str.get_dummies uses astype(str)

[hayd]

I wrote this so my fault. This probably means some stuff with break when passing stuff with unicode.

Will add example.

The reason is to include integers as strings before getting dummies, maybe should just drop that functionality?

[jreback]

maybe only operate on object dtypes? and if passed nothing to work on raise an error?

[jreback]

closing in favor of #6885 (good examples their)

[hayd]

IIRC this one was about str.get_dummies rather than pd.get_dummies. So slightly different but related/can use same test case.

[jreback]

oh...ok....suspect same issue though

[jreback]

I merged fix for #6885 this still open?

[hayd]

This is also a "decision" type issue, atm we apply .astype(str) to the Series before doing get dummies.

This means you can do stuff like:

In [11]: s = pd.Series(['a', 1, '1', np.nan])

In [12]: s.str.get_dummies()
Out[12]:
   1  a
0  0  1
1  1  0
2  1  0
3  0  0

[jreback]

ahh...makes sense, that is fine, does this blow up on unicode?

[hayd]

That's the claim, suspect on py3 only ?

[hayd]

So we could just kill this functionality, I had thought this was what all the string methods did... :s (it isn't)

[jreback]

well you can do:

try:
    data = astype('S')
except:
    data = astype('U')

(or just do the 'U' on py3)

[hayd]

I can't repo this on py3, I'm sure this came up for an actual bug.