What is the numpy operation to make a dot product over an axis-CodePudding

I've got an array (L) of shape (2,2) and an array (W) of shape (2, 5, 3) I'd like to know what is the operation of that does a dot product for each element in axis 2. the result should be of shape (2,5,3). I've tried:

np.malmul(L, W)
ValueError: matmul: Input operand 1 has a mismatch in its core dimension 0

and

np.tensordot(L, W) 
ValueError: shape-mismatch for sum

both return me an error. The slow non pythonic solution is:

W_corr = []
for i in range(W.shape[-1]):
     res_ = L.dot(W[:,:,i])
     W_corr.append(res_)
W_corr = np.moveaxis(np.array(W_corr), 0, -1)

But I'm sure there's a better way. Any idea?

CodePudding user response：

Use .swapaxes() for this:

L = np.random.rand(2, 2)
W = np.random.rand(2, 5, 3)

W_corr = np.dot(W.T, L.T).swapaxes(0, 2)

CodePudding user response：

Swap the first two axes of W, and then do the dot product:

>>> np.random.seed(0)
>>> L = np.random.rand(2, 2)
>>> W = np.random.rand(2, 5, 3)
>>> L.dot(W.transpose(1, 0, 2))   # or W.swapaxes(0, 1)
array([[[0.85473091, 1.05437284, 0.81170348],
        [0.8194622 , 1.0870973 , 0.2950265 ],
        [0.8921741 , 0.39278942, 0.98736769],
        [0.8812003 , 0.33554736, 0.2370251 ],
        [0.56481983, 0.78318688, 0.83360085]],

       [[0.72941859, 0.92255399, 0.69920961],
        [0.78898045, 1.00615784, 0.29557025],
        [0.82590506, 0.39690928, 0.85713066],
        [0.84226213, 0.26876025, 0.19667025],
        [0.43405383, 0.75042139, 0.77877449]]])