stack

paddle.fluid.layers. stack ( x, axis=0 ) [源代码]

该OP沿 axis 轴对输入 x 进行堆叠操作。

  • 例1:

输入:
    x[0].shape = [1, 2]
    x[0].data = [ [1.0 , 2.0 ] ]
    x[1].shape = [1, 2]
    x[1].data = [ [3.0 , 4.0 ] ]
    x[2].shape = [1, 2]
    x[2].data = [ [5.0 , 6.0 ] ]

参数:
    axis = 0 #沿着第0维对输入x进行堆叠操作。

输出:
    Out.shape = [3, 1, 2]
    Out.data = [ [ [1.0, 2.0] ],
                [ [3.0, 4.0] ],
                [ [5.0, 6.0] ] ]
  • 例2:

输入:
    x[0].shape = [1, 2]
    x[0].data = [ [1.0 , 2.0 ] ]
    x[1].shape = [1, 2]
    x[1].data = [ [3.0 , 4.0 ] ]
    x[2].shape = [1, 2]
    x[2].data = [ [5.0 , 6.0 ] ]

参数:
    axis = 1 or axis = -2 #沿着第1维对输入进行堆叠操作。

输出:
    Out.shape = [1, 3, 2]
    Out.data = [ [ [1.0, 2.0]
                  [3.0, 4.0]
                  [5.0, 6.0] ] ]

参数

  • x (list(Variable)|tuple(Variable)) – 输入 x 可以是单个Tensor,或是多个Tensor组成的列表。如果 x 是一个列表,那么这些Tensor的维度必须相同。假设输入是N维Tensor \([d_0,d_1,...,d_{n−1}]\),则输出变量的维度为N+1维 \([d_0,d_1,...d_{axis-1},len(x),d_{axis}...,d_{n−1}]\)。支持的数据类型:float32,float64,int32,int64。

  • axis (int,可选) – 指定对输入Tensor进行堆叠运算的轴,有效 axis 的范围是:\([-(R+1), R+1)\),R是输入中第一个Tensor的rank。如果 axis < 0,则 \(axis=axis+rank(x[0])+1\) 。axis默认值为0。

返回

堆叠运算后的Tensor,数据类型与输入Tensor相同。输出维度等于 \(rank(x[0])+1\) 维。

返回类型

Variable

代码示例

import paddle.fluid as fluid
import paddle.fluid.layers as layers
# set batch size=None
x1 = fluid.data(name='x1', shape=[None, 1, 2], dtype='int32')
x2 = fluid.data(name='x2', shape=[None, 1, 2], dtype='int32')
# stack Tensor list
data = layers.stack([x1,x2]) # stack according to axis 0, data.shape=[2, None, 1, 2]

data = layers.stack([x1,x2], axis=1) # stack according to axis 1, data.shape=[None, 2, 1, 2]