Let's say I have a 3D numpy matrix M with shape (c, b, a). I want to shuffle a specific cell (x, y) in M along the z-axis --- that is, I want to shuffle the array [M[z][y][x] for z in range(c)] and assign it back (somehow, I have no idea how that would work --- maybe with M[:, y, x]?)
I can think of one way to do this: Flatten each 2D matrix, so M1 is a 2D matrix; transpose M1 to M2, shuffle M2[x+y*a], transpose M2 back to M1, then reform the 2D matrices.
However, this is clearly clunky. Is there a cleaner way to do this?
I found a solution: np.random.shuffle(M[:, y, x]). Note that this modifies M (you can make a copy by first executing M1=copy.deecopy(M) before shuffling).