creating Panel from dict of DataFrames loses dtypes

[CRP]

in the example below, "a" is a mixed type dataframe. Once included in the Panel, all columns become of type object:

In [249]: a=DataFrame(randn(30,3),columns=('a','b','c'))

In [250]: b=DataFrame(randn(30,3),columns=('a','b','c'))

In [251]: a['b']=a['b'].astype(object)

In [252]: a.dtypes
Out[252]:
a float64
b object
c float64

In [253]: b.dtypes
Out[253]:
a float64
b float64
c float64

In [254]: Panel({'1':a,'2':b})['1'].dtypes
Out[254]:
a object
b object
c object

In [255]: Panel({'1':a,'2':b})['2'].dtypes
Out[255]:
a float64
b float64
c float64

[wesm]

Panel can only be heterogeneous in the items. I just added an orient option to Panel so you can do:

In [4]: panel = Panel.from_dict({'1':a,'2':b}, orient='minor')

In [5]: panel
Out[5]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 3 (items) x 30 (major) x 2 (minor)
Items: a to c
Major axis: 0 to 29
Minor axis: 1 to 2

In [6]: panel['a'].values.dtype
Out[6]: dtype('float64')

In [7]: panel['b'].values.dtype
Out[7]: dtype('object')

In [8]: panel['c'].values.dtype
Out[8]: dtype('float64')