Groupby transform idxmax return floats

[jesrael]

Transform with idxmax return wrong output - floats instead datetimes

Sample:

rng = pd.date_range('1/1/2011', periods=15, freq='D') 
np.random.seed(4)

stocks = pd.DataFrame({ 
    'price':(np.random.randn(15).cumsum() + 10) },index = rng)

stocks['week_id'] = pd.to_datetime(stocks.index).week #used for the groupby
print (stocks)
                price  week_id
2011-01-01  10.050562       52
2011-01-02  10.550513       52
2011-01-03   9.554604        1
2011-01-04  10.248203        1
2011-01-05   9.829901        1
2011-01-06   8.245324        1
2011-01-07   7.597617        1
2011-01-08   8.196192        1
2011-01-09   8.528442        1
2011-01-10   7.380966        2
2011-01-11   7.999635        2
2011-01-12   7.911648        2
2011-01-13   8.336721        2
2011-01-14   8.668974        2
2011-01-15   7.512158        2

print (stocks.groupby(stocks['week_id'])['price'].transform('idxmax'))
2011-01-01    1.293926e+18
2011-01-02    1.293926e+18
2011-01-03    1.294099e+18
2011-01-04    1.294099e+18
2011-01-05    1.294099e+18
2011-01-06    1.294099e+18
2011-01-07    1.294099e+18
2011-01-08    1.294099e+18
2011-01-09    1.294099e+18
2011-01-10    1.294963e+18
2011-01-11    1.294963e+18
2011-01-12    1.294963e+18
2011-01-13    1.294963e+18
2011-01-14    1.294963e+18
2011-01-15    1.294963e+18
Freq: D, Name: price, dtype: float64

print (stocks.groupby(stocks['week_id'])['price'].idxmax())
week_id
1    2011-01-04
2    2011-01-14
52   2011-01-02
Name: price, dtype: datetime64[ns]

SO question.

print (pd.show_versions())

INSTALLED VERSIONS
------------------
commit: None
python: 3.5.1.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 42 Stepping 7, GenuineIntel
byteorder: little
LC_ALL: None
LANG: sk_SK
LOCALE: None.None

pandas: 0.19.2+0.g825876c.dirty
nose: 1.3.7
pip: 8.1.1
setuptools: 20.3
Cython: 0.23.4
numpy: 1.11.0
scipy: 0.17.0
statsmodels: 0.6.1
xarray: None
IPython: 4.1.2
sphinx: 1.3.1
patsy: 0.4.0
dateutil: 2.5.1
pytz: 2016.2
blosc: None
bottleneck: 1.0.0
tables: 3.2.2
numexpr: 2.5.1
matplotlib: 1.5.1
openpyxl: 2.3.2
xlrd: 0.9.4
xlwt: 1.0.0
xlsxwriter: 0.8.4
lxml: 3.6.0
bs4: 4.4.1
html5lib: 0.999
httplib2: None
apiclient: None
sqlalchemy: 1.0.12
pymysql: None
psycopg2: None
jinja2: 2.8
boto: 2.39.0
pandas_datareader: 0.2.1
None

[jreback]

yeah .idxmax() is just defined as a passthru back to Series, then it is coerce. so need to be directly defined on GroupBy, xref to #15260

you can achieve this like this currently

In [16]: stocks.index[stocks.reset_index().groupby(stocks.index.week).price.transform('idxmax').astype(int)]
Out[16]: 
DatetimeIndex(['2011-01-02', '2011-01-02', '2011-01-04', '2011-01-04', '2011-01-04', '2011-01-04', '2011-01-04', '2011-01-04', '2011-01-04', '2011-01-14', '2011-01-14', '2011-01-14', '2011-01-14',
               '2011-01-14', '2011-01-14'],
              dtype='datetime64[ns]', freq=None)

[jreback]

if you'd like to put up a PR would be appreciated.

[dsm054]

@jreback: would it suffice just to add idxmax/min to cython_cast_blacklist?