Get special group in pandas multiindex

问题

I have a DataFrame with MultiIndex like this:

In [5]: df
Out[5]:
                 a   b
lvl0 lvl1 lvl2
A0   B0   C0     0   1
          C1     2   3
          C2     4   5
          C3     6   7
     B1   C0     8   9
          C1    10  11
          C2    12  13
          C3    14  15
A1   B0   C0    16  47
          C1    18  49
          C2    20  41
          C3    22  43
     B1   C0    24  25
          C1    26  27
          C2    28  29
          C3    30  31
A2   B0   C0    32  33
          C1    34  35
          C2    36  37
          C3    38  39
     B1   C0    40  41
          C1    42  43
          C2    44  45
          C3    46  47

I want get the special lvl1 group in each lvl0 index. In this case, choice the group where column b has max value, result may like this:

                 a   b
lvl0 lvl1 lvl2
A0   B1   C0     8   9
          C1    10  11
          C2    12  13
          C3    14  15
A1   B0   C0    16  47
          C1    18  49
          C2    20  41
          C3    22  43
A2   B1   C0    40  41
          C1    42  43
          C2    44  45
          C3    46  47

Is there a indexing method like df[(('A0','B1'),('A1','B0'),('A2','B1')),:]? I have try my best, Thanks for any help.

回答1:

You can use:

df1 = df.reset_index(level=2, drop=True)
mask = df1.index.isin(df1.groupby(level=[0])['b'].idxmax())
df = df[mask]

print (df)
                 a   b
lvl0 lvl1 lvl2        
A0   B1   C0     8   9
          C1    10  11
          C2    12  13
          C3    14  15
A1   B0   C0    16  47
          C1    18  49
          C2    20  41
          C3    22  43
A2   B1   C0    40  41
          C1    42  43
          C2    44  45
          C3    46  47

Explanation:

First remove 3 level of MultiIndex by reset_index and groupby with idxmax for indices of max values in column b:

df1 = df.reset_index(level=2, drop=True)
idx = df1.groupby(level=[0])['b'].idxmax()
print (idx)
 lvl0
A0    (A0, B1)
A1    (A1, B0)
A2    (A2, B1)
Name: b, dtype: object

Then create boolean mask by compare by isin:

print (df1.index.isin(idx))
[False False False False  True  True  True  True  True  True  True  True
 False False False False False False False False  True  True  True  True]

and last filter by boolean indexing:

df = df[df1.index.isin(idx)]
print (df)
                 a   b
lvl0 lvl1 lvl2        
A0   B1   C0     8   9
          C1    10  11
          C2    12  13
          C3    14  15
A1   B0   C0    16  47
          C1    18  49
          C2    20  41
          C3    22  43
A2   B1   C0    40  41
          C1    42  43
          C2    44  45
          C3    46  47

来源：https://stackoverflow.com/questions/48090565/get-special-group-in-pandas-multiindex