Inconsistent integer column indexing

[alexfields]

When columns have an integer as their name, indexing them becomes inconsistent:

>>> import pandas as pd
>>> df = pd.DataFrame({1:[1,1],2:[2,2],'a':['a','a']})
>>> df
   1  2  a
0  1  2  a
1  1  2  a

[2 rows x 3 columns]
>>> df[1]
0    1
1    1
Name: 1, dtype: int64
>>> df[['a',1]]
   a  1
0  a  1
1  a  1

[2 rows x 2 columns]
>>> df[[1,2]]
   2  a
0  2  a
1  2  a

[2 rows x 2 columns]

Desired behavior would be:

>>> df[[1,2]]
   1  2
0  1  2
1  1  2

[2 rows x 2 columns]

This is in pandas 0.13.1

[jreback]

Actually it IS consistent with using ix. You are pretty much asking for troublie mixing
positions and labels, here are some solns. Read the docs: http://pandas.pydata.org/pandas-docs/stable/indexing.html#different-choices-for-indexing-loc-iloc-and-ix

Things are laid out very explicty, and you have all of the tools to do what you want. However, pandas cannot guess at what you are after; you need to be more explicit.

In [5]: df.ix[:,[1,2]]
Out[5]: 
   1  2
0  1  2
1  1  2

[2 rows x 2 columns]

In [6]: df.iloc[:,[1,2]]
Out[6]: 
   2  a
0  2  a
1  2  a

[2 rows x 2 columns]

In [7]: df.loc[:,[1,2]]
Out[7]: 
   1  2
0  1  2
1  1  2

[2 rows x 2 columns]

[alexfields]

Thanks for the response.

Perhaps you can comment on this behavior?

>>> df.loc[0,[1,2]]
1    1
2    2
Name: 0, dtype: object
>>> df.loc[0,[1,2]] = [3,4]
>>> df
   1  2  a
0  1  3  4
1  1  2  a

[2 rows x 3 columns]

Apparently loc selects a different part of the dataframe for getting than it does for setting?

[jreback]

this was a hidden bug; the index code has an amazing number of cases! fixed by #6548