numpy:矩阵数组移位/按索引插入 [英] Numpy: Matrix Array Shift / Insert by Index

问题描述

我有一个对象，在该对象中，我已将大量的for循环方法转换为一系列矢量化的numpy数组(速度提高了约50倍).现在，我正在尝试添加一种新方法，该方法需要处理一个numpy矩阵，然后根据矩阵中的数组索引移动"子数组内容(即插入值).我知道我可以使用for循环来完成此操作，但是我试图通过使用向量数学来提高速度来避免这种情况.

I have an object where I have converted a massive for-loop method into a series of vectorized numpy arrays (about 50x faster). Now I am trying to add a new method where I need to deal with a numpy matrix and then "shift" sub-array contents (i.e. insert values) based on the array index within the matrix. I know I can accomplish this with a for-loop, but am trying to avoid that with the speed-up gains achieved by using vector math instead.

我想知道是否有一种快速有效的方法来完成以下任务:

I was wondering if there is a fast and efficient way to accomplish the following:

import numpy as np

period = [1, 2, 3]

x = [1, 10, 100]
y = [.2, .4, .6]

z = np.outer(x,y)

print(z)

结果:

[[  0.2   0.4   0.6]
 [  2.    4.    6. ]
 [ 20.   40.   60. ]]

我想移动z中的行，以基于句点的方式将零的数量添加为z中的行索引，基本上是以下内容:

I'd like to shift the rows in z to add the number of zeros based on period as the row index in z, basically the following:

[[   0.0   0.2   0.4    0.6 ]
 [   0.0   0.0   2.0    4.0    6.0 ]
 [   0.0   0.0   0.0   20.0   40.0   60.0 ]]

最终，我希望在垂直/列轴(轴= 1)上求和.我需要一个如下的最终数组:

Ultimately, I'd be looking to sum on the vertical / column axis (axis=1). I'd need a final array like the following:

[   0.0   0.2   2.4   24.6   46.0   60.0]

推荐答案

您还可以先计算索引并立即分配:

You can also calculate the indices first and assign at once:

a = np.array(
    [[0.2 ,  0.4 ,  0.6],
     [2.,    4.,    6. ],
     [20.,   40.,   60. ]])

s0, s1 = a.shape
rows = np.repeat(np.arange(s0), s1).reshape(a.shape)
cols = (np.add.outer(np.arange(0, s0), np.arange(s1)) + 1)
res = np.zeros((s0, s0 + s1))
res[rows, cols] = a
np.sum(res,axis=0)

>>> np.sum(res,axis=0)
array([  0. ,   0.2,   2.4,  24.6,  46. ,  60. ])

这篇关于numpy:矩阵数组移位/按索引插入的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！