我可以在不复制数据的情况下将非相邻维度合并到 NumPy 数组中吗？

Question

我想将 3-D NumPy 数组的第一个和最后一个维度合并为一个维度，而不复制数据：

import numpy as np

data = np.empty((3, 4, 5))
data = data.transpose([0, 2, 1])

try:
  # this fails, indicating that it is not possible:
  # AttributeError: incompatible shape for a non-contiguous array
  data.shape = (-1, 4)
except AttributeError:
  # this creates a copy of the data:
  data = data.reshape((-1, 4))

这可能吗？

Answer 1

In [55]: arr = np.arange(24).reshape(2,3,4)                                                               
In [56]: arr1 = arr.transpose(2,1,0)                                                                      
In [57]: arr                                                                                              
Out[57]: 
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]]])
In [58]: arr1                                                                                             
Out[58]: 
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]]])

查看值在 1d 数据缓冲区中的布局方式：

In [59]: arr.ravel()                                                                                      
Out[59]: 
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])

比较转置后的顺序：

In [60]: arr1.ravel()                                                                                     
Out[60]: 
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])

如果拼出来的值顺序不同，就无法避免复制。

reshape 有这条注释：

You can think of reshaping as first raveling the array (using the given index order), then inserting the elements from the raveled array into the new array using the same kind of index ordering as was used for the raveling.

In [63]: arr1.reshape(-1,2)                                                                               
Out[63]: 
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]])

我可以在不复制数据的情况下将非相邻维度合并到 NumPy 数组中吗？

Can I combine non-adjacent dimensions in a NumPy array without copying data?

python

memory

arrays

numpy

reshape