Series.str.cat() with NaN in series returns NaN, rather than ignoring NaN

[hack-c]

I find this surprising as the rest of the pandas Series.str.* API ignores NaN values.

In [1]: import pandas as pd

In [2]: import numpy as np

In [3]: pd.__version__
Out[3]: u'0.17.0'

In [4]: s = pd.Series(['asdf','sdfg',np.nan,'qwer','wert'])

In [5]: s.str.cat(sep=' ')
Out[5]: nan

I think this should return

In [5]: s.str.cat(sep=' ')
Out[5]:'asdf sdfg qwer wert'

[hack-c]

Didn't read the docs closely enough-- the na_rep keyword more or less takes care of this. It's still a little strange that even if you set

In [6]: s.str.cat(sep=' ', na_rep='')
Out[6]: 'asdf sdfg  qwer wetr'

You get a double sep around the NaN. Easy to fix with a regex, but shouldn't NaN just be ignored?

[hack-c]

I actually think Series.str.cat() should ignore NaN by default, to comply with expected behavior for the rest of the API. But I'm curious if there's a reason for why this is a bad idea.

[Winterflower]

I think this would benefit from an additional doc example involving na_rep. I just read the API doc and it's not very clear from the explanation that doing Series.str.cat() without specifying anything for na_rep.

[hack-c]

That sounds like a good approach. I can do a pull request for the docs. I opened another issue (#11334) for str_cat as well that I haven't gotten around to yet that might also be best solved with a doc example as opposed to changing internals.

[jorisvandenbossche]

I agree it seems more logical to ignore the NaN value by default.