Python 在 NumPy 中重塑数组
声明：本页面是StackOverFlow热门问题的中英对照翻译，遵循CC BY-SA 4.0协议，如果您需要使用它，必须同样遵循CC BY-SA许可，注明原文地址和作者信息，同时你必须将它归于原作者（不是我）：StackOverFlow
原文地址: http://stackoverflow.com/questions/14476415/
Warning: these are provided under cc-by-sa 4.0 license. You are free to use/share it, But you must attribute it to the original authors (not me):
StackOverFlow
Reshape an array in NumPy
提问by user1876864
Consider an array of the following form (just an example):
考虑以下形式的数组（只是一个例子）：
[[ 0 1]
[ 2 3]
[ 4 5]
[ 6 7]
[ 8 9]
[10 11]
[12 13]
[14 15]
[16 17]]
It's shape is [9,2]. Now I want to transform the array so that each column becomes a shape [3,3], like this:
它的形状是 [9,2]。现在我想转换数组，使每一列变成一个形状 [3,3]，如下所示：
[[ 0 6 12]
[ 2 8 14]
[ 4 10 16]]
[[ 1 7 13]
[ 3 9 15]
[ 5 11 17]]
The most obvious (and surely "non-pythonic") solution is to initialise an array of zeroes with the proper dimension and run two for-loops where it will be filled with data. I'm interested in a solution that is language-conform...
最明显的（当然也是“非 Pythonic”）解决方案是用适当的维度初始化一个零数组，并运行两个 for 循环，其中将填充数据。我对符合语言的解决方案感兴趣......
采纳答案by eumiro
a = np.arange(18).reshape(9,2)
b = a.reshape(3,3,2).swapaxes(0,2)
# a:
array([[ 0, 1],
[ 2, 3],
[ 4, 5],
[ 6, 7],
[ 8, 9],
[10, 11],
[12, 13],
[14, 15],
[16, 17]])
# b:
array([[[ 0, 6, 12],
[ 2, 8, 14],
[ 4, 10, 16]],
[[ 1, 7, 13],
[ 3, 9, 15],
[ 5, 11, 17]]])
回答by Or_K
numpy has a great tool for this task ("numpy.reshape") link to reshape documentation
numpy 有一个很好的工具来完成这个任务（“numpy.reshape”）链接到重塑文档
a = [[ 0 1]
[ 2 3]
[ 4 5]
[ 6 7]
[ 8 9]
[10 11]
[12 13]
[14 15]
[16 17]]
`numpy.reshape(a,(3,3))`
you can also use the "-1" trick
你也可以使用“-1”技巧
`a = a.reshape(-1,3)`
the "-1" is a wild card that will let the numpy algorithm decide on the number to input when the second dimension is 3
“-1”是一个通配符，当第二维为 3 时，它将让 numpy 算法决定要输入的数字
so yes.. this would also work:
a = a.reshape(3,-1)
所以是的..这也可以：
a = a.reshape(3,-1)
and this:
a = a.reshape(-1,2)would do nothing
而这：
a = a.reshape(-1,2)什么都不做
and this:
a = a.reshape(-1,9)would change the shape to (2,9)
这：
a = a.reshape(-1,9)将形状更改为 (2,9)
回答by Alleo
There are two possible result rearrangements (following example by @eumiro). Einopspackage provides a powerful notation to describe such operations non-ambigously
有两种可能的结果重新排列（以下示例来自@eumiro）。Einops包提供了一种强大的符号来明确地描述此类操作
>> a = np.arange(18).reshape(9,2)
# this version corresponds to eumiro's answer
>> einops.rearrange(a, '(x y) z -> z y x', x=3)
array([[[ 0, 6, 12],
[ 2, 8, 14],
[ 4, 10, 16]],
[[ 1, 7, 13],
[ 3, 9, 15],
[ 5, 11, 17]]])
# this has the same shape, but order of elements is different (note that each paer was trasnposed)
>> einops.rearrange(a, '(x y) z -> z x y', x=3)
array([[[ 0, 2, 4],
[ 6, 8, 10],
[12, 14, 16]],
[[ 1, 3, 5],
[ 7, 9, 11],
[13, 15, 17]]])
