def forward(self, x):
"""
Calculate the output.
Parameters
----------
x : ``torch.LongTensor``, required.
the input tensor, of shape (seq_len, batch_size, input_dim).
Returns
----------
output: ``torch.FloatTensor``.
The output of RNNs.
"""
if self.droprate > 0:
new_x = F.dropout(x, p=self.droprate, training=self.training)
else:
new_x = x
out, new_hidden = self.layer(new_x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
out = out.contiguous()
return torch.cat([x, out], 2)
def forward(self, x):
# set_trace()
out, new_hidden = self.layer(x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
if self.droprate > 0:
out = F.dropout(out, p=self.droprate, training=self.training)
return out
def forward(self, x):
if self.droprate > 0:
new_x = F.dropout(x, p=self.droprate, training=self.training)
else:
new_x = x
out, new_hidden = self.layer(new_x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
out = out.contiguous()
return torch.cat([x, out], 2)
def forward(self, x):
out = 0
# n_w = F.softmax(self.unit_weight, dim=0)
for ind in range(self.unit_number):
nout, new_hidden = self.unit_list[ind](x[ind],
self.hidden_list[ind])
self.hidden_list[ind] = utils.repackage_hidden(new_hidden)
out = out + nout
# out = out + n_w[ind] * self.unit_number * nout
if self.droprate > 0:
out = F.dropout(out, p=self.droprate, training=self.training)
x.append(out)
return x
def forward(self, x, p_out):
if self.droprate > 0:
new_x = F.dropout(x, p=self.droprate, training=self.training)
else:
new_x = x
out, new_hidden = self.layer(new_x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
out = out.contiguous()
if self.training and random.uniform(0, 1) < self.layer_drop:
deep_out = torch.autograd.Variable( torch.zeros(x.size(0), x.size(1), self.increase_rate) ).cuda()
else:
deep_out = out
o_out = torch.cat([p_out, out], 2)
d_out = torch.cat([x, deep_out], 2)
return d_out, o_out
def forward(self, x):
"""
Calculate the output.
Parameters
----------
x : ``torch.LongTensor``, required.
the input tensor, of shape (seq_len, batch_size, input_dim).
Returns
----------
output: ``torch.FloatTensor``.
The output of RNNs.
"""
out, new_hidden = self.layer(x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
if self.droprate > 0:
out = F.dropout(out, p=self.droprate, training=self.training)
return out
def forward(self, x, p_out):
"""
Calculate the output.
Parameters
----------
x : ``torch.LongTensor``, required.
the input tensor, of shape (seq_len, batch_size, input_dim).
p_out : ``torch.LongTensor``, required.
the final output tensor for the softmax, of shape (seq_len, batch_size, input_dim).
Returns
----------
out: ``torch.FloatTensor``.
The undropped outputs of RNNs to the softmax.
p_out: ``torch.FloatTensor``.
The dropped outputs of RNNs to the next_layer.
"""
if self.droprate > 0:
new_x = F.dropout(x, p=self.droprate, training=self.training)
else:
new_x = x
out, new_hidden = self.layer(new_x, self.hidden_state)
self.hidden_state = utils.repackage_hidden(new_hidden)
out = out.contiguous()
if self.training and random.uniform(0, 1) < self.layer_drop:
deep_out = torch.autograd.Variable(
torch.zeros(x.size(0), x.size(1), self.increase_rate)).cuda()
else:
deep_out = out
o_out = torch.cat([p_out, out], 2)
d_out = torch.cat([x, deep_out], 2)
return d_out, o_out