Python numpy array multiplication

Question

If I have to arrays X (X has n rows and k columns) and Y (Y has n rows and q columns) how do I multiply the two in the vector form, such that I obtain array Z with following characteristics:

Z[0]=X[:,0]*Y
Z[1]=X[:,1]*Y
Z[2]=X[:,2]*Y
...
Z[K-1]=X[:,k-1]*Y
Z[K]=X[:,k]*Y

for c in range(X.shape[1]):
    Z[c]=X[:,c].dot(Y)

What's the final shape of `Z`? (k,)? (k,q)? Are you summing over the `n` size? Details please. May be even a small 2x3x4 example so we are sure we understand you. — hpaulj, Nov 19 '16 at 23:48
final shape of Z is (k,q). I will try posting loop version shortly if that may help. — 1nsg, Nov 19 '16 at 23:53
See my answer for the kind of test case that we like to see. — hpaulj, Nov 20 '16 at 00:21

hpaulj · Answer 1 · 2016-11-20T00:21:38.140

From your description, and almost no thinking:

Z=np.einsum('nk,nq->kq',X,Y)

I could also write it with np.dot, with a transpose or two. np.dot does the matrix sum over the last dim of the 1st and 2nd to last of 2nd

Z = np.dot(X.T, Y)

=================

In [566]: n,k,q=2,3,4
In [567]: X=np.arange(n*k).reshape(n,k)
In [568]: Y=np.arange(n*q).reshape(n,q)
In [569]: Z=np.einsum('nk,nq->kq',X,Y)
In [570]: Z
Out[570]: 
array([[12, 15, 18, 21],
       [16, 21, 26, 31],
       [20, 27, 34, 41]])
In [571]: Z1=np.empty((k,q))
In [572]: Z1=np.array([X[:,c].dot(Y) for c in range(k)])
In [573]: Z1
Out[573]: 
array([[12, 15, 18, 21],
       [16, 21, 26, 31],
       [20, 27, 34, 41]])
In [574]: X.T.dot(Y)
Out[574]: 
array([[12, 15, 18, 21],
       [16, 21, 26, 31],
       [20, 27, 34, 41]])

Python numpy array multiplication

1 Answers1