Python 在张量流中移动轴
我有两个张量。主张量如下所示:Python 在张量流中移动轴,python,tensorflow,Python,Tensorflow,我有两个张量。主张量如下所示: array([[[ 298, 1217, 298, 1217], [ 298, 1217, 298, 1217], [ 298, 1217, 298, 1217], [ 298, 1217, 298, 1217], [ 298, 1217, 298, 1217], [ 298, 1217, 298, 1217], [ 298, 1217, 298, 121
array([[[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217]],
[[ 450, 607, 493, 662],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 950, 1277, 1028, 1335],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]]], dtype=int32)
我想根据以下张量移动这个张量:
array([0, 2, 5], dtype=int32)
上面的张量包含我们希望当前轴移动到的轴
最后一个张量应如下所示:
array([[[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 450, 607, 493, 662],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 950, 1277, 1028, 1335],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]]], dtype=int32)
您可以使用tensorflow分散功能来实现这一点 定义您的
输入
张量:
input = tf.constant([[[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217]],
[[ 450, 607, 493, 662],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 950, 1277, 1028, 1335],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]]])
因为我们只对第零维上的前3个元素感兴趣,所以让我们将其切成一个新的张量:
sliced_input = tf.slice(input, [0, 0, 0], [3, -1, -1])
定义您的目标索引
:
indices = tf.constant([[0], [2], [5]])
定义目标输出的形状
,此处与您的输入
形状相同:
shape = tf.shape(input)
现在,使用分散功能获取输出
:
output = tf.scatter_nd(indices, sliced_input, shape)
array([[[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 450, 607, 493, 662],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 950, 1277, 1028, 1335],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]]], dtype=int32)
输出
:
output = tf.scatter_nd(indices, sliced_input, shape)
array([[[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217],
[ 298, 1217, 298, 1217]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 450, 607, 493, 662],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 950, 1277, 1028, 1335],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]],
[[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0],
[ 0, 0, 0, 0]]], dtype=int32)
如果我理解正确,您正在寻找offeltoffel的numpy.moveaxis
(),但我想要tensorflow中的这个。对不起!然后是tf.transpose(张量数组,perm=[0,2,5])
。看看这个:从文本中看不太清楚,但您想要的是,对于输入[0,2,5]
,交换轴0中位置2和5的内容?@jdehesa在位置2,我想要1的内容;在第5位,我想要3的内容。