Setting DataFrame column headers to a MultiIndex

Question

How do I convert an existing dataframe with single-level columns to have hierarchical index columns (MultiIndex)?

Example dataframe:

In [1]:
import pandas as pd
from pandas import Series, DataFrame

df = DataFrame(np.arange(6).reshape((2,3)),
               index=['A','B'],
               columns=['one','two','three'])
df
Out [1]:
   one  two  three
A    0    1      2
B    3    4      5

I'd have thought that reindex() would work, but I get NaN's:

In [2]:
df.reindex(columns=[['odd','even','odd'],df.columns])
Out [2]:
   odd  even    odd
   one   two  three
A  NaN   NaN    NaN
B  NaN   NaN    NaN

Same if I use DataFrame():

In [3]:
DataFrame(df,columns=[['odd','even','odd'],df.columns])
Out [3]:
   odd  even    odd
   one   two  three
A  NaN   NaN    NaN
B  NaN   NaN    NaN

This last approach actually does work if I specify df.values:

In [4]:
DataFrame(df.values,index=df.index,columns=[['odd','even','odd'],df.columns])
Out [4]:
   odd  even    odd
   one   two  three
A    0     1      2
B    3     4      5

What's the proper way to do this? Why does reindex() give NaN's?

Answer 1

You were close, just set the columns directly to a new (equal sized) index-like (which if its a list-of-list will convert to a multi-index)

In [8]: df
Out[8]: 
   one  two  three
A    0    1      2
B    3    4      5

In [10]: df.columns = [['odd','even','odd'],df.columns]

In [11]: df
Out[11]: 
   odd  even    odd
   one   two  three
A    0     1      2
B    3     4      5

Reindex will reorder / filter the existing index. The reason you get all nans is you are saying, hey find the existing columns that match this new index; none match, so that's what you get