Numpy tensor: Tensordot over frontal slices of tensor

Question

I\'m trying to perform a matrix multiplication with frontal slices of a 3D tensor, shown below. If X.shape == (N, N), and Y.shape == (N, N, Y), the

Answer 1

The reduction is along axis=1 for X and axis=0 for tensor, thus np.tensordot based solution would be -

np.tensordot(X,tensor, axes=([1],[0]))

Explanation :

Let's take your iterative solution for explanation and in it the first iteration :

output[:, :, 0] = X.dot(tensor[:, :, 0])

In the dot product, the first input is X, whose shape is (N x N) and the second input is tensor[:, :, 0], which is the first slice along the last axis and its shape is (N x N). That dot product is causing reduction along the second axis of X, i.e. axis=1 and along the first axis, i.e. axis=0 of tensor[:, :, 0], which also happens to be the first axis of the entire array tensor. Now, this continues across all iterations. Therefore, even in the big picture, we need to do the same : Reduce/ Lose axis=1 in X and axis=0 in tensor, just like we did!

---

Integrating @hlin117's answer

np.tensordot(X,tensor, axes=([1],[0]))

Timing:

>>> N = 200
>>> tensor = np.random.rand(N, N, 30)
>>> X = np.random.rand(N, N)
>>> 
>>> %timeit np.tensordot(X, tensor, axes=([1], [0]))
100 loops, best of 3: 14.7 ms per loop
>>> %timeit np.tensordot(X, tensor, axes=1)
100 loops, best of 3: 15.2 ms per loop