def scoreABSA(self, content):
predict_data = mydatasets.SemEval(self.text_field, self.as_field,
self.sm_field, content)
predict_iter = data.Iterator(predict_data,
batch_size=len(predict_data),
train=False,
sort=False,
sort_within_batch=False)
labels, scores = self.absa_predict(predict_iter, self.absa_model,
self.args_absa)
labels, scores = self._reorder_sm_field(labels, scores)
return [self.ontology[l] for l in labels], scores
def load_semeval_data(text_field, as_field, sm_field, years, aspects, **kargs):
if not args.atsa:
semeval_train, semeval_test = get_semeval(years, aspects, args.r_l,
args.use_attribute)
else:
semeval_train, semeval_test = get_semeval_target(years, args.r_l)
print(semeval_test)
predict_test = [{
"aspect":
"food",
"sentiment":
"positive",
"sentence":
"good food in cute - though a bit dank - little hangout, but service terrible"
}, {
"aspect":
"service",
"sentiment":
"negative",
"sentence":
"good food in cute - though a bit dank - little hangout, but service terrible"
}, {
"aspect":
"service",
"sentiment":
"negative",
"sentence":
"good food in cute - though a bit dank - little hangout, but service terrible"
}]
predict_data = mydatasets.SemEval(text_field, as_field, sm_field,
predict_test)
train_data, dev_data, mixed_data = mydatasets.SemEval.splits_train_test(
text_field, as_field, sm_field, semeval_train, semeval_test)
text_field.build_vocab(train_data, dev_data)
as_field.build_vocab(train_data, dev_data)
sm_field.build_vocab(train_data, dev_data)
train_iter, test_iter, mixed_test_iter, predict_iter = data.Iterator.splits(
(train_data, dev_data, mixed_data, predict_data),
batch_sizes=(args.batch_size, len(dev_data), len(mixed_data),
len(predict_data)),
**kargs)
return train_iter, test_iter, mixed_test_iter, predict_iter