[Python-ideas] PEP 335: Another use case

[Greg Ewing]

I've received the following comments concerning another
potential use case for PEP 335.

-------- Original Message --------
Subject: 	PEP 335 and NumPy NA
Date: 	Tue, 4 Oct 2011 15:42:50 -0700
From: 	Mark Wiebe <mwwiebe at gmail.com>
To: 	Gregory Ewing <greg at cosc.canterbury.ac.nz>

Hi Greg,

I took a glance at the python-ideas list, and saw that you're proposing
an update to PEP 335. I recently did a short internship at Enthought
where they asked me to look into the "missing data" problem for NumPy,
and I believe the work from that provides further motivation for the
PEP. Basically, the approach to allow good handling of missing data is
emulating the R programming language by introducing a new value called
NA, similar to the built-in None. For an NA with a boolean type, you get
a three-valued logic, as described
in http://en.wikipedia.org/wiki/Three-valued_logic#Kleene_logic. In the
NA I added to NumPy, this truth table cannot be satisfied because of the
issue PEP 335 addresses:

      >>> import numpy as np
      >>> if True or np.NA: print "yes"
     ...
     yes
      >>> if np.NA or True: print "yes"
     ...
     Traceback (most recent call last):
       File "<stdin>", line 1, in <module>
     ValueError: numpy.NA represents an unknown missing value, so its
     truth value cannot be determined


This can be worked around by using the bitwise operators, as mentioned
in the PEP for the array case:

      >>> if True | np.NA: print "yes"
     ...
     yes
      >>> if np.NA | True: print "yes"
     ...
     yes


Here are the documents with more information:

https://github.com/numpy/numpy/blob/master/doc/source/reference/arrays.maskna.rst
https://github.com/numpy/numpy/blob/master/doc/neps/missing-data.rst

Cheers,
Mark