考虑以下形式的数组(仅作为示例):
[[ 0 1]
[ 2 3]
[ 4 5]
[ 6 7]
[ 8 9]
[10 11]
[12 13]
[14 15]
[16 17]]
它的形状是[9,2]。现在我想转换数组,使每列成为一个形状[3,3],如下所示:
[[ 0 6 12]
[ 2 8 14]
[ 4 10 16]]
[[ 1 7 13]
[ 3 9 15]
[ 5 11 17]]
最明显的(当然也是“非pythonic”)解决方案是使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...
答案 0 :(得分:55)
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]]])
答案 1 :(得分:0)
numpy具有完成此任务的出色工具(“ numpy.reshape”)link to reshape documentation
a = [[ 0 1]
[ 2 3]
[ 4 5]
[ 6 7]
[ 8 9]
[10 11]
[12 13]
[14 15]
[16 17]]
`numpy.reshape(a,(3,3))`
您还可以使用“ -1”技巧
`a = a.reshape(-1,3)`
“-1”是通配符,当第二维为3时,将使numpy算法决定要输入的数字
是的,这也可以:
a = a.reshape(3,-1)
:
a = a.reshape(-1,2)
什么都不做
:
a = a.reshape(-1,9)
会将形状更改为(2,9)
答案 2 :(得分:0)
有两种可能的结果重排(以下为@eumiro的示例)。 Einops
程序包提供了一种功能强大的注释,可以清楚地描述此类操作
>> 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]]])