1

This is my first Question here, let me know if I could've done anything better.

I'm trying to do a element-wise operation between two arrays, but the broadcasting wont work like I want it to.

I have an array of shape (N,4).

square_list = np.array([[1,2,255,255], [255,255,4,4], [255,255,8,8], [255,255,16,16], [255,255,8,4], [255,1,8,8], [1,255,8,8]], dtype='B')

I also have an array of shape (4,).

square = np.array([1, 8, 8, 1], dtype='B')

What I am able to do is compare my square against each element in the square_list and it is being broadcast into shape (N,4) as expected.

Now I want to compare my square in each possible rotation against the square_list. I've written a function which returns an array of shape (4,4), which contains each possible rotation.

square.rotations
    array([[1, 8, 8, 1],
           [1, 1, 8, 8],
           [8, 1, 1, 8],
           [8, 8, 1, 1]], dtype=uint8)

I know how to do this using a loop. I'd prefer however to use an element-wise operator that returns my desired shape.

What I get:

rotations & square_list
ValueError: operands could not be broadcast together with shapes (4,4) (6,4)

What I'd like to get:

rotations & square_list
array([[[1, 0, 8, 1],
        [1, 8, 0, 0],
        [1, 8, 8, 0],
        [1, 8, 0, 0],
        [1, 8, 8, 0],
        [1, 0, 8, 0]],
       [[1, 0, 8, 8],
        [1, 1, 0, 0],
        [1, 1, 8, 8],
        [1, 1, 0, 0],
        [1, 1, 8, 0],
        [1, 1, 8, 8]],
       [[0, 0, 1, 8],
        [8, 1, 0, 0],
        [8, 1, 0, 8],
        [8, 1, 0, 0],
        [8, 1, 0, 0],
        [8, 1, 0, 8],
        [0, 1, 0, 8]],
       [[0, 0, 1, 1],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 0, 0, 0],
        [0, 8, 0, 0]]], dtype=uint8)

This is just to visualize what I want, I don't particularly care about the order of the axis'. A shape of either (4, N, 4) or (N, 4, 4) would be great. I have the feeling that this can be achieved easily by just reshaping one of the input arrays but I couldn't figure it out.

Thanks in advance!

Improve this question

1 Answer 1

Reset to default
1

Add an extra dimension to rotations:

square_list & rotations[:,None]

output:

array([[[1, 0, 8, 1],
        [1, 8, 0, 0],
        [1, 8, 8, 0],
        [1, 8, 0, 0],
        [1, 8, 8, 0],
        [1, 0, 8, 0],
        [1, 8, 8, 0]],

       [[1, 0, 8, 8],
        [1, 1, 0, 0],
        [1, 1, 8, 8],
        [1, 1, 0, 0],
        [1, 1, 8, 0],
        [1, 1, 8, 8],
        [1, 1, 8, 8]],

       [[0, 0, 1, 8],
        [8, 1, 0, 0],
        [8, 1, 0, 8],
        [8, 1, 0, 0],
        [8, 1, 0, 0],
        [8, 1, 0, 8],
        [0, 1, 0, 8]],

       [[0, 0, 1, 1],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 8, 0, 0],
        [8, 0, 0, 0],
        [0, 8, 0, 0]]], dtype=uint8)
Improve this answer
Sign up to request clarification or add additional context in comments.

2 Comments

Thank you for your quick reply! Do you mind adding a short explanation or a link why this works so I can figure out similar problems by myself?
In your initial attempt, the dimensions were (7, 4) and (4, 4), so no possibility to align them on both. After reshaping "rotations", it becomes (4, 1, 4), so the 1 can broadcast to the 7. Is this clear?

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.