suppose a 2D numpy array has to be created from two 1D numpy arrays x and y. The fortran pseudocode is as follows:

   do i = 1, n
   do j = 1, m
       A[i, j] = x[i]*y[j]
   enddo
   enddo

what is the most numpythonic way to create this 2D array this without using loops?

CodePudding user response：

The numpythonic way would be to use broadcasting.

You can do a broadcasting operation if the dimensions of your arrays are compatible based on two criteria. Starting from the rightmost dimensions:

1. they are equal

2. they are 1.

If you want there to be length-x number of rows, you could to give it an appropriate dimension, with 1, for it to be broadcast with the * operation:

>>> import numpy as np
>>> x = np.array([1,2,3,4])
>>> y = np.array([10, 11])
>>> print(x[:, None] ,  y, sep="\n"*2)
[[1]
 [2]
 [3]
 [4]]

[10 11]
>>> x[:, None] * y
array([[10, 11],
       [20, 22],
       [30, 33],
       [40, 44]])

Or using the other dimensions (corresponding to switching the orders of the loop):

>>> print(x , y[:, None], sep="\n"*2)
[1 2 3 4]

[[10]
 [11]]
>>> x * y[:, None]
array([[10, 20, 30, 40],
       [11, 22, 33, 44]])