Converting numpy array to structured array

Question

Let's say I have the following array:

arr = np.array([[1,2], [3,4]], dtype='u1')

and I want to convert it into a structured array like this one:

strarr = np.array([(1,2), (3,4)], dtype=[('a', 'u1'), ('b', 'u1')])

If I just try

arr.astype([('a', 'u1'), ('b', 'u1')])

it returns

>>> array([[(1, 1), (2, 2)],
       [(3, 3), (4, 4)]], dtype=[('a', 'u1'), ('b', 'u1')])

How can I convert the array so that it uses all elements of a row to fill the fields (provided that the numbers match) instead of duplicating each element?

Answer 1

There are special helper functions for this:

>>> from numpy.lib.recfunctions import unstructured_to_structured

So,

>>> import numpy as np
>>> arr = np.array([[1,2], [3,4]], dtype='u1')
>>> unstructured_to_structured(arr, dtype=np.dtype([('a', 'u1'), ('b', 'u1')]))
array([(1, 2), (3, 4)], dtype=[('a', 'u1'), ('b', 'u1')])

You can also create a view:

>>> arr.ravel().view(dtype=np.dtype([('a', 'u1'), ('b', 'u1')]))
array([(1, 2), (3, 4)], dtype=[('a', 'u1'), ('b', 'u1')])

And in this simple case, that is fine, but if you choose to use a view you sometimes have to worry about how the array is packed. Note, a view doesn't copy the underlying buffer! Which can make it much more efficient if you are working with large arrays.

Answer 2

The data for a structured array is supposed to be a list of tuples:

In [5]: arr = np.array([[1,2], [3,4]], dtype='u1')
In [6]: alist = [tuple(i) for i in arr]
In [7]: alist
Out[7]: [(1, 2), (3, 4)]
In [9]: np.array(alist, dtype=[('a', 'u1'), ('b', 'u1')])
Out[9]: array([(1, 2), (3, 4)], dtype=[('a', 'u1'), ('b', 'u1')])

Note that the input mirrors the output display.

In this case, a view is also possible (there's no change in underlying data):

In [10]: arr.view(dtype=[('a', 'u1'), ('b', 'u1')])
Out[10]: 
array([[(1, 2)],
       [(3, 4)]], dtype=[('a', 'u1'), ('b', 'u1')])
In [11]: _.ravel()      # but needs a shape correction
Out[11]: array([(1, 2), (3, 4)], dtype=[('a', 'u1'), ('b', 'u1')])