how to select a subset of columns of a row of a numpy record array?

Question

I understand the statements like x[['col1','col2']] can be used to select columns from a numpy record array. My question is how to perform the same operation on a single row of a record array. Here is what I would like to do:

Python 2.7.7 (default, Jun  2 2014, 01:41:14)
[...]
IPython 2.1.0 -- An enhanced Interactive Python.
[...]

In [1]: import numpy as np

In [2]: x = np.ones(3, dtype=[('a',float),('b',float),('c',int)])

In [3]: x[['b','c']][0]
Out[3]: (1.0, 1)

In [4]: row0 = x[0]

In [5]: row0['b']
Out[5]: 1.0

In [6]: row0['c']
Out[6]: 1

In [7]: row0[['b','c']]
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-7-287722bfeeaa> in <module>()
----> 1 row0[['b','c']]

IndexError: invalid index

I hoped that the last statement gives the result (1.0, 1), but instead I get the error shown above. How can I extract the specified columns from row0?

My real application has many more columns, and which columns I need depends on the row I'm in, so I cannot select the columns before the rows. Since the equivalent of row0[['b','c']] will be executed many times, I would like to avoid using a python loop here.

Part of my problem may be that I don't know the data type of row0:

In [8]: row0
Out[8]: (1.0, 1.0, 1)

In [9]: row0.dtype
Out[9]: dtype([('a', '<f8'), ('b', '<f8'), ('c', '<i8')])

row0 has a dtype, so it clearly is not a plain Python tuple. But it seems not to be a numpy array either:

In [10]: type(row0)
Out[10]: numpy.void

Answer 1

The problem is that when you do x[0] you get a np.void object with ndim=0. You can do x[0:1] instead in order to get a np.ndarray object with ndim=1, which will accept the column selection:

r0 = x[0:1]
r0[['b', 'c']]
#array([(1.0, 1)],
#      dtype=[('b', '<f8'), ('c', '<i4')])

You can also transform the np.void into a np.ndarray in this case doing:

r0 = x[0]
r0 = r0.reshape(-1)

which will accept the column selection.