本文整理汇总了Python中dragnn.python.digraph_ops.RootPotentialsFromTokens方法的典型用法代码示例。如果您正苦于以下问题：Python digraph_ops.RootPotentialsFromTokens方法的具体用法？Python digraph_ops.RootPotentialsFromTokens怎么用？Python digraph_ops.RootPotentialsFromTokens使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类dragnn.python.digraph_ops的用法示例。
在下文中一共展示了digraph_ops.RootPotentialsFromTokens方法的4个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: testRootPotentialsFromTokens
# 需要导入模块: from dragnn.python import digraph_ops [as 别名]
# 或者: from dragnn.python.digraph_ops import RootPotentialsFromTokens [as 别名]
def testRootPotentialsFromTokens(self):
with self.test_session():
root = tf.constant([1, 2], tf.float32)
tokens = tf.constant([[[4, 5, 6],
[5, 6, 7],
[6, 7, 8]],
[[6, 7, 8],
[5, 6, 7],
[4, 5, 6]]], tf.float32)
weights = tf.constant([[2, 3, 5],
[7, 11, 13]],
tf.float32)
roots = digraph_ops.RootPotentialsFromTokens(root, tokens, weights)
self.assertAllEqual(roots.eval(), [[375, 447, 519],
[519, 447, 375]]) 示例2: testRootPotentialsFromTokens
# 需要导入模块: from dragnn.python import digraph_ops [as 别名]
# 或者: from dragnn.python.digraph_ops import RootPotentialsFromTokens [as 别名]
def testRootPotentialsFromTokens(self):
with self.test_session():
root = tf.constant([1, 2], tf.float32)
tokens = tf.constant([[[4, 5, 6],
[5, 6, 7],
[6, 7, 8]],
[[6, 7, 8],
[5, 6, 7],
[4, 5, 6]]], tf.float32) # pyformat: disable
weights_arc = tf.constant([[2, 3, 5],
[7, 11, 13]],
tf.float32) # pyformat: disable
weights_source = tf.constant([11, 10], tf.float32)
roots = digraph_ops.RootPotentialsFromTokens(root, tokens, weights_arc,
weights_source)
self.assertAllEqual(roots.eval(), [[406, 478, 550],
[550, 478, 406]]) # pyformat: disable 示例3: create
# 需要导入模块: from dragnn.python import digraph_ops [as 别名]
# 或者: from dragnn.python.digraph_ops import RootPotentialsFromTokens [as 别名]
def create(self,
fixed_embeddings,
linked_embeddings,
context_tensor_arrays,
attention_tensor,
during_training,
stride=None):
"""Requires |stride|; otherwise see base class."""
check.NotNone(stride,
'BiaffineDigraphNetwork requires "stride" and must be called '
'in the bulk feature extractor component.')
# TODO(googleuser): Add dropout during training.
del during_training
# Retrieve (possibly averaged) weights.
weights_arc = self._component.get_variable('weights_arc')
weights_source = self._component.get_variable('weights_source')
root = self._component.get_variable('root')
# Extract the source and target token activations. Use |stride| to collapse
# batch and beam into a single dimension.
sources = network_units.lookup_named_tensor('sources', linked_embeddings)
targets = network_units.lookup_named_tensor('targets', linked_embeddings)
source_tokens_bxnxs = tf.reshape(sources.tensor,
[stride, -1, self._source_dim])
target_tokens_bxnxt = tf.reshape(targets.tensor,
[stride, -1, self._target_dim])
num_tokens = tf.shape(source_tokens_bxnxs)[1]
# Compute the arc, source, and root potentials.
arcs_bxnxn = digraph_ops.ArcPotentialsFromTokens(
source_tokens_bxnxs, target_tokens_bxnxt, weights_arc)
sources_bxnxn = digraph_ops.ArcSourcePotentialsFromTokens(
source_tokens_bxnxs, weights_source)
roots_bxn = digraph_ops.RootPotentialsFromTokens(
root, target_tokens_bxnxt, weights_arc)
# Combine them into a single matrix with the roots on the diagonal.
adjacency_bxnxn = digraph_ops.CombineArcAndRootPotentials(
arcs_bxnxn + sources_bxnxn, roots_bxn)
return [tf.reshape(adjacency_bxnxn, [-1, num_tokens])] 示例4: create
# 需要导入模块: from dragnn.python import digraph_ops [as 别名]
# 或者: from dragnn.python.digraph_ops import RootPotentialsFromTokens [as 别名]
def create(self,
fixed_embeddings,
linked_embeddings,
context_tensor_arrays,
attention_tensor,
during_training,
stride=None):
"""Requires |stride|; otherwise see base class."""
check.NotNone(stride,
'BiaffineDigraphNetwork requires "stride" and must be called '
'in the bulk feature extractor component.')
# TODO(googleuser): Add dropout during training.
del during_training
# Retrieve (possibly averaged) weights.
weights_arc = self._component.get_variable('weights_arc')
weights_source = self._component.get_variable('weights_source')
root = self._component.get_variable('root')
# Extract the source and target token activations. Use |stride| to collapse
# batch and beam into a single dimension.
sources = network_units.lookup_named_tensor('sources', linked_embeddings)
targets = network_units.lookup_named_tensor('targets', linked_embeddings)
source_tokens_bxnxs = tf.reshape(sources.tensor,
[stride, -1, self._source_dim])
target_tokens_bxnxt = tf.reshape(targets.tensor,
[stride, -1, self._target_dim])
num_tokens = tf.shape(source_tokens_bxnxs)[1]
# Compute the arc, source, and root potentials.
arcs_bxnxn = digraph_ops.ArcPotentialsFromTokens(
source_tokens_bxnxs, target_tokens_bxnxt, weights_arc)
sources_bxnxn = digraph_ops.ArcSourcePotentialsFromTokens(
source_tokens_bxnxs, weights_source)
roots_bxn = digraph_ops.RootPotentialsFromTokens(
root, target_tokens_bxnxt, weights_arc, weights_source)
# Combine them into a single matrix with the roots on the diagonal.
adjacency_bxnxn = digraph_ops.CombineArcAndRootPotentials(
arcs_bxnxn + sources_bxnxn, roots_bxn)
# The adjacency matrix currently has sources on rows and targets on columns,
# but we want targets on rows so that maximizing within a row corresponds to
# selecting sources for a given target.
adjacency_bxnxn = tf.matrix_transpose(adjacency_bxnxn)
return [tf.reshape(adjacency_bxnxn, [-1, num_tokens])]