This seems like it should be a simple operation, but for the life of me I can't figure it out. I have two arrays of incompatible shape that can't be broadcast together.

A1.shape == (2, 10, 10)
A2.shape == (2, 300)

I would like to add the two arrays along the first dimension, so that the result is an array with shape:

Result.shape == (2, 10, 10, 300)

In other words:

Result[0, 2, 3, 122] == A1[0, 2, 3]   A2[0, 122]
Result[1, 2, 3, 122] == A1[1, 2, 3]   A2[1, 122]

Can I do this vectorised, without resorting to looping?

CodePudding user response：

To make numpy do the broadcasting, you should insert new axes to broadcast over. (this is pointed out by Heisenbugs in the comments)

Result = A1[:,:,:,np.newaxis]  A2[:,np.newaxis,np.newaxis,:]

Do note that np.newaxis is None, so you can write None if you like. But I think np.newaxis is more readable.