Given a 3d array and a 2d array,
a = np.arange(10*4*3).reshape((10,4,3))
b = np.arange(30).reshape((10,3))
How can I run elementwise-multiplication across the final axis of each, resulting in c where c has the shape .shape as a? I.e.
c[0] = a[0] * b[0]
c[1] = a[1] * b[1]
# ...
c[i] = a[i] * b[i]
Without any sum-reduction involved, a simple broadcasting would be really efficient after extending b to 3D with np.newaxis/None -
a*b[:,None,:] # or simply a*b[:,None]
Runtime test -
In [531]: a = np.arange(10*4*3).reshape((10,4,3))
...: b = np.arange(30).reshape((10,3))
...:
In [532]: %timeit np.einsum('ijk,ik->ijk', a, b) #@Brad Solomon's soln
...: %timeit a*b[:,None]
...:
100000 loops, best of 3: 1.79 µs per loop
1000000 loops, best of 3: 1.66 µs per loop
In [525]: a = np.random.rand(100,100,100)
In [526]: b = np.random.rand(100,100)
In [527]: %timeit np.einsum('ijk,ik->ijk', a, b)
...: %timeit a*b[:,None]
...:
1000 loops, best of 3: 1.53 ms per loop
1000 loops, best of 3: 1.08 ms per loop
In [528]: a = np.random.rand(400,400,400)
In [529]: b = np.random.rand(400,400)
In [530]: %timeit np.einsum('ijk,ik->ijk', a, b)
...: %timeit a*b[:,None]
...:
10 loops, best of 3: 128 ms per loop
10 loops, best of 3: 94.8 ms per loop