我有一个 [n x n] 矩阵,其中包含属于不同组的值,以及一个 [1 x n] 向量,定义每个元素属于哪个组。 (n通常~1E4,在本例中n=4)
我想计算一个将属于同一组的所有元素相加得到的矩阵。
我使用 np.where() 来计算每个组的元素所在的索引。 当我使用计算出的索引时,我没有获得预期的元素,因为我选择的是位置对而不是范围(我习惯了 Matlab,我可以简单地选择 M(idx1,idx2) )。
import numpy as np
n=4
M = np.random.rand(n,n)
print(M)
# This vector defines to which group each element belong
belongToGroup = np.array([0, 1, 0, 2])
nGroups=np.max(belongToGroup);
# Calculate a matrix obtained by summing elements belonging to the same group
M_sum = np.zeros((nGroups+1,nGroups+1))
for g1 in range(nGroups+1):
idxG1 = np.where(belongToGroup==g1)
for g2 in range(nGroups+1):
idxG2 = np.where(belongToGroup==g2)
print('g1 = ' + str(g1))
print('g2 = ' + str(g2))
print(idxG1[0])
print(idxG2[0])
print(M[idxG1[0],idxG2[0]])
print(np.sum(M[idxG1[0],idxG2[0]]))
M_sum[g1,g2]=np.sum(M[idxG1[0],idxG2[0]])
print('')
print('Example of the problem:')
print('Elements I would like to sum to obtain M_sum[0,0]')
print(M[0:2,0:2])
print('Elements that are summed instead')
print(M[[0,1],[0,1]])
问题示例: 在上面的示例中,元素 M_sum[0,0] 应该是 M[0,0]、M[0,1]、M[1,0] 和 M[1,1] 的和 相反,它被计算为 M[0,0] 和 M[1,1] 之和
最佳答案
在 MATLAB 中,使用 2 个列表(实际上是矩阵)进行索引会选择一个 block 。另一方面,numpy 尝试相互广播索引数组,并返回选定的点。其行为类似于 sub2ind
在 MATLAB 中的行为。
In [971]: arr = np.arange(16).reshape(4,4)
In [972]: arr
Out[972]:
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
In [973]: i1, i2 = np.array([0,2,3]), np.array([1,2,0])
使用 2 个相同大小的一维数组进行索引:
In [974]: arr[i1,i2]
Out[974]: array([ 1, 10, 12])
这实际上返回[arr[0,1], arr[2,2], arr[3,0]]
,每个匹配索引点一个元素。
但是,如果我将一个索引转换为“列向量”,它会从行中进行选择,而 i2
会从列中进行选择。
In [975]: arr[i1[:,None], i2]
Out[975]:
array([[ 1, 2, 0],
[ 9, 10, 8],
[13, 14, 12]])
MATLAB 使 block 索引变得容易,而单独访问则更困难。在 numpy 中, block 访问有点困难,尽管底层机制是相同的。
在您的示例中,i1[0]
和 i2[0]
可以是如下数组:
array([0, 2]), array([3])
(2,) (1,)
形状 (1,) 数组也可以与 (2,) 或 (2,1) 数组一起广播。如果 is[0]
是 np.array([0,1,2])
(一个无法与以下对象配对的 (3,) 数组),您的代码将会失败(2,) 数组。但对于 (2,1),它会生成 (2,3) block 。
关于python - 在 Python 中对矩阵的选定元素求和,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/55480343/