Pandas：使用 MultiIndex 旋转数据框时出现 ValueError

Question

我有以下代码，其目的是对从 Oracle 数据库检索到的 SQL 表进行透视：

s = "SELECT Country || '_' || Product || '_' || Flow Ref, " + \
        "Country, Product, Flow, zm, Qty " + \
    "FROM Volumes "

#Following will simply pull from db into a dataframe 
df = fb.QueryDB(s)

#Put ZM as column headers
df = df.pivot(values = 'QTY', index = 'REF', columns = 'ZM')

#Format the column headers
df.columns = [x.strftime('%b-%Y') for x in df.columns]

一切都很好，我得到一个数据框，例如：

         Mar-2017    Apr-2017
 Ref
A_B_C      100          110
D_E_F      500          210
G_H_I      310          150

除了现在我想创建一个多索引，如下：

                                    Mar-2017    Apr-2017
 Ref    Country   Product   Flow
A_B_C     A          B       C        100          110
D_E_F     D          E       F        500          210
G_H_I     G          H       I        310          150    

为此，我编辑了将数据框旋转到的行：

df = df.pivot(values = 'QTY', index = ['REF','COUNTRY','PRODUCT','FLOW'], columns = 'ZM')

这会产生以下错误

ValueError: 传递的项目数错误 1859796，位置暗示 4

非常感谢您的帮助。

Answer 1

首先尝试set_index + unstack：

data = {'REF' : ['A_B_C','D_E_F','G_H_I','A_B_C','D_E_F','G_H_I'],
        'COUNTRY' : list('ADGADG'),
        'PRODUCT' : list('BEHBEH'),
        'FLOW' : list('CFICFI'),
         'QTY':[100,500,310,110,210,150],
         'ZM':pd.to_datetime(['2017-03-01'] * 3 + ['2017-04-01'] * 3 )}
df = pd.DataFrame(data)
print (df)
  COUNTRY FLOW PRODUCT  QTY    REF         ZM
0       A    C       B  100  A_B_C 2017-03-01
1       D    F       E  500  D_E_F 2017-03-01
2       G    I       H  310  G_H_I 2017-03-01
3       A    C       B  110  A_B_C 2017-04-01
4       D    F       E  210  D_E_F 2017-04-01
5       G    I       H  150  G_H_I 2017-04-01

df = df.set_index(['REF','COUNTRY','PRODUCT','FLOW', 'ZM'])['QTY']
       .unstack()
       .rename_axis(None, axis=1)
df.columns = df.columns.strftime('%b-%Y')
print (df)
                            Mar-2017  Apr-2017
REF   COUNTRY PRODUCT FLOW                    
A_B_C A       B       C          100       110
D_E_F D       E       F          500       210
G_H_I G       H       I          310       150

---

如果返回错误：

ValueError：索引包含重复条目，无法重塑

需要 pivot_table 和一些聚合函数，如果重复则适用：

data = {'REF' : ['A_B_C','A_B_C','G_H_I','A_B_C','D_E_F','G_H_I'],
        'COUNTRY' : list('AAGADG'),
        'PRODUCT' : list('BBHBEH'),
        'FLOW' : list('CCICFI'),
         'QTY':[100,500,310,110,210,150],
         'ZM':pd.to_datetime(['2017-03-01'] * 3 + ['2017-04-01'] * 3 )}
df = pd.DataFrame(data)
print (df)
  COUNTRY FLOW PRODUCT  QTY    REF         ZM
0       A    C       B  100  A_B_C 2017-03-01 <-dupe COUNTRY,FLOW,PRODUCT,QTY,REF 
1       A    C       B  500  A_B_C 2017-03-01 <-dupe COUNTRY,FLOW,PRODUCT,QTY,REF 
2       G    I       H  310  G_H_I 2017-03-01
3       A    C       B  110  A_B_C 2017-04-01
4       D    F       E  210  D_E_F 2017-04-01
5       G    I       H  150  G_H_I 2017-04-01


df = df.pivot_table(values = 'QTY', 
                    index = ['REF','COUNTRY','PRODUCT','FLOW'], 
                    columns = 'ZM', 
                    aggfunc='mean')

df.columns = df.columns.strftime('%b-%Y')
print (df)
                            Mar-2017  Apr-2017
REF   COUNTRY PRODUCT FLOW                    
A_B_C A       B       C        300.0     110.0
D_E_F D       E       F          NaN     210.0
G_H_I G       H       I        310.0     150.0

或者groupby + aggregate function + unstack：

df = df.groupby(['REF','COUNTRY','PRODUCT','FLOW', 'ZM'])['QTY'].mean().unstack()
df.columns = df.columns.strftime('%b-%Y')
print (df)
                            Mar-2017  Apr-2017
REF   COUNTRY PRODUCT FLOW                    
A_B_C A       B       C        300.0     110.0
D_E_F D       E       F          NaN     210.0
G_H_I G       H       I        310.0     150.0