def __call__(self, batch):
myid, text_input, humor_rating,humor_contr = map(list, zip(*batch))
input_lengths = torch.tensor(
[len(s) for s in text_input], device=self.device)
# attention mask
max_length = max(input_lengths)
inputs_pad_mask = pad_mask(input_lengths, max_length=max_length,
device=self.device)
# Pad inputs and targets
padded_inputs = (
pad_sequence(text_input, batch_first=True,
padding_value=self.pad_indx)
.to(self.device))
humor_rating = mktensor(humor_rating, dtype=torch.float)
humor_contr = mktensor(humor_contr, dtype=torch.long)
return myid, padded_inputs,inputs_pad_mask, humor_rating, humor_contr
def __getitem__(self, index):
text, humor = self.data[index]
is_humor = self.label2idx(humor)
if not self.transforms:
text = self.tokenizer(text)
text = mktensor(text['input_ids'], dtype=torch.long)
else:
for t in self.transforms:
text = t(text)
is_humor = int(is_humor)
return text, is_humor
def __getitem__(self, index):
if self.splitname == 'train':
myid, text, is_humor, humor_rating, humor_controversy, offense_rating\
= \
self.data[index]
else:
myid, text = self.data[index]
humor_rating = None
humor_controversy = None
if not self.transforms:
text = self.tokenizer(text)
text = mktensor(text['input_ids'], dtype=torch.long)
else:
for t in self.transforms:
text = t(text)
myid = int(myid)
if self.splitname == 'train':
humor_rating = float(humor_rating)
humor_controversy = int(humor_controversy)
return myid, text, humor_rating, humor_controversy
