二维 ndarray：逐行操作

Question

I have 2D numpy array, with example shape:

>>> a.shape
(48, 160)

and I want to do simple operation between elements or each row, like a[0] - a[1] but for every row against all other rows.

I know how to do it simply by using for loop and iterating rows, but I was wondering if there is some numpy slicing specific instruction, that can do this without using for loops

Answer 1

You can use broadcasting magic to do this.

import numpy as np
a = np.arange(12).reshape((4, 3))
b = np.arange(15).reshape((5, 3))
diff = a[np.newaxis, :, :] - b[:, np.newaxis, :]
diff.shape
# (5, 4, 3)

This is a good broadcasting tutorial. In this case I make a (1, 4, 3) and b (5, 1, 3) and I get a result that's (5, 4, 3), the difference of every row pair in a and b.