本文整理匯總了Python中dragnn.python.network_units.get_input_tensor方法的典型用法代碼示例。如果您正苦於以下問題：Python network_units.get_input_tensor方法的具體用法？Python network_units.get_input_tensor怎麽用？Python network_units.get_input_tensor使用的例子？那麽, 這裏精選的方法代碼示例或許可以為您提供幫助。您也可以進一步了解該方法所在類dragnn.python.network_units的用法示例。
在下文中一共展示了network_units.get_input_tensor方法的2個代碼示例，這些例子默認根據受歡迎程度排序。您可以為喜歡或者感覺有用的代碼點讚，您的評價將有助於係統推薦出更棒的Python代碼示例。
示例1: create
# 需要導入模塊: from dragnn.python import network_units [as 別名]
# 或者: from dragnn.python.network_units import get_input_tensor [as 別名]
def create(self,
fixed_embeddings,
linked_embeddings,
context_tensor_arrays,
attention_tensor,
during_training,
stride=None):
"""See base class."""
# NB: This cell pulls the lstm's h and c vectors from context_tensor_arrays
# instead of through linked features.
check.Eq(
len(context_tensor_arrays), 2 * len(self._hidden_layer_sizes),
'require two context tensors per hidden layer')
# Rearrange the context tensors into a tuple of LSTM sub-states.
length = context_tensor_arrays[0].size()
substates = []
for index, num_units in enumerate(self._hidden_layer_sizes):
state_c = context_tensor_arrays[2 * index].read(length - 1)
state_h = context_tensor_arrays[2 * index + 1].read(length - 1)
# Fix shapes that for some reason are not set properly for an unknown
# reason. TODO(googleuser): Why are the shapes not set?
state_c.set_shape([tf.Dimension(None), num_units])
state_h.set_shape([tf.Dimension(None), num_units])
substates.append(tf.contrib.rnn.LSTMStateTuple(state_c, state_h))
state = tuple(substates)
input_tensor = dragnn.get_input_tensor(fixed_embeddings, linked_embeddings)
cell = self._train_cell if during_training else self._inference_cell
def _cell_closure(scope):
"""Applies the LSTM cell to the current inputs and state."""
return cell(input_tensor, state, scope)
unused_h, state = self._apply_with_captured_variables(_cell_closure)
# Return tensors to be put into the tensor arrays / used to compute
# objective.
output_tensors = []
for new_substate in state:
new_c, new_h = new_substate
output_tensors.append(new_c)
output_tensors.append(new_h)
return self._append_base_layers(output_tensors) 示例2: create
# 需要導入模塊: from dragnn.python import network_units [as 別名]
# 或者: from dragnn.python.network_units import get_input_tensor [as 別名]
def create(self,
fixed_embeddings,
linked_embeddings,
context_tensor_arrays,
attention_tensor,
during_training,
stride=None):
"""See base class."""
# NB: This cell pulls the lstm's h and c vectors from context_tensor_arrays
# instead of through linked features.
check.Eq(
len(context_tensor_arrays), 2 * len(self._hidden_layer_sizes),
'require two context tensors per hidden layer')
# Rearrange the context tensors into a tuple of LSTM sub-states.
length = context_tensor_arrays[0].size()
substates = []
for index, num_units in enumerate(self._hidden_layer_sizes):
state_c = context_tensor_arrays[2 * index].read(length - 1)
state_h = context_tensor_arrays[2 * index + 1].read(length - 1)
# Fix shapes that for some reason are not set properly for an unknown
# reason. TODO(googleuser): Why are the shapes not set?
state_c.set_shape([tf.Dimension(None), num_units])
state_h.set_shape([tf.Dimension(None), num_units])
substates.append(tf.contrib.rnn.LSTMStateTuple(state_c, state_h))
state = tuple(substates)
input_tensor = dragnn.get_input_tensor(fixed_embeddings, linked_embeddings)
cell = self._train_cell if during_training else self._inference_cell
def _cell_closure(scope):
"""Applies the LSTM cell to the current inputs and state."""
return cell(input_tensor, state, scope=scope)
unused_h, state = self._apply_with_captured_variables(_cell_closure)
# Return tensors to be put into the tensor arrays / used to compute
# objective.
output_tensors = []
for new_substate in state:
new_c, new_h = new_substate
output_tensors.append(new_c)
output_tensors.append(new_h)
return self._append_base_layers(output_tensors)