Question

我有一个包含三列（CUST_ID，TOPIC，VALUE）的数据模型

data = pd.DataFrame({"CUST_ID":["C1", "C1", "C2", "C3", "C3"],
                       "TOPIC":["TOPIC1", "TOPIC2", "TOPIC2", "TOPIC1", "TOPIC2"],
                       "VALUE":[10, 15, 8, 5, 20]})

我想按CUST_ID分组，将“ TOPIC”列转换为“ TOPIC_a_VALUE”和“ TOPIC_b_VALUE”两列

我知道如何通过SQL来实现，但是如何通过熊猫来实现？

SELECT CUST_ID,
       MAX(CASE WHEN TOPIC = "TOPIC1" THEN VALUE ELSE 0 END) AS TOPIC_a_VALUE
       MAX(CASE WHEN TOPIC = "TOPIC2" THEN VALUE ELSE 0 END) AS TOPIC_b_VALUE
FROM TABLE
GROUP BY CUST_ID

我想要的结果如下，

result = pd.DataFrame({"CUST_ID":["C1", "C2", "C3"],
                       "TOPIC_a_VALUE":[10, np.nan, 5],
                       "TOPIC_b_VALUE":[15, 8, 20]})

Answer 1

IIUC，您需要类似的东西：

df=data.pivot_table(index=['CUST_ID','TOPIC'],columns=['TOPIC']).reset_index()
df.columns=[''.join(col) for col in df.columns.values]
df.loc[df.CUST_ID.duplicated(keep=False)]=df.loc[df.CUST_ID.duplicated(keep=False)].bfill()
df=df.drop_duplicates('CUST_ID')
df=df.drop([col for col in df.columns if 'Key' in col],axis=1).reset_index(drop=True)

print(df)

  CUST_ID   TOPIC  VALUETOPIC1  VALUETOPIC2
0      C1  TOPIC1         10.0         15.0
1      C2  TOPIC2          NaN          8.0
2      C3  TOPIC1          5.0         20.0

Answer 2

也许比其他建议的答案更具可读性，我会同意：

data.groupby(['CUST_ID', 'TOPIC'])['VALUE'].max().unstack()
# Output
#TOPIC   TOPIC1 TOPIC2
#CUST_ID              
#C1        10.0   15.0
#C2         NaN    8.0
#C3         5.0   20.0

如果愿意，您当然可以重命名列：

.rename(columns={'TOPIC1': 'TOPIC_a_VALUE', 'TOPIC2': 'TOPIC_b_VALUE'})

Answer 3

您的查询在SQL中没有意义。我认为您打算这样做：

SELECT CUST_ID,
       MAX(CASE WHEN TOPIC = 'a' THEN VALUE ELSE 0 END) AS TOPIC_a_VALUE
       MAX(CASE WHEN TOPIC = 'b' THEN VALUE ELSE 0 END) AS TOPIC_b_VALUE
FROM TABLE
GROUP BY CUST_ID;

这对Pandas解决方案没有直接帮助，但至少查询是有意义的。

Answer 4

您可以通过以下方式使用groupby：

df=data.pivot_table(index=['CUST_ID','TOPIC'],columns=['TOPIC']).reset_index()
df.columns=[''.join(col) for col in df.columns.values]

df1 = df.groupby('CUST_ID').ffill()\
        .groupby('CUST_ID').last()\
        .reset_index()

清理数据框

df1 = df1.drop(columns = ['TOPIC']).
rename(columns{'VALUETOPIC1':'TOPIC_a_VALUE','VALUETOPIC2':'TOPIC_b_VALUE'})

在其他列中与熊猫一起按+大小写分组

4 个答案: