BUG: consistency between logical ops & type casts

[behzadnouri]

Closes #6528.
Currently series logical operations are bound to return a series with dtype='bool'. That being a good decision or not, it is good to have some internal consistency:

>>> a = pd.Series([nan, nan, nan])
>>> b = pd.Series([False, True, nan])
>>> a | b == b | a  # not symmetric
0     True
1    False
2     True
dtype: bool
>>> a & b == a.astype(bool) & b.astype(bool)  # not consistent w/ type casts
0     True
1    False
2    False
dtype: bool

With this patch, all these tests pass.

[jreback]

this is a valid test
don't change it
you can add to tests but not change w/o a really good reason

[behzadnouri]

@jreback in python

>>> bool(np.nan)
True

in pandas:

>>> pd.Series([np.nan], dtype='bool')
0    True
dtype: bool

this test passes only if np.nan is casted to False.

same comment for the other test.

[jreback]

need a release note

[jreback]

@behzadnouri can you rebase this. I believe there was another pr with modified this code.

[behzadnouri]

this pr is based on the assumption that the element-wise logical operations are bound to return a series with dtype=bool. #9338 has changed this; so this way may not be the best way to go forward.

i feel like following python's mechanism of boolean operations is a better path.

[jreback]

can u provide an example of what you think is wrong / what needs changing?

[behzadnouri]

I mean element-wise and/or etc should actually do element-wise logical operations, so then a & b and a | b should be equivalent of

>>> Series(map(lambda x, y: x and y, a, b))  # element-wise and
>>> Series(map(lambda x, y: x or y, a, b))  # element-wise or

for example:

>>> a = pd.Series(['a', '', np.nan])
>>> b = pd.Series([0, 1, 2])
>>> pd.Series(map(lambda x, y: x and y, a, b))
0    0
1
2    2
dtype: object
>>> pd.Series(map(lambda x, y: x or y, a, b))
0      a
1      1
2    NaN
dtype: object

#9338 is sort of doing this for integers, but same can be done for all types as in python "abc" and 3.14 is a valid expression.

[jreback]

@behzadnouri we ought to reexamine this for 0.17.0. pls rebase when you have a chance.

[behzadnouri]

@jreback I close this one per my comment earlier that i no longer feel this is a good approach. I will do a new pr when i get a chance and if i could make it work.

[jreback]

great thanks!