So the inputs x y and z must all be 3d right?

>>> x = np.arange(3)
>>> y = np.arange(4)
>>> z = np.arange(5)

>>> np.broadcast_arrays(x, y, z)
ValueError: shape mismatch: two or more arrays have incompatible dimensions on axis 0.

>>> x = np.arange(3)
>>> y = np.arange(4)
>>> z = np.arange(5)
>>> x.shape = (-1, 1, 1)
>>> y.shape = (1, -1, 1)
>>> z.shape = (1, 1, -1)

>>> X, Y, Z = np.broadcast_arrays(x, y, z)
>>> print Y.shape
(3, 4, 5)
>>> print y.shape
(1, 4, 1)

On Sunday, August 12, 2012, Phil Elson wrote:

So the inputs x y and z must all be 3d right?

x = np.arange(3)
y = np.arange(4)
z = np.arange(5)

np.broadcast_arrays(x, y, z)
ValueError: shape mismatch: two or more arrays have incompatible dimensions on axis 0.

x = np.arange(3)
y = np.arange(4)
z = np.arange(5)
x.shape = (-1, 1, 1)
y.shape = (1, -1, 1)
z.shape = (1, 1, -1)

X, Y, Z = np.broadcast_arrays(x, y, z)
print Y.shape
(3, 4, 5)
print y.shape
(1, 4, 1)

First, the broadcasted results have to be 2d, not 3d. Second, your first
example fails because that isn't how broadcasting works. You might want to
read up on the concept. Third, the example you provided makes no sense
because how should the z portion be interpreted?