在numpy数组中非对齐插入未对齐的元素

Question

我正在使用numpy 1.9来处理一组数组.假设我有类似的东西我有两个2D阵列A和B和一个1-d阵列C,其看起来像这样:

>>> A
array([[ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.],
       [ 1.,  1.,  1.,  1.,  1.]])
>>> B
array([[-1., -1., -1., -1., -1.],
       [-1., -1., -1., -1., -1.],
       [-1., -1., -1., -1., -1.],
       [-1., -1., -1., -1., -1.],
       [-1., -1., -1., -1., -1.]])
>>> C
array([1, 3, 2, 4, 0])

我的目标是根据C在A中插入所有元素.更具体地说,如果位置0处的C具有1,则应在A [0,1]之后插入B [0,1].

这是预期的结果:

array([[ 1,  1, -1,  1,  1,  1],
       [ 1,  1,  1,  1, -1,  1],
       [ 1,  1,  1, -1,  1,  1],
       [ 1,  1,  1,  1,  1, -1],
       [ 1, -1,  1,  1,  1,  1]])

我试图像这样实现它,但它不是很快:

for i in xrange(size(C, 0)):
    j = C[i]
    A[i, :] = numpy.insert(A[i], j, B[i, j])

有办法让它更快吗？(通过单个numpy操作,如面具或类似的东西)

Answer 1

一个令人讨厌的单线程怎么样？

一,数据; 数组与你的形状相同,但我使用整数来使示例更容易阅读.

In [81]: A
Out[81]: 
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, 24]])

In [82]: B
Out[82]: 
array([[   0,  100,  200,  300,  400],
       [ 500,  600,  700,  800,  900],
       [1000, 1100, 1200, 1300, 1400],
       [1500, 1600, 1700, 1800, 1900],
       [2000, 2100, 2200, 2300, 2400]])

In [83]: C
Out[83]: array([1, 3, 2, 4, 0])

这是令人讨厌的单线:

In [84]: np.insert(A.ravel(), np.ravel_multi_index((range(A.shape[0]), C), A.shape) + 1, B[range(B.shape[0]), C]).reshape(A.shape[0], A.shape[1]+1)
Out[84]: 
array([[   0,    1,  100,    2,    3,    4],
       [   5,    6,    7,    8,  800,    9],
       [  10,   11,   12, 1200,   13,   14],
       [  15,   16,   17,   18,   19, 1900],
       [  20, 2000,   21,   22,   23,   24]])

这是分解版本:

A.ravel()变平A为一维数组,我称之为F:

In [87]: F = A.ravel()

In [88]: F
Out[88]: 
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, 24])

(编辑:事实证明这是第一步 - 扁平化 - A没有必要.正如@hpaulj在他的回答中指出的那样,np.insert默认会压扁数组.)

np.ravel_multi_index用于将所需的2-d位置转换为指数到扁平化阵列中.的+ 1,因为你要插入的元素在年底是必要的后给出的指标C:

In [89]: insert_indices = np.ravel_multi_index((range(A.shape[0]), C), A.shape) + 1

In [90]: insert_indices
Out[90]: array([ 2,  9, 13, 20, 21])

B[range(B.shape[0]), C]拉出所需的值B:

In [91]: values = B[range(B.shape[0]), C]

In [92]: values
Out[92]: array([ 100,  800, 1200, 1900, 2000])

np.insert 实际插入并创建一个新数组:

In [93]: np.insert(F, insert_indices, values)
Out[93]: 
array([   0,    1,  100,    2,    3,    4,    5,    6,    7,    8,  800,
          9,   10,   11,   12, 1200,   13,   14,   15,   16,   17,   18,
         19, 1900,   20, 2000,   21,   22,   23,   24])

现在只需重塑一下即可获得最终结果:

In [94]: np.insert(F, insert_indices, values).reshape(A.shape[0], A.shape[1]+1)
Out[94]: 
array([[   0,    1,  100,    2,    3,    4],
       [   5,    6,    7,    8,  800,    9],
       [  10,   11,   12, 1200,   13,   14],
       [  15,   16,   17,   18,   19, 1900],
       [  20, 2000,   21,   22,   23,   24]])