Add hierarchy to numpy structured array

Question

I would like to take an existing array with several named fields, and create a new array (or change it in place) with one field with a hierarchical dtype equal to the original dtype. That is,

newarray = np.array(oldarray, dtype=[('old',oldarray.dtype)])

such that newarray['old'] is identical in shape and structure to oldarray

Here's an example:

In [1]: import numpy as np

In [2]: dt = np.dtype([('name',np.str_,2),('val',np.float_)])

In [3]: constants = np.array([('pi',3.14),('e',2.72)],dtype=dt)

In [4]: constants
Out[4]: 
array([('pi', 3.14), ('e', 2.72)], 
      dtype=[('name', '|S2'), ('val', '<f8')])

In [5]: numbers = constants.astype([('constants',dt)])

But this gives me all zeros:

In [6]: numbers
Out[6]: 
array([(('', 0.0),), (('', 0.0),)], 
      dtype=[('constants', [('name', '|S2'), ('val', '<f8')])])

I have the same problem if I try to make a copy:

In [7]: numbers = np.array(constants,dtype=[('constants',dt)])

In [8]: numbers
Out[8]: 
array([(('', 0.0),), (('', 0.0),)], 
      dtype=[('constants', [('name', '|S2'), ('val', '<f8')])])

Also: Does anybody know why this is happening?

Answer 1

You can take a view of the original array with the new dtype (http://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.view.html):

>>> import numpy as np
>>> dt = np.dtype([('name',np.str_,2),('val',np.float_)])
>>> constants = np.array([('pi',3.14),('e',2.72)],dtype=dt)
>>> 
>>> numbers = constants.view([('constants',dt)])
>>> 
>>> numbers['constants']
array([('pi', 3.14), ('e', 2.72)], 
      dtype=[('name', '|S2'), ('val', '<f8')])

Be aware that the resulting array numbers is a view of the origingal array, so changes in one of them will also affect the other.

Answer 2

I can solve the problem by making a list of the original array:

In [9]: numbers = np.array([constants],dtype=[('constants',dt)])

In [10]: numbers
Out[10]: 
array([[(('pi', 3.14),), (('e', 2.72),)]], 
      dtype=[('constants', [('name', '|S2'), ('val', '<f8')])])

But when I look at it, of course I have an extra nesting in the list:

In [11]: numbers['constants']
Out[11]: 
array([[('pi', 3.14), ('e', 2.72)]], 
      dtype=[('name', '|S2'), ('val', '<f8')])

In [12]: numbers['constants']['name']
Out[12]: 
array([['pi', 'e']], 
      dtype='|S2')

I really just want the first item in the list:

In [13]: numbers[0]
Out[13]: 
array([(('pi', 3.14),), (('e', 2.72),)], 
      dtype=[('constants', [('name', '|S2'), ('val', '<f8')])])

I can also achieve this by flattening the array afterward:

In [14]: numbers.flatten()
Out[14]: 
array([(('pi', 3.14),), (('e', 2.72),)], 
      dtype=[('constants', [('name', '|S2'), ('val', '<f8')])])

In [15]: numbers.flatten()['constants']
Out[15]: 
array([('pi', 3.14), ('e', 2.72)], 
      dtype=[('name', '|S2'), ('val', '<f8')])

In [16]: numbers.flatten()['constants']['name']
Out[16]: 
array(['pi', 'e'], 
      dtype='|S2')

But isn't that a hack? What I really want in the end is this array:

In [17]: numbers = np.array([(('pi', 3.14),), (('e', 2.72),)],dtype=[('constants',dt)])

In [18]: numbers['constants']
Out[18]: 
array([('pi', 3.14), ('e', 2.72)], 
      dtype=[('name', '|S2'), ('val', '<f8')])

In [19]: numbers['constants']['name']
Out[19]: 
array(['pi', 'e'], 
      dtype='|S2')

Without having to make a one-item list and then flatten it. Any better ideas?