本文整理汇总了Python中dragnn.python.lexicon.create_lexicon_context方法的典型用法代码示例。如果您正苦于以下问题：Python lexicon.create_lexicon_context方法的具体用法？Python lexicon.create_lexicon_context怎么用？Python lexicon.create_lexicon_context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.lexicon的用法示例。
在下文中一共展示了lexicon.create_lexicon_context方法的2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: testCreateLexiconContext
# 需要导入模块: from dragnn.python import lexicon [as 别名]
# 或者: from dragnn.python.lexicon import create_lexicon_context [as 别名]
def testCreateLexiconContext(self):
expected_context = task_spec_pb2.TaskSpec()
text_format.Parse(_EXPECTED_CONTEXT, expected_context)
self.assertProtoEquals(
lexicon.create_lexicon_context('/tmp'), expected_context) 示例2: fill_from_resources
# 需要导入模块: from dragnn.python import lexicon [as 别名]
# 或者: from dragnn.python.lexicon import create_lexicon_context [as 别名]
def fill_from_resources(self, resource_path, tf_master=''):
"""Fills in feature sizes and vocabularies using SyntaxNet lexicon.
Must be called before the spec is ready to be used to build TensorFlow
graphs. Requires a SyntaxNet lexicon built at the resource_path. Using the
lexicon, this will call the SyntaxNet custom ops to return the number of
features and vocabulary sizes based on the FML specifications and the
lexicons. It will also compute the number of actions of the transition
system.
This will often CHECK-fail if the spec doesn't correspond to a valid
transition system or feature setup.
Args:
resource_path: Path to the lexicon.
tf_master: TensorFlow master executor (string, defaults to '' to use the
local instance).
"""
check.IsTrue(
self.spec.transition_system.registered_name,
'Set a transition system before calling fill_from_resources().')
context = lexicon.create_lexicon_context(resource_path)
for key, value in self.spec.transition_system.parameters.iteritems():
context.parameter.add(name=key, value=value)
context.parameter.add(
name='brain_parser_embedding_dims',
value=';'.join(
[str(x.embedding_dim) for x in self.spec.fixed_feature]))
context.parameter.add(
name='brain_parser_features',
value=';'.join([x.fml for x in self.spec.fixed_feature]))
context.parameter.add(
name='brain_parser_predicate_maps',
value=';'.join(['' for x in self.spec.fixed_feature]))
context.parameter.add(
name='brain_parser_embedding_names',
value=';'.join([x.name for x in self.spec.fixed_feature]))
context.parameter.add(
name='brain_parser_transition_system',
value=self.spec.transition_system.registered_name)
# Propagate information from SyntaxNet C++ backends into the DRAGNN
# self.spec.
with tf.Session(tf_master) as sess:
feature_sizes, domain_sizes, _, num_actions = sess.run(
gen_parser_ops.feature_size(task_context_str=str(context)))
self.spec.num_actions = int(num_actions)
for i in xrange(len(feature_sizes)):
self.spec.fixed_feature[i].size = int(feature_sizes[i])
self.spec.fixed_feature[i].vocabulary_size = int(domain_sizes[i])
for i in xrange(len(self.spec.linked_feature)):
self.spec.linked_feature[i].size = len(
self.spec.linked_feature[i].fml.split(' '))
for resource in context.input:
self.spec.resource.add(name=resource.name).part.add(
file_pattern=resource.part[0].file_pattern)