Unstacked data in unexpected order

[sniemi]

After unstacking data that were grouped using groupby the data appears in an unexpected order. The groupby data are in the ascending order, but after unstacking this is no longer true.

The following code demonstrates:
import numpy as np
import pandas as pd

print pd.version

random data

data = pd.DataFrame()
data['a'] = np.random.random(1000)_10. - 2
data['b'] = np.random.random(1000)_10. - 2
data['c'] = np.random.random(1000)*10. - 2

bins in a and b

ba = np.linspace(-5., 5, 8)
bb = np.linspace(-5., 5, 9)

groupby bins

groups = data.groupby([pd.cut(data['a'], ba), pd.cut(data['b'], bb)])

mask, needs at least 10 to get robust std estimate

msk = groups.count()['c'] > 10

std(c) per bin, these are ordered as expected

stdc = groups['c'].agg([np.std, len])[msk]
print stdc

now unstack

result = stdc['std'].unstack()

and the results are not ordered as expected

print result

and the result is:
b (-1.25, 0] (-2.5, -1.25] (0, 1.25] (1.25, 2.5]
a
(-0.714, 0.714] 2.878018 NaN 3.019754 2.908735
(-2.143, -0.714] 2.490683 NaN 2.809243 2.754959
(0.714, 2.143] 2.656060 2.940749 2.421975 2.677205
(2.143, 3.571] 3.128538 2.879978 3.038488 3.191041
(3.571, 5] 2.534340 2.943076 3.215626 2.687091

b (2.5, 3.75] (3.75, 5]
a
(-0.714, 0.714] 3.328777 3.305938
(-2.143, -0.714] 2.716552 3.090601
(0.714, 2.143] 3.093307 1.669262
(2.143, 3.571] 2.885055 3.295693
(3.571, 5] 3.333355 2.972684

[5 rows x 6 columns]

This seems to happen if the bins cross zero.

[jreback]

this is a duplicate of #5314

ATM the moment its not easy to sort categoricals; these are being treated as 'strings' right now
because categorical support at the frame level is not fully their yet.

stay tuned and/or pls contribute to this effort!