ValueError: operands could not be broadcast together with shapes

Question

I have a Numpy array with shape (6,2),

a = array([[1, 2, 3, 3, 2, 1],
           [4, 5, 6, 7, 8, 9]])

I want to raise each element of array to two different powers (-6, -8) using np.power(a, [-6,-8]) but this raises ValueError: operands could not be broadcast together with shapes. How can I do this? Thanks!

Praveen Praveen · Accepted Answer · 2014-07-08T16:10:56

Are you trying to get this:

>>> a = np.array([[1, 2, 3, 3, 2, 1], [4, 5, 6, 7, 8, 9]], dtype=float)
>>> np.array([np.power(a, -6), np.power(a, -8)])
array([[[  1.00000000e+00,   1.56250000e-02,   1.37174211e-03,
           1.37174211e-03,   1.56250000e-02,   1.00000000e+00],
        [  2.44140625e-04,   6.40000000e-05,   2.14334705e-05,
           8.49985975e-06,   3.81469727e-06,   1.88167642e-06]],
       [[  1.00000000e+00,   3.90625000e-03,   1.52415790e-04,
           1.52415790e-04,   3.90625000e-03,   1.00000000e+00],
        [  1.52587891e-05,   2.56000000e-06,   5.95374181e-07,
           1.73466526e-07,   5.96046448e-08,   2.32305731e-08]]])

Power doesn't work that way. According to the in-built doc, you can either

Raise the whole array to a single power, as in np.power(a, 3), or
Raise one array by different powers, as in np.power(a, b), where b is an array with the same shape as a, or

Raise one array to many powers, as in np.power(a, b), where b has "at least" the shape of a, with extra dimensions containing different powers (as I understand it, anyway). So, for your case, the following would also produce the same result:

>>> b = np.array([-6 * np.ones(a.shape), -8 * np.ones(a.shape)])
>>> b
array([[[-6., -6., -6., -6., -6., -6.],
        [-6., -6., -6., -6., -6., -6.]],    
       [[-8., -8., -8., -8., -8., -8.],
        [-8., -8., -8., -8., -8., -8.]]])
>>> np.power(a, b)

ValueError: operands could not be broadcast together with shapes

1 Answers