pandas 0.20:如何对具有多级索引的列进行自定义排序?

pandas 0.20: How do I do a custom sort of the columns with multi-level indexes?

pandas 的新手。使用 pandas 0.20,因此没有 CategoricalDtype。我想在将几个 df 与 concat 合并后对列进行自定义排序。合并后的 df 将具有多级索引列。

使用分类,它不适用于自定义排序。

dfs=[table2, table3,table4]
L_list=['A', 'B', 'C']     
test=pd.Categorical(L_list, categories=['B', 'C','A'], ordered=True)

merged_df = pd.concat(

    dfs, axis=1,

    keys=pd.MultiIndex.from_arrays([

        test,       # Top Level Keys

        ['Cat1', 'Cat2', 'Cat1']  # Second Level Keys

    ], names=['Importance', 'Category'])

)

output=merged_df.sort_index(axis=1, level=[0])

当前状态

**Merged_df**

Importance| A         |  B   | C       |
Category | Cat1       | Cat2 | Cat1    |
         |Total Assets| AUMs | Revenue |
Firm 1   | 100        | 300  | 300     |
Firm 2   | 200        | 3400 | 200     |
Firm 3   | 300        | 800  | 400     |
Firm 4   | NaN        | 800  | 350     |


期望状态

**Merged_df**

Importance|  B   | C       | A            |
Category  | Cat2 | Cat1    | Cat1         |
         |AUMs | Revenue | Total Assets |
Firm 1   | 300  | 300     | 100          |  
Firm 2   | 3400 | 200     | 200          |  
Firm 3   |  800  | 400     | 300          |  
Firm 4   |  800  | 350     | NaN          |  

不确定 0.20 的所有可能性,但一个想法是将多索引列转换为框架,将每个级别更改为分类数据(就像您在问题中对测试所做的那样),然后 sort_values数据框,保持列的索引按照您想要重新排列 merged_df 列的顺序排序。看这个例子:

# simple example
dfs=[
    pd.DataFrame({'a':[0,1]}, index=[0,1]), 
    pd.DataFrame({'b':[0,1]}, index=[0,1]), 
    pd.DataFrame({'c':[0,1]}, index=[0,1]),
]

L_list=['A', 'B', 'B'] # changed C to have 2 B with 2 Cat  

merged_df = pd.concat(
    dfs, axis=1,
    keys=pd.MultiIndex.from_arrays([
        test,       # Top Level Keys
        ['Cat1', 'Cat1', 'Cat2']  # Second Level Keys
    ], names=['Importance', 'Category'])
)
print(merged_df)
#      A    B     
#   Cat1 Cat1 Cat2
#      a    b    c
# 0    0    0    0
# 1    1    1    1

所以你可以做到

# create dataframes of columns names
col_df = merged_df.columns.to_frame()

# change to catagorical each level you want
col_df[0] = pd.Categorical(col_df[0], categories=['B', 'C','A'], ordered=True)
col_df[1] = pd.Categorical(col_df[1], categories=['Cat2', 'Cat1'], ordered=True)

# sort values and get the index
print(col_df.sort_values(by=col_df.columns.tolist()).index)
# MultiIndex([('B', 'Cat2', 'c'), # B is before A and Cat2 before Cat1
#             ('B', 'Cat1', 'b'),
#             ('A', 'Cat1', 'a')],
#            )
output = merged_df[col_df.sort_values(by=col_df.columns.tolist()).index]
print(output)  
#      B         A
#   Cat2 Cat1 Cat1
#      c    b    a
# 0    0    0    0
# 1    1    1    1