complex_frames.py

# Practically, it's useful if we have complex observations (rows) and variables (columns):

df = pandas.DataFrame(
   data=pandas.np.array(
      [[1,2,3,4,5],
       [6,7,8,9,10],
       [11,12,13,14,15]]).T,
   index=pandas.MultiIndex.from_arrays(
       [['x','x','x','y','z'],
        ['a','a','b','b','c'],
        [1,2,1,2,3]],
       names=['level1', 'level2', 'level3']),
   columns=pandas.MultiIndex.from_arrays(
      [['VAR', 'VAR', 'VAR'],
       ['VAR-A','VAR-A', 'VAR-B'],
       ['var-a','var-b', 'var-c']],
      names=['LEVEL1', 'LEVEL2', 'LEVEL3']
   )
)

LEVEL1                 VAR            
LEVEL2               VAR-A       VAR-B
LEVEL3               var-a var-b var-c
level1 level2 level3                  
x      a      1          1     6    11
              2          2     7    12
       b      1          3     8    13
y      b      2          4     9    14
z      c      3          5    10    15


# This allows us to filter by levels of indexing.

df[df.index.get_level_values('level2').isin(['a', 'c'])]

LEVEL1                 VAR            
LEVEL2               VAR-A       VAR-B
LEVEL3               var-a var-b var-c
level1 level2 level3                  
x      a      1          1     6    11
              2          2     7    12
z      c      3          5    10    15

# If we want to select an intersection from multiple levels, obviously, we can do:

ix1 = df.index.get_level_values('level2').isin(['a', 'c'])
ix2 = df.index.get_level_values('level3').isin([1, 3])
df[ix1 & ix2]

LEVEL1                 VAR            
LEVEL2               VAR-A       VAR-B
LEVEL3               var-a var-b var-c
level1 level2 level3                  
x      a      1          1     6    11
z      c      3          5    10    15

cx1 = df.columns.get_level_values('LEVEL2').isin(['VAR-A'])
cx2 = df.columns.get_level_values('LEVEL3').isin(['var-b'])
df.T[cx1 & cx2].T

LEVEL1                 VAR
LEVEL2               VAR-A
LEVEL3               var-b
level1 level2 level3      
x      a      1          6
              2          7
       b      1          8
y      b      2          9
z      c      3         10

# Selecting by both:
df[ix1 & ix2].T[cx1 & cx2].T

LEVEL1                 VAR
LEVEL2               VAR-A
LEVEL3               var-b
level1 level2 level3      
x      a      1          6
z      c      3         10


# Examples of slicing such frames:
https://pandas.pydata.org/pandas-docs/stable/advanced.html