Pandas像嵌套的国家一样出类拔萃

Question

我在这里碰到一个街区。我必须翻译这个excel公式

=IF(COUNTIFS(advisor!$C:$C,$A3)=0,"0 disclosed",
IF(COUNTIFS(advisor!$C:$C,$A3,advisor!$E:$E,2)>0,"Dependent",
IF(IF(COUNTIFS(advisor!$C:$C,$A3,advisor!$B:$B,"auditor")>0,1,0)+IF(COUNTIFS(advisor!$C:$C,$A3,advisor!$B:$B,"compensation")>0,1,0)=2,"Independent","1 disclosed")))

到目前为止，这是我的解决方案：

df['auditor_compensation'] = np.where(df['id'].isin(df_advisor['company_id']).count() == 0,
                                          '0 disclosed',
                                          np.where(df_advisor['dependent'] == 2, 'dependent',
                                          np.where((np.where(df_advisor['type']=='auditor', 1, 0)+np.where(df_advisor['type']=='compensation', 1, 0)) == 2, 'independent', '1 disclosed')))

我一直得到ValueError: Length of values does not match length of index

df样本数据：公司数据

id      ticker       iq_id     company              auditor_compensation
48299   ENXTAM:AALB  IQ881736   Aalberts Industries       ?
48752   ENXTAM:ABN   IQ1090191  ABN AMRO Group            ?
48865   ENXTAM:ACCEL IQ4492981  Accell Group              ?
49226   ENXTAM:AGN   IQ247906   AEGON                     ?
49503   ENXTAM:AD    IQ373545   Koninklijke               ?

下面是df\U advisor示例数据

id    type          company_id  advisor_company_id  dependent
1     auditor       4829        6091                    1
17    auditor       4875        16512                   1
6359  auditor       4886        7360                    1
37    auditor       4922        8187                    1
4415  compensation  4922        9025                    1
53    auditor       4950        8187                    1

任何帮助都将不胜感激。你知道吗

Answer 1

numpy.where函数不会生成与原始数据帧长度相同的数组或序列。这是因为它试图组合不一致的条件，例如df['id']和df_advisor['dependent']将具有不同的长度。你知道吗

尽管将Excel公式翻译成Pandas/NumPy很有诱惑力，但使用^{}、^{}和^{}可能会更高效、更可读。你知道吗

步骤1：组映射数据帧

df_advisor_grouped = df_advisor.groupby('company_id')\
                               .agg({'type': '|'.join, 'dependent': 'sum'})\
                               .reset_index()

print(df_advisor_grouped)

   company_id                  type  dependent
0        4829               auditor          1
1        4875               auditor          1
2        4886               auditor          1
3        4922  auditor|compensation          2
4        4950               auditor          1

步骤2：与主数据帧合并

# merge dataframes based on key column
res = df.merge(df_advisor_grouped, left_on='id', right_on='company_id', how='left')

步骤3：应用条件逻辑

# define 3 conditions
conds = [res['company_id'].isnull(), res['dependent'].eq(2),
         res['type'].str.contains('auditor') & res['type'].str.contains('compensation')]

# define 3 choices
choices = ['0 disclosed', 'dependent', 'independent'] 

# apply np.select logic, including default argument if 3 conditions are not met
res['auditor_compensation'] = np.select(conds, choices, '1 disclosed')