Is there a way to partition and reshape a 2D array to a 3D one. Like the following example:
Basically, I have a 4x4 matrix on the left and I want a 2x2x4 matrix in the way shown, so that I can apply numpy.mean on the 3rd axis. In reality the matrix I have is really huge, so that's why looping over the blocks is not an option.
Any help is greatly appreciated.
For your example, you can use numpy.lib.stride_tricks.as_strided.
In [1]: A = np.arange(16).reshape(4, 4)
In [2]: A
Out[2]:
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
In [3]: strides = A.itemsize * np.array([8, 2, 4, 1])
In [4]: x = p.lib.stride_tricks.as_strided(A, shape = (2, 2, 2, 2), strides = strides)
In [4]: x
Out[4]:
array([[[[ 0, 1],
[ 4, 5]],
[[ 2, 3],
[ 6, 7]]],
[[[ 4, 5],
[ 8, 9]],
[[ 6, 7],
[10, 11]]]])
In [5]: x.reshape(4, 2, 2)
Out[5]:
array([[[ 0, 1],
[ 4, 5]],
[[ 2, 3],
[ 6, 7]],
[[ 8, 9],
[12, 13]],
[[10, 11],
[14, 15]]])
The strides define offsets in bytes to use when traversing the array, and the as_strided function enables you to build a new array with user-define strides.
However, I do not know how efficient it is and how well it will scale for your use.
