Numpy Broadcasting

问题:

What happens when i make this operation in Numpy?

a = np.ones([500,1])
b = np.ones([5000,])/2
c = a + b 

# a.shape (500,1)
# b.shape (5000, )
# c.shape (500, 5000)

I'm having a hard time to figure out what is actually happening in this broadcast.

回答1:

Numpy assumes for 1 dimensional arrays row vectors, so your summation is indeed between shapes (500, 1) and (1, 5000), which leads to matrix summation.

Since this is not very clear, you should extend your dimensions explicitly:

>>> np.arange(5)[:, None] + np.arange(8)[None, :]
array([[ 0,  1,  2,  3,  4,  5,  6,  7],
       [ 1,  2,  3,  4,  5,  6,  7,  8],
       [ 2,  3,  4,  5,  6,  7,  8,  9],
       [ 3,  4,  5,  6,  7,  8,  9, 10],
       [ 4,  5,  6,  7,  8,  9, 10, 11]])