Groupby max erroneously returns NaN

[cancan101]

Using:

df =pd.read_csv(StringIO(""",Date,app,File
2013-04-23,2013-04-23 00:00:00,,log080001.log
2013-05-06,2013-05-06 00:00:00,,log.log
2013-05-07,2013-05-07 00:00:00,OE,xlsx"""), parse_dates=[0])

This does not work:

In [8]: df.groupby("Date")[["File"]].max()
Out[8]:
                     File
Date
2013-04-23 00:00:00   NaN
2013-05-06 00:00:00   NaN
2013-05-07 00:00:00  xlsx

[3 rows x 1 columns]

but this does:

In [9]: df.groupby("Date")["File"].max()
Out[9]:
Date
2013-04-23 00:00:00    log080001.log
2013-05-06 00:00:00          log.log
2013-05-07 00:00:00             xlsx
Name: File, dtype: object

[hayd]

Good news, the subselection bit is a red herring:

In [65]: df.groupby("Date").max()
Out[65]:
                              Unnamed: 0  app  File
Date
2013-04-23 00:00:00                  NaN  NaN   NaN
2013-05-06 00:00:00                  NaN  NaN   NaN
2013-05-07 00:00:00  2013-05-07 00:00:00   OE  xlsx

it's the app columns which is screwing stuff up

In [81]: df[['Date', 'File', 'app']].groupby("Date").max()
Out[81]:
                     File  app
Date
2013-04-23 00:00:00   NaN  NaN
2013-05-06 00:00:00   NaN  NaN
2013-05-07 00:00:00  xlsx   OE

In [83]: df[['Date', 'File', 'Unnamed: 0']].groupby("Date").max()
Out[83]:
                              File Unnamed: 0
Date
2013-04-23 00:00:00  log080001.log 2013-04-23
2013-05-06 00:00:00        log.log 2013-05-06
2013-05-07 00:00:00           xlsx 2013-05-07

[dsm054]

This seems to be fixed now as a result of the groupby mods; close?

[hayd]

Yes this is fixed in 0.14! Let's add a regression test and then close.

[cpcloud]

i'll add the test

[hayd]

Thanks!