Index with Categorical dtype

[mrocklin]

Currently categorical data is converted to full form when made an index.

In [1]: import pandas as pd

In [2]: df = pd.DataFrame({'a': pd.Categorical(['a', 'b', 'a'])}, 
...                       index=pd.Categorical(['a', 'b', 'a'])) 
In [3]: df.dtypes 
Out[3]: 
a    category
dtype: object

In [4]: df.index.dtype 
Out[4]: dtype('O')

In [5]: df.index 
Out[5]: Index([u'a', u'b', u'a'], dtype='object')

Is there a way around this? I'm trying to keep my DataFrames free of object dtypes to reduce serialization costs. I would like to index off of a text column using categoricals.

This is with 0.15.2

[jreback]

requested in #7629 already

best way is simply to leave it as a coulumn and not set the index; it will then be serialized as a Categorical type