堆疊 DataFrame

使用 pandas 重塑資料

Maria Eugenia Inzaugarat

Data Scientist

列的多重索引（MultiIndex）

具有列 MultiIndex 的 DataFrame

設定索引

churn

   credit_score age  country  num_products exited
0           619  43   France             1    Yes
1           608  34  Germany             0     No
2           502  23   France             1    Yes

設定索引

churn.set_index(['country', 'age'], inplace=True)

              credit_score num_products exited
age  country  
 43   France           619            1    Yes
 34  Germany           608            0     No
 23   France           502            1    Yes

由陣列建立 MultiIndex

new_array = [['yes', 'no', 'yes'], ['no', 'yes', 'yes']]

churn.index = pd.MultiIndex.from_arrays(new_array, names=['member', 'credit_card'])

churn

                    credit_score age  country  num_products exited
member credit_card           
   yes          no           619  43   France             1    Yes
    no         yes           608  34  Germany             0     No
   yes         yes           502  23   France             1    Yes

MultiIndex DataFrame

同時具有列與欄多重索引的 DataFrame

MultiIndex DataFrame

index = pd.MultiIndex.from_arrays([['Wick', 'Wick', 'Shelley', 'Shelley'],
                                   ['John', 'Julien', 'Mary', 'Frank']], 
                      names=['last', 'first'])
columns = pd.MultiIndex.from_arrays([['2019', '2019', '2020', '2020'],
                                     ['age', 'weight', 'age', 'weight']],
                        names=['year', 'feature'])

patients = pd.DataFrame(data, index=index, columns=columns)
patients

            year        2019        2020
         feature  age weight  age weight
   last    first
   Wick     John   25     68   26     72
          Julien   31     72   32     73
Shelley     Mary   41     68   42     69
           Frank   32     75   33     74

.stack() 方法

從兩層列索引的 DataFrame 指向三層列索引的箭頭

stack 方法

.stack() 方法

重新排列欄位的一個層級，將其堆疊為列的最內層索引，得到重塑後的 DataFrame。

用方框標示索引重新排列

堆疊成 Series

churn

   credit_score age  country  num_products exited
0           619  43   France             1    Yes
1           608  34  Germany             0     No
2           502  23   France             1    Yes

churned_stacked = churn.stack()
churned_stacked.head(10)

member  credit_card              
yes     no           credit_score        619
                     age                  43
                     country          France
                     num_products          1
                     churn               Yes
no      yes          credit_score        608
                     age                  34
                     country         Germany
                     num_products          0
                     churn                No

堆疊成 DataFrame

patients

            year        2019        2020
         feature  age weight  age weight
   last    first
   Wick     John   25     68   26     72
          Julien   31     72   32     73
Shelley     Mary   41     68   42     69
           Frank   32     75   33     74

patients_stacked = patients.stack()
patients_stacked

                   year  2019 2020
  last   first  feature
  Wick    John      age    25   26 
                 weight    68   72
        Julien      age    31   32
                 weight    72   73
Shelley  Mary       age    41   42
                 weight    68   69
        Frank       age    32   33
                 weight    75   74

依層級編號堆疊

patients

            year        2019        2020
         feature  age weight  age weight
   last    first
   Wick     John   25     68   26     72
          Julien   31     72   32     73
Shelley     Mary   41     68   42     69
           Frank   32     75   33     74

patients.stack(level=0)

               feature   age weight
   last  first    year
   Wick   John    2019    25     68 
                  2020    26     72
        Julien    2019    31     72
                  2020    32     73
Shelley   Mary    2019    41     68
                  2020    42     69
         Frank    2019    32     75
                  2020    33     74

依層級名稱堆疊

patients

            year        2019        2020
         feature  age weight  age weight
   last    first
   Wick     John   25     68   26     72
          Julien   31     72   32     73
Shelley     Mary   41     68   42     69
           Frank   32     75   33     74

patients.stack(level='year')

               feature   age weight
   last  first    year               
   Wick   John    2019    25     68 
                  2020    26     72
        Julien    2019    31     72
                  2020    32     73
Shelley   Mary    2019    41     68
                  2020    42     69
         Frank    2019    32     75
                  2020    33     74

一起來練習吧！

使用 pandas 重塑資料