number", i,
_instance_help(att, phrasebook),
_instance_help(rel, phrasebook),
file=sys.stderr)
prob, best = _combine_single_prob(attach, relate, att, rel)
edu1, edu2 = edu_pair(att)
distrib.append((edu1, edu2, prob, best))
return distrib
def _get_attach_prob_orange(phrasebook, attach):
"""
Attachment probabilities (only) for each EDU pair in the data
"""
# orange-based models
edu_pair = mk_edu_pairs(phrasebook, attach.data.domain)
prob_distrib = []
for inst in attach.data:
edu1, edu2 = edu_pair(inst)
prob = _get_inst_attach_orange(attach.model, inst)
prob_distrib.append((edu1, edu2, prob, "unlabelled"))
return prob_distrib
def _get_attach_prob_orange(config, attach):
"""
Attachment probabilities (only) for each EDU pair in the data
"""
# orange-based models
edu_pair = mk_edu_pairs(config.phrasebook, attach.data.domain)
prob_distrib = []
for inst in attach.data:
edu1, edu2 = edu_pair(inst)
probs = attach.model(inst, Classifier.GetProbabilities)[1]
prob_distrib.append((edu1, edu2, probs, "unlabelled"))
return prob_distrib
def _get_attach_prob_orange(config, attach):
"""
Attachment probabilities (only) for each EDU pair in the data
"""
# orange-based models
edu_pair = mk_edu_pairs(config.phrasebook, attach.data.domain)
prob_distrib = []
for inst in attach.data:
edu1, edu2 = edu_pair(inst)
probs = attach.model(inst, Classifier.GetProbabilities)[1]
prob_distrib.append((edu1, edu2, probs, "unlabelled"))
return prob_distrib
def _get_attach_prob_orange(phrasebook, attach):
"""
Attachment probabilities (only) for each EDU pair in the data
"""
# orange-based models
edu_pair = mk_edu_pairs(phrasebook, attach.data.domain)
prob_distrib = []
for inst in attach.data:
edu1, edu2 = edu_pair(inst)
prob = _get_inst_attach_orange(attach.model, inst)
prob_distrib.append((edu1, edu2, prob, "unlabelled"))
return prob_distrib
def get_edu_pair(self, orange_inst):
return mk_edu_pairs(self.__meta_features, self.__domain)(orange_inst)
def get_edu_pair(self, orange_inst):
return mk_edu_pairs(self.__phrasebook, self.__domain)(orange_inst)
def get_edu_pair(self, orange_inst):
return mk_edu_pairs(self.__phrasebook, self.__domain)(orange_inst)
def get_edu_pair(self, orange_inst):
return mk_edu_pairs(self.__meta_features, self.__domain)(orange_inst)
