本文整理汇总了Python中tensor2tensor.layers.common_attention.get_layer_timing_signal_sinusoid_1d方法的典型用法代码示例。如果您正苦于以下问题：Python common_attention.get_layer_timing_signal_sinusoid_1d方法的具体用法？Python common_attention.get_layer_timing_signal_sinusoid_1d怎么用？Python common_attention.get_layer_timing_signal_sinusoid_1d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tensor2tensor.layers.common_attention的用法示例。
在下文中一共展示了common_attention.get_layer_timing_signal_sinusoid_1d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: add_step_timing_signal
# 需要导入模块: from tensor2tensor.layers import common_attention [as 别名]
# 或者: from tensor2tensor.layers.common_attention import get_layer_timing_signal_sinusoid_1d [as 别名]
def add_step_timing_signal(x, step, hparams):
"""Add n-dimensional embedding as the step (vertical) timing signal.
Args:
x: a tensor with shape [batch, length, depth]
step: step
hparams: model hyper parameters
Returns:
a Tensor with the same shape as x.
"""
num_steps = (
hparams.act_max_steps
if hparams.recurrence_type == "act" else hparams.num_rec_steps)
channels = common_layers.shape_list(x)[-1]
if hparams.step_timing_signal_type == "learned":
signal = common_attention.get_layer_timing_signal_learned_1d(
channels, step, num_steps)
elif hparams.step_timing_signal_type == "sinusoid":
signal = common_attention.get_layer_timing_signal_sinusoid_1d(
channels, step, num_steps)
if hparams.add_or_concat_timing_signal == "add":
x_with_timing = x + signal
elif hparams.add_or_concat_timing_signal == "concat":
batch_size = common_layers.shape_list(x)[0]
length = common_layers.shape_list(x)[1]
signal_tiled = tf.tile(signal, [batch_size, length, 1])
x_with_timing = tf.concat((x, signal_tiled), axis=-1)
return x_with_timing 示例2: add_step_timing_signal
# 需要导入模块: from tensor2tensor.layers import common_attention [as 别名]
# 或者: from tensor2tensor.layers.common_attention import get_layer_timing_signal_sinusoid_1d [as 别名]
def add_step_timing_signal(x, step, hparams):
"""Add n-dimensional embedding as the step (vertical) timing signal.
Args:
x: a tensor with shape [batch, length, depth]
step: step
hparams: model hyper parameters
Returns:
a Tensor with the same shape as x.
"""
if hparams.recurrence_type == "act":
num_steps = hparams.act_max_steps
else:
num_steps = hparams.num_rec_steps
channels = common_layers.shape_list(x)[-1]
if hparams.step_timing_signal_type == "learned":
signal = common_attention.get_layer_timing_signal_learned_1d(
channels, step, num_steps)
elif hparams.step_timing_signal_type == "sinusoid":
signal = common_attention.get_layer_timing_signal_sinusoid_1d(
channels, step, num_steps)
if hparams.add_or_concat_timing_signal == "add":
x_with_timing = x + common_layers.cast_like(signal, x)
elif hparams.add_or_concat_timing_signal == "concat":
batch_size = common_layers.shape_list(x)[0]
length = common_layers.shape_list(x)[1]
signal_tiled = tf.tile(signal, [batch_size, length, 1])
x_with_timing = tf.concat((x, signal_tiled), axis=-1)
return x_with_timing