在Python中建立三维阵列，以取代优化循环 [英] Building 3D arrays in Python to replace loops for optimization

问题描述

我想更好地了解蟒蛇的优化，所以这是一个虚拟的情况，但希望勾勒我的想法......

I'm trying to better understand python optimization so this is a dummy case, but hopefully outlines my idea...

说我有一个函数，它接受两个变量：

Say I have a function which takes two variables:

def func(param1, param2):
    return some_func(param1) + some_const*(param2/2)

和我有参数1和参数（不同长度的）阵列，在此我要进行评估的功能，（some_func是参数1的任意功能）例如。

and I have arrays for param1 and param2 (of different lengths), at which I want the function to be evaluated, (some_func is an arbitrary function of param1) e.g.

param1 = np.array((1,2,3,4,5))
param2 = np.array((5,2,3,1,9, 9, 10))

我可以通过做评估对所有参数空间：

I can evaluate over all parameter space by doing:

result = []
for p in param1:
    result.append(func(p, param2))
result = np.asarray(result)

然而，在Python环比数组操作慢。因此，我不知道有没有办法实现了三维数组，它包含两个参数1和参数数组FUNC的所有值的结果？

However, loops in Python are slower than array operations. Therefore, I wonder is there a way to achieve a 3D array which contains the results of func for all values in both param1 and param2 arrays?

推荐答案

您例如：

In [198]: result=[]
In [199]: for p in param1:
   .....:     result.append(p+3*(param2/2))
In [200]: result=np.array(result)

同样的结果通过广播（和 np.newaxis ）

In [197]: param1[:,None] + 3*(param2[None,:]/2)
Out[197]: 
array([[ 7,  4,  4,  1, 13, 13, 16],
       [ 8,  5,  5,  2, 14, 14, 17],
       [ 9,  6,  6,  3, 15, 15, 18],
       [10,  7,  7,  4, 16, 16, 19],
       [11,  8,  8,  5, 17, 17, 20]])

some_func 的细节将决定是否使用 some_func（参数1 [：，无]）或 some_func（参数1）:,无] 。

Details of some_func would determine whether you use some_func(param1[:,None]) or some_func(param1)[:,None].

这篇关于在Python中建立三维阵列，以取代优化循环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！