BUG: Pandas treats NaN and None differently when multiplying booleans

[darindillon]

Using pandas 0.10.
Pandas throws error when multiplying a series with booleans and a series with None. However, it works fine when you reverse the order. Also, it works fine if you use NaN instead of None. But isn't pandas supposed to treat NaN and None interchangably?

import pandas
import numpy as np

nans = pandas.Series([np.NaN, np.NaN])
nones = pandas.Series([None, None])
bools= pandas.Series([True,False])

x = nans * bools #works fine
x = bools * nans #works fine
x = nones * bools #works fine
x = bools * nones #ERROR! THIS BLOWS UP

[jreback]

You have different dtypes for these series.

In [9]: nans
Out[9]: 
0   NaN
1   NaN
dtype: float64

In [10]: nones
Out[10]: 
0    None
1    None
dtype: object

In [11]: bools
Out[11]: 
0     True
1    False
dtype: bool

Nones are are not treated like np.nan, they are an object type (like a string).

They will be coered into np.nan in a constructor if there are other types.

n [12]: Series([None,1])
Out[12]: 
0   NaN
1     1
dtype: float64

however in your example, they are left as object because pandas doesn't know what to do with them (similary to how mixed types are handled)

The return type for a binary operation (e.g. * here), is whatever numpy says it is, which is the first type, maybe upcasted (e.g. an int * float == float type, unless the float is equiv to an int)

your last example blows up because the return type is bool, which you can't do any operations on except for comparison to other bool types

the bools * nans is a special case in numpy I think where you get NaN output, not sure why