def double_linear_logits(args,
size,
bias,
bias_start=0.0,
mask=None,
wd=0.0,
input_drop_prob=0.0,
is_train=None):
first = torch.tanh(
linear(args,
size,
bias,
bias_start=bias_start,
wd=wd,
input_drop_prob=input_drop_prob,
is_train=is_train))
second = linear(first,
1,
bias,
bias_start=bias_start,
squeeze=True,
wd=wd,
input_drop_prob=input_drop_prob,
is_train=is_train)
if mask is not None:
second = exp_mask(second, mask)
return second
def softmax(logits, mask=None):
if mask is not None:
logits = exp_mask(logits, mask)
flat_logits = flatten(logits, 1)
flat_out = F.softmax(flat_logits)
out = reconstruct(flat_out, logits, 1)
return out
def linear_logits(linear_layer,
args,
bias,
bias_start=0.0,
mask=None,
wd=0.0,
input_drop_prob=0.0,
is_train=None):
logits = linear(linear_layer,
args,
1,
bias,
bias_start=bias_start,
squeeze=True,
wd=wd,
input_drop_prob=input_drop_prob,
is_train=is_train)
if mask is not None:
logits = exp_mask(logits, mask)
return logits
def sum_logits(args, mask=None):
rank = len(args[0].size())
logits = sum(torch.sum(arg, rank - 1) for arg in args)
if mask is not None:
logits = exp_mask(logits, mask)
return logits