本文整理汇总了Python中tensorflow.python.training.training_ops.resource_apply_gradient_descent方法的典型用法代码示例。如果您正苦于以下问题：Python training_ops.resource_apply_gradient_descent方法的具体用法？Python training_ops.resource_apply_gradient_descent怎么用？Python training_ops.resource_apply_gradient_descent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tensorflow.python.training.training_ops的用法示例。
在下文中一共展示了training_ops.resource_apply_gradient_descent方法的3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: _resource_apply_dense
# 需要导入模块: from tensorflow.python.training import training_ops [as 别名]
# 或者: from tensorflow.python.training.training_ops import resource_apply_gradient_descent [as 别名]
def _resource_apply_dense(self, grad, var):
momentum_buffer = self.get_slot(var, "momentum")
learning_rate = math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype)
momentum = math_ops.cast(self._momentum_tensor, var.dtype.base_dtype)
nu = math_ops.cast(self._nu_tensor, var.dtype.base_dtype)
momentum_op = training_ops.resource_apply_momentum(
var.handle,
momentum_buffer.handle,
nu * (1.0 - momentum) * learning_rate,
grad,
momentum,
use_locking=self._use_locking,
use_nesterov=False,
)
with ops.control_dependencies([momentum_op]):
gd_op = training_ops.resource_apply_gradient_descent(
var.handle, (1.0 - nu) * learning_rate, grad, use_locking=self._use_locking
)
return control_flow_ops.group(momentum_op, gd_op) 示例2: _resource_apply_dense
# 需要导入模块: from tensorflow.python.training import training_ops [as 别名]
# 或者: from tensorflow.python.training.training_ops import resource_apply_gradient_descent [as 别名]
def _resource_apply_dense(self, grad, handle):
return training_ops.resource_apply_gradient_descent(
handle.handle, math_ops.cast(self._learning_rate_tensor,
grad.dtype.base_dtype),
grad, use_locking=self._use_locking) 示例3: _resource_apply_dense
# 需要导入模块: from tensorflow.python.training import training_ops [as 别名]
# 或者: from tensorflow.python.training.training_ops import resource_apply_gradient_descent [as 别名]
def _resource_apply_dense(self, grad, handle):
return training_ops.resource_apply_gradient_descent(
handle, math_ops.cast(self._learning_rate_tensor,
grad.dtype.base_dtype),
grad, use_locking=self._use_locking)