Every NumPy array has a natural 1D order to its items. This is the order that
you see when you
ravel
the array. Reshaping (with the default order='C') does not change the order of
the items in the array. Therefore, x.reshape(shape) is the same as
x.ravel().reshape(shape). The key take-away message here is that an array
y is "reachable" via reshaping x if and only if y.ravel() equals x.ravel().
So consider the raveled (1-dimensional) order of the items in the given array and the desired array:
In [21]: x = np.array([[[-0.00047776], [-0.00065181]], [[-0.00065181], [ 0.00130446]], [[ 0.00130446], [ 0.00151989]], [[ 0.00151989], [ 0.00121407]], [[ 0.00121407], [-0.00121259]]], dtype=np.float32); x.ravel()
Out[21]:
array([-0.00047776, -0.00065181, -0.00065181, 0.00130446, 0.00130446,
0.00151989, 0.00151989, 0.00121407, 0.00121407, -0.00121259], dtype=float32)
versus
In [22]: y = np.array([ [[-0.00047776], [-0.00065181], [ 0.00130446], [ 0.00151989], [ 0.00121407]], [[-0.00065181], [ 0.00130446], [ 0.00151989], [ 0.00121407], [-0.00121259]] ]); y.ravel()
Out[22]:
array([-0.00047776, -0.00065181, 0.00130446, 0.00151989, 0.00121407,
-0.00065181, 0.00130446, 0.00151989, 0.00121407, -0.00121259])
Notice that the item order is different. Thus, to achieve the desired array, you must first (somehow) reorder the items in x. In this case using swapaxes to swap the first and second axes does the trick:
In [23]: x.swapaxes(0,1)
Out[25]:
array([[[-0.00047776],
[-0.00065181],
[ 0.00130446],
[ 0.00151989],
[ 0.00121407]],
[[-0.00065181],
[ 0.00130446],
[ 0.00151989],
[ 0.00121407],
[-0.00121259]]], dtype=float32)
In [26]: np.allclose(x.swapaxes(0,1), y)
Out[26]: True
No reshape is necessary since x.swapaxes(0,1) already has shape (2,5,1).
