Multiplying two 2D numpy arrays to a 3D array

Question

I've got two 2D numpy arrays called A and B, where A is M x N and B is M x n. My problem is that I wish to multiply each element of each row of B with corresponding row of A and create a 3D matrix C which is of size M x n x N, without using for-loops.

As an example, if A is:

A = np.array([[1, 2, 3],
              [4, 5, 6]])

and B is

B = np.array([[1, 2],
              [3, 4]])

Then the resulting multiplication C = A x B would look something like

C = [
     [[1, 2],
      [12, 16]],
     [[2, 4],
      [15, 20]],
     [[3, 6],
      [18, 24]]
     ]

Is it clear what I'm trying to achieve, and is it possible doing without any for-loops? Best, tingis

score 6 · Accepted Answer · edited Aug 18 '23 at 19:31

6

You can use the NumPy Einstein summation function, einsum():

C=np.einsum('ij,ik->jik',A,B)

edited Aug 18 '23 at 19:31

swimfar2

103
8

answered Aug 20 '15 at 11:55

yevgeniy

888
7
14

1

Thanks for the quick answer! This funcition helps a lot, will go with your solution! – Tingiskhan Aug 20 '15 at 12:01
`einsum` can do wondrous things, i think it is also faster than other numpy vector product methods. Also, by changing the index order after ->, you can do a reshape without an extra call to the actual `reshape`. – yevgeniy Aug 20 '15 at 12:09
2

@yevgeniy about being faster, [It depends on the size of the array](http://stackoverflow.com/a/30645430/832621). Usually it is faster for smaller arrays, while for larger arrays I measured speeds comparable to the method using `newaxis` – Saullo G. P. Castro Aug 20 '15 at 12:12
2

@SaulloCastro. Thanks for running an actual test. I was also following [this](http://stackoverflow.com/questions/32083997/huge-speed-difference-in-numpy-between-similar-code), where `einsum` really showed great performance, so it seems at array size >10000 is where one should consider alternative solutions. – yevgeniy Aug 20 '15 at 12:26

score 4 · Answer 2 · answered Aug 20 '15 at 11:54

4

It is possible by creating a new axis in each array and transposing the modified A:

A[np.newaxis,...].T * B[np.newaxis,...]

giving:

array([[[ 1,  2],
        [12, 16]],

       [[ 2,  4],
        [15, 20]],

       [[ 3,  6],
        [18, 24]]])

answered Aug 20 '15 at 11:54

Saullo G. P. Castro

56,802
26
179
234

3

Wow, thanks for the quick answer! Going with yev's answer though, as he needs the points ;D – Tingiskhan Aug 20 '15 at 12:00

Multiplying two 2D numpy arrays to a 3D array

2 Answers2