inconsistent indexing behaviour with multi index

[behzadnouri]

i have a dataframe with multi-index as

>>> df
              val  diffs
tag day                 
A   26   0.208105    NaN
    37   2.709899    NaN
    57   0.933981    NaN
B   75   0.624029    NaN
C   0   -1.645489    NaN
    27  -1.749618    NaN
    27  -0.725497    NaN
    37  -0.381485    NaN
    67   0.110426    NaN
    82   1.253857    NaN

[10 rows x 2 columns]

however

>>> df.val[ 'A' ]
A   NaN
Name: val, dtype: float64
>>> df.val[ 'X' ]
X   NaN
Name: val, dtype: float64



>>> pd.__version__
'0.13.0rc1-92-gf6fd509'

[jreback]

believe it or not, this is ony under PY3, when you happend to use 'A' as a column label (because that is a valid Timestamp....wierd but true)

[jreback]

pls update to master and give a try..thanks for the report

[behzadnouri]

this works:

>>> pd.__version__
'0.13.0rc1-99-gbac7c6e'
>>> df.val[ 'A' ]
day
26     0.208105
37     2.709899
57     0.933981
Name: val, dtype: float64

but i do not understand below, and why it does not throw KeyError. there is no X in df:

>>> df.val[ 'X' ]
X   NaN
Name: val, dtype: float64

[jreback]

PY3 does such weird things with exceptions. I think they were trying to make it better, but just is more confusing. #5737 should fix this. This seems very simple, except that since pandas allows indexing with all kinds of things, several things needs to be tried in order to index properly, e.g. imagine that the first level of the index was not 'A','B','C' but a dates, then s['2011'] should work! (which is IMHO somewhat confusing, but a convenience)

[jreback]

ok...master is update....give it a go

[jreback]

You might find these other access methods useful

This is the recommended actually

In [7]: df.loc['D','val']
Out[7]: 
day
26    -0.269229
37    -0.330613
57    -0.654839
Name: val, dtype: float64

Specify a tuple for each element you want (could be a list of tuples too)

In [8]: df.loc[('D',57),'val']
Out[8]: 
tag  day
D    57    -0.654839
Name: val, dtype: float64

Cross sectional

In [9]: df.xs(57,level=1)
Out[9]: 
          val
tag          
D   -0.654839

[1 rows x 1 columns]

[behzadnouri]

confirmed it raises KeyError,

thanks for quick fix