BUG: string methods on empty series (GH7241)

[wabu]

closes #7241

[jreback]

this will need a release note (though 0.14.1 section is not there yet)

[jreback]

can u squash to a single commit

[wabu]

ok

[hayd]

This is good.

I wonder if for consistency perhaps these should return empty Series/DataFRames with object dtype?

In [8]: s.str.findall('a')
Out[8]: Series([], dtype: object)

[hayd]

FYI this is equivalent to Series(dtype=object).

[jreback]

agreed @wabu can u put in tests for all string methods in empty series (their maybe some already existing) and see what breaks if u enforce object dtype

[hayd]

It's not necessarily object for all string methods, I guess it should match the dtype of a non-empty example... for example str.count and str.len returns int64 (actually these could be the only examples).

[jreback]

right those are counting methods so they should be int64 in empty

[wabu]

sounds like the right thing to do, I'll have a look.
I realized some time ago that other methods behave strange with empty series/dfs, but did not have time to look at this (perhaps it's already fixed).

[wabu]

I changed _na_map. as noted in the release notes, now str_extract returns series with only nan values as dtype=object instead of float. I think it's strange if you have object for empty, but float for not-matching strings.

old:

>>> pd.Series(['foo']).str.extract('(bar)')
0 NaN
dtype: float

new:

>>> pd.Series(['foo']).str.extract('(bar)')
0 NaN
dtype: object

[jreback]

maybe say something about non-matching here (also do need to mention this is str.extract doc section (basics.rst)? and/or doc-string?

[hayd]

thanks for putting the empty tests together!

You have a merge conflict (probably in release notes) so please rebase off master.

[hayd]

closed via f24f2e8

(there was another merge conflict on the release notes!)

[hayd]

Thanks!