如何使用列表（例如数据类）将 MultiIndex DataFrame 转换为对象？

Question

First, suppose you have the following DataFrame.

import pandas as ps

df = ps.DataFrame([
    [0, 'test0', 0, 'sub0', 'one'],
    [0, 'test0', 1, 'sub1', 'two'],
    [1, 'test1', 0, 'sub0', 'one'],
    [1, 'test1', 1, 'sub1', 'two'],
], columns=['id', 'name', 'sub_id', 'sub_name', 'value'])

df = df.set_index(['id', 'sub_id'])

            name sub_name value
id sub_id                      
0  0       test0     sub0   one
   1       test0     sub1   two
1  0       test1     sub0   one
   1       test1     sub1   two

I want to convert this to a list object like the one below (Here we use dataclass).

from typing import List
from dataclasses import dataclass

@dataclass
class SubObj:
    id: int
    name: str
    value: str

@dataclass
class MainObj:
    id: int
    name: str
    sub_obj: List[SubObj]

The output should look like this:

result = [
    MainObj(
        id=0,
        name='test0',
        sub_obj=[
            SubObj(
                id=0,
                name='sub0',
                value='one'
            ),
            SubObj(
                id=1,
                name='sub1',
                value='two'
            )
        ]
    ),
    MainObj(
        id=1,
        name='test1',
        sub_obj=[
            SubObj(
                id=0,
                name='sub0',
                value='one'
            ),
            SubObj(
                id=1,
                name='sub1',
                value='two'
            )
        ]
    ),
]

print(result)

[MainObj(id=0, name='test0', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')]), MainObj(id=1, name='test1', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')])]

I want to implement it so that it outputs a list of MainObj with as short and easy-to-understand code as possible.

Do you know how to do it?

Answer 1

How about a little list comprehension like this?

result = [MainObj(
    row[0][0], 
    row[1]['name'], 
    SubObj(
        row[0][1],
        row[1]['sub_name'],
        row[1]['value']
    )
) for row in df.iterrows()]

Returns

[MainObj(id=0, name='test0', sub_obj=SubObj(id=0, name='sub0', value='one')),
 MainObj(id=0, name='test0', sub_obj=SubObj(id=1, name='sub1', value='two')),
 MainObj(id=1, name='test1', sub_obj=SubObj(id=0, name='sub0', value='one')),
 MainObj(id=1, name='test1', sub_obj=SubObj(id=1, name='sub1', value='two'))]

Update

Just realized you want sub_obj's as lists. I think this would be a better way:

results = list()
for _, g in df.groupby(level=0):  # Groupby on first index
    results.append(
        MainObj(
            g.index[0][0],  # Get the first index value
            g['name'].iloc[0],
            [SubObj(row[0][1], row[1]['sub_name'], row[1]['value']) for row in g.iterrows()]))  # List comp iterrating over group rows

[MainObj(id=0, name='test0', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')]),
 MainObj(id=1, name='test1', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')])]

Answer 2

Here's a way to do it with pandas constructs

1. aggregate rows to SubObj
2. drop sub_id to create a dataframe that contains only the MainObj level info
3. aggregate rows to MainObj

>>> sub = df.reset_index('sub_id')[['sub_id', 'sub_name', 'value']].agg(lambda row: SubObj(*row), axis='columns')
>>> sub
id
0    SubObj(id=0, name='sub0', value='one')
0    SubObj(id=1, name='sub1', value='two')
1    SubObj(id=0, name='sub0', value='one')
1    SubObj(id=1, name='sub1', value='two')
>>> sub.groupby('id').agg(list)
id
0    [SubObj(id=0, name='sub0', value='one'), SubOb...
1    [SubObj(id=0, name='sub0', value='one'), SubOb...
Name: obj, dtype: object
>>> maindf = df[['name']].droplevel('sub_id').drop_duplicates().join(sub.groupby('id').agg(list))
>>> maindf
     name                                                obj
id                                                          
0   test0  [SubObj(id=0, name='sub0', value='one'), SubOb...
1   test1  [SubObj(id=0, name='sub0', value='one'), SubOb...
>>> maindf.reset_index().agg(lambda row: MainObj(*row), axis='columns').to_list()
[MainObj(id=0, name='test0', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')]), MainObj(id=1, name='test1', sub_obj=[SubObj(id=0, name='sub0', value='one'), SubObj(id=1, name='sub1', value='two')])]