Selecting sublevels of Multiindex columns in pandas

Question

I generate a multiindex dataframe like this example

import pandas as pd
import numpy as np

iterables = [ ['co1', 'co2', 'co3', 'co4'], ['age','weight'] ]
multi = pd.MultiIndex.from_product(iterables, names= ["Spread", "attribute"])

df = pd.DataFrame(np.random.rand(80).reshape(10,8),index = range(0,10), columns = multi)

The columns each have a sublevel attribute called 'weight'

I need to generate a list or (preferably) Series that contains, for a given row, all the 'weight' sub-columns in that row. In the example picture, I'd want a Series that gave me 0.02, 0.46, 0.33, 0.47.

Can anyone suggest a nice way to do this? The solutions I've thought of are all gross, and I suspect I have an incomplete understanding of the indexing capabilities of pandas.

Answer 1

IIUC then you can use loc and pass a tuple consisting of a slice and column label to access the col of interest at that level:

In [59]:
iterables = [ ['co1', 'co2', 'co3', 'co4'], ['age','weight'] ]
multi = pd.MultiIndex.from_product(iterables, names= ["Spread", "attribute"])
df = pd.DataFrame(np.random.rand(80).reshape(10,8),index = range(0,10), columns = multi)
df

Out[59]:
Spread          co1                 co2                 co3            \
attribute       age    weight       age    weight       age    weight   
0          0.600947  0.509537  0.605538  0.496002  0.215206  0.075079   
1          0.152956  0.922832  0.167788  0.024761  0.622378  0.983030   
2          0.712478  0.603798  0.407014  0.625474  0.445592  0.903240   
3          0.420569  0.576604  0.220097  0.401624  0.929464  0.512026   
4          0.273088  0.032303  0.607577  0.836231  0.751845  0.181522   
5          0.859699  0.274760  0.456812  0.666109  0.349961  0.237894   
6          0.632754  0.603252  0.157416  0.221576  0.068355  0.121864   
7          0.090595  0.035526  0.698262  0.525770  0.792618  0.220601   
8          0.670236  0.805195  0.310680  0.100464  0.875299  0.853238   
9          0.020501  0.405245  0.447614  0.999340  0.659616  0.709312   

Spread          co4            
attribute       age    weight  
0          0.297421  0.415730  
1          0.235259  0.156014  
2          0.365762  0.198299  
3          0.695431  0.478457  
4          0.331657  0.338436  
5          0.943810  0.097999  
6          0.638720  0.033747  
7          0.646969  0.475316  
8          0.623225  0.024976  
9          0.023494  0.959514  

In [61]:
df.loc[1,(slice(None),'weight')]

Out[61]:
Spread  attribute
co1     weight       0.922832
co2     weight       0.024761
co3     weight       0.983030
co4     weight       0.156014
Name: 1, dtype: float64

To explain the syntax :

df.loc[1,(slice(None),'weight')]

So the first param is just your index lave, the second param is a tuple consisting of a slice and a col label, the first member being slice(None) selects all cols 'col1' to 'col4' in effect, then the second param selects at the next level cols that match the label 'weight'