Pandas: calculate by multi-index

Question

I have a dataframe like:

I would like to subtract values ​​like:

minus

What I have tried so far (dataframe: http://pastebin.com/PydRHxcz ):

index = pd.MultiIndex.from_tuples([key for key in dfdict], names = ['a','b','c','d'])
dfl = pd.DataFrame([dfdict[key] for key in dfdict],index=index)
dfl.columns = ['data']
dfl.sort(inplace=True)
d = dfl.unstack(['a','b'])

I can do:

d[0:5] - d[0:5]

And I get zeros for all values.

But if I do:

d[0:5] - d[5:]

I get Nans for all values. Any ideas how I can perform such an operation?

EDIT:

What works

dfl.unstack(['a','b'])['data'][5:] - dfl.unstack(['a','b'])['data'][0:5].values

But he feels a little awkward

Answer 1

You can use locto select all the lines corresponding to one label at the first level, for example:

In [8]: d.loc[0]
Out[8]:
          data                                                        ...
a         0.17                                       1.00
b          0          5         10         500        0          5
d
0.0  11.098909   9.223784  8.003650  10.014445  13.231898  10.372040
0.3  14.349606  11.420565  9.053073  10.252542  26.342501  25.219403
0.5   1.336937   2.522929  3.875139  11.161803   3.168935   6.287555
0.7   0.379158   1.061104  2.053024  12.358577   0.678352   2.133887
1.0   0.210244   0.631631  1.457333  15.117805   0.292904   1.053916

So, the subtraction is as follows:

In [11]: d.loc[0] - d.loc[1000]
Out[11]:
           data                                                         ...
a          0.17                                        1.00
b           0          5          10        500         0          5
d
0.0   -3.870946  -3.239915  -3.504068 -0.722377   -2.335147  -2.460035
0.3  -65.611418 -42.225811 -25.712668 -1.028758  -65.106473 -44.067692
0.5  -84.494748 -55.186368 -34.184425 -1.619957  -89.356417 -69.008567
0.7  -92.681688 -61.636548 -37.386604 -4.227343 -110.501219 -78.925078
1.0 -101.071683 -61.758741 -37.080222 -3.081782 -103.779698 -80.337487