【Numpy】类转置操作

前言

本篇总结、介绍Numpy数组（ndarray）的基本操作之一——类转置操作 [1]。

1. moveaxis

numpy.moveaxis(a, source, destination)：将数组（a）的轴（source）移动到新的位置（destination），其他轴维持原有顺序不变

• a：类数组
• source：整数或整数序列。待移动的轴的原始位置
• destination：整数或整数序列。待移动的轴要移动到的位置

>>> arr = np.arange(24).reshape(2,3,4)
>>> arr
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11]],

       [[12, 13, 14, 15],
        [16, 17, 18, 19],
        [20, 21, 22, 23]]])

>>> np.moveaxis(arr,0,2)
array([[[ 0, 12],
        [ 1, 13],
        [ 2, 14],
        [ 3, 15]],

       [[ 4, 16],
        [ 5, 17],
        [ 6, 18],
        [ 7, 19]],

       [[ 8, 20],
        [ 9, 21],
        [10, 22],
        [11, 23]]])
>>> np.moveaxis(arr,0,2).shape
(3, 4, 2)

>>> np.moveaxis(arr,0,1)
array([[[ 0,  1,  2,  3],
        [12, 13, 14, 15]],

       [[ 4,  5,  6,  7],
        [16, 17, 18, 19]],

       [[ 8,  9, 10, 11],
        [20, 21, 22, 23]]])
>>> np.moveaxis(arr,0,1).shape
(3, 2, 4)

>>> np.moveaxis(arr,1,2)
array([[[ 0,  4,  8],
        [ 1,  5,  9],
        [ 2,  6, 10],
        [ 3,  7, 11]],

       [[12, 16, 20],
        [13, 17, 21],
        [14, 18, 22],
        [15, 19, 23]]])
>>> np.moveaxis(arr,1,2).shape
(2, 4, 3)

>>> np.moveaxis(arr,(0,1),(1,2))
array([[[ 0,  4,  8],
        [12, 16, 20]],

       [[ 1,  5,  9],
        [13, 17, 21]],

       [[ 2,  6, 10],
        [14, 18, 22]],

       [[ 3,  7, 11],
        [15, 19, 23]]])
>>> np.moveaxis(arr,(0,1),(1,2)).shape
(4, 2, 3)

2. rollaxis

numpy.rollaxis(a, axis, start=0)：向后滚动特定的轴到一个特定位置

• a：类数组。输入数组
• axis：整数。要向后滚动的轴，其它轴的相对位置不会改变
• start：整数，可选参数。指示上述轴要滚动到的特定位置
  • 如果start <= axis：axis指定的轴滚动到start指定的位置
  • 如果start > axis：axis指定的轴滚动到start指定的前一个位置

>>> arr = np.arange(24).reshape(1,2,3,4)
>>> arr
array([[[[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11]],

        [[12, 13, 14, 15],
         [16, 17, 18, 19],
         [20, 21, 22, 23]]]])

>>> np.rollaxis(arr,1,3)
array([[[[ 0,  1,  2,  3],
         [12, 13, 14, 15]],

        [[ 4,  5,  6,  7],
         [16, 17, 18, 19]],

        [[ 8,  9, 10, 11],
         [20, 21, 22, 23]]]])
>>> np.rollaxis(arr,1,3).shape
(1, 3, 2, 4)

>>> np.rollaxis(arr,3,1)
array([[[[ 0,  4,  8],
         [12, 16, 20]],

        [[ 1,  5,  9],
         [13, 17, 21]],

        [[ 2,  6, 10],
         [14, 18, 22]],

        [[ 3,  7, 11],
         [15, 19, 23]]]])
>>> np.rollaxis(arr,3,1).shape
(1, 4, 2, 3)

3. swapaxes

numpy.swapaxes(a, axis1, axis2)：交换数组的两个轴

• a：类数组。输入数组
• axis1：整数。第一个轴
• axis2：整数。第二个轴

>>> arr = np.arange(24).reshape(1,2,3,4)
>>> arr
array([[[[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11]],

        [[12, 13, 14, 15],
         [16, 17, 18, 19],
         [20, 21, 22, 23]]]])

>>> np.swapaxes(arr,0,1)
array([[[[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11]]],


       [[[12, 13, 14, 15],
         [16, 17, 18, 19],
         [20, 21, 22, 23]]]])

>>> np.swapaxes(arr,0,2)
array([[[[ 0,  1,  2,  3]],

        [[12, 13, 14, 15]]],


       [[[ 4,  5,  6,  7]],

        [[16, 17, 18, 19]]],


       [[[ 8,  9, 10, 11]],

        [[20, 21, 22, 23]]]])

>>> np.swapaxes(arr,0,3)
array([[[[ 0],
         [ 4],
         [ 8]],

        [[12],
         [16],
         [20]]],


       [[[ 1],
         [ 5],
         [ 9]],

        [[13],
         [17],
         [21]]],


       [[[ 2],
         [ 6],
         [10]],

        [[14],
         [18],
         [22]]],


       [[[ 3],
         [ 7],
         [11]],

        [[15],
         [19],
         [23]]]])

4. transpose

numpy.transpose(a, axes=None)：反转或置换数组的轴，返回修改后的数组

• a：类数组。输入数组
  • 一维数组：返回原始数组形状未变的视图
• axes：整数序列，可选参数
  • 如果指定：axes应该是[0,1,…,a.ndim-1]的一个排列，表示置换后数组的每个轴对应原始数组的哪个轴。例如，[2,0,1]表示置换后数组的第一个轴对应原始数组的第三个轴，置换后数组的第二个轴对应原始数组的第一个轴，置换后数组的第三个轴对应原始数组的第二个轴。假设数组a的形状为(2,3,4)，那么numpy.transpose(a,axes=[2,0,1])返回的数组形状为(4,2,3)
  • 如果未指定：axes默认为[a.ndim-1,…,1,0]，表示翻转数组的轴的顺序

>>> arr = np.arange(24).reshape(2,3,4)
>>> arr
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11]],

       [[12, 13, 14, 15],
        [16, 17, 18, 19],
        [20, 21, 22, 23]]])

>>> np.transpose(arr)
array([[[ 0, 12],
        [ 4, 16],
        [ 8, 20]],

       [[ 1, 13],
        [ 5, 17],
        [ 9, 21]],

       [[ 2, 14],
        [ 6, 18],
        [10, 22]],

       [[ 3, 15],
        [ 7, 19],
        [11, 23]]])
>>> np.transpose(arr).shape
(4, 3, 2)

>>> np.transpose(arr,axes=[2,0,1])
array([[[ 0,  4,  8],
        [12, 16, 20]],

       [[ 1,  5,  9],
        [13, 17, 21]],

       [[ 2,  6, 10],
        [14, 18, 22]],

       [[ 3,  7, 11],
        [15, 19, 23]]])
>>> np.transpose(arr,axes=[2,0,1]).shape
(4, 2, 3)

Reference

[1]: https://numpy.org/doc/stable/reference/routines.array-manipulation.html

[2]: https://zhuanlan.zhihu.com/p/162874970