NaN in columns produces TypeError

[alefnula]

>>> pd.DataFrame([[1,2,3]], columns=[1.1,2.2,np.nan])
Traceback (most recent call last):
...
TypeError: ("'NoneType' object is not iterable", u'occurred at index 2.2')

Is this a bug or a feature? I don't see anything about that in the v0.13.0.txt, and it worked in 0.12:

>>> pd.DataFrame([[1,2,3]], columns=[1.1,2.2,np.nan])
    1.1   2.2  NaN
0     1     2     3

[jreback]

This is a quite tricky and non-trivial bug. It didn't work before, just didn't error. Not really sure it should even be allowed. (but is supported in some places).

[jreback]

Its not very useful; you end up defeating name based indexing, eg. df[np.nan] doesn't even make sense

[alefnula]

@jreback I found this when I tried to reproduce the issue #4987 in 0.13.

>>> df = pd.read_html(
             'http://en.wikipedia.org/wiki/Vancouver',
             match='Municipality', header=0
         )[0]
>>> df
0         Country         Municipality   NaN
1         Ukraine               Odessa  1944
2           Japan             Yokohama  1965
3  United Kingdom  Edinburgh[198][199]  1978
4           China       Guangzhou[200]  1985
5   United States          Los Angeles  1986
6     South Korea                Seoul  2007

This is the result in 0.12, in 0.13 it just brakes... Maybe this should be a bug on the read_html function then?

[jreback]

@alefnula thanks for pointing this out!