如何扩展3d numpy数组的值?

Question

假设我有一个3d数组(3x3x1)像这样:

[[[149]
[121]
[189]]
[[ 32]
[225]
[ 44]]
[[ 33]
[133]
[ 11]]]

我如何展开所有的值，使它们在最深的一个(3x3x3)中相同，像这样:

[[[149 149 149]
[121 121 121]
[189 189 189]]
[[ 32  32  32]
[225 225 225]
[ 44  44  44]]
[[ 33  33  33]
[133 133 133]
[ 11  11  11]]]

我已经试过了:

for i in range(len(array)):
for j in range(len(array[i])):
array[i][j] = np.array(list(array[i][j]) * 3)
print(array)

但是它给了我一个错误:

could not broadcast input array from shape (3,) into shape (1,)

为了一般化的目的，我如何用m x n x p形状格式实现这一点?

Answer 1

有多种选择。例如:np.repeat,np.tile,np.broadcast_to:

import numpy as np
arr = np.array([
[[149], [121], [189]],
[[ 32], [225], [ 44]],
[[ 33], [133], [ 11]]
])
out = np.repeat(arr, 3, axis=2)
# or
out = np.broadcast_to(arr, (3, 3, 3))
# or
out = np.tile(arr, (1, 1, 3))

选择最适合你的。

还请注意，形状为(3, 3, 1)的数组可以自动地作为形状为(3, 3, 3)的数组运行，而无需由于广播而手动重复。

Answer 2

也有np.c_[]方法:

import numpy as np
arr = np.array([
[[149], [121], [189]],
[[ 32], [225], [ 44]],
[[ 33], [133], [ 11]]
])
print(np.c_[arr, arr, arr])