根据限制规则对交易进行分组

Question

我需要编写一个 SQL 脚本，根据以下规则对每个客户端的事务进行分组

1. 每个组的交易限额为 500000 笔或 365 天，以先到者为准

2. 我有三栏

   transaction_id

   、

   transaction_date

   和

   client_id

3. 结果必须显示

   group_id

   、

   client_id

   、

   number_of_transaction_in_group

   、

   group_start_date

   和

   group_end_date

我尝试了 ChatGPT 中的这个脚本，但它没有返回正确的结果，如下所示：

客户端ID	组ID	群组交易	群组开始日期	组结束日期
8	0	33101	2022-08-14	2023-08-13
8	1	966899	2023-08-14	2024-05-07
8	2	500000	2024-05-07	2024-08-12
8	3	417142	2024-08-12	2024-11-27

预期的结果应该是这样的

客户端ID	组ID	群组交易	群组开始日期	组结束日期
8	0	33101	2022-08-14	2023-08-13
8	1	500000	2023-08-14	2024-05-07
8	2	500000	2024-05-07	2024-08-12
8	3	417142	2024-08-12	2025-08-12
8	4	300000	2025-08-13	2026-08-12

代码：

WITH NumberedTransactions AS 
(
   -- Assign a row number for each transaction per client, ordered by date
   SELECT 
       ClientId,
       TransactionId,
       TransactionDate,
       ROW_NUMBER() OVER (PARTITION BY ClientId ORDER BY TransactionDate) AS RowNum
   FROM 
       Transactions
),
GroupsByTransactionCount AS 
(
   -- Group transactions into sets of 5 based on RowNum
   SELECT
       ClientId,
       TransactionId,
       TransactionDate,
       (RowNum - 1) / 5 AS TransactionGroup
   FROM 
       NumberedTransactions
),
GroupsByDate AS  
(
   -- Assign a start date for each 365-day window for each client
   SELECT
       ClientId,
       TransactionId,
       TransactionDate,
       DATEDIFF(DAY, MIN(TransactionDate) OVER (PARTITION BY ClientId), TransactionDate) / 365 AS DateGroup
   FROM 
       NumberedTransactions
),
FinalGroups AS 
(
   -- Combine both grouping methods into one
   SELECT
       ClientId,
       TransactionId,
       TransactionDate,
       TransactionGroup,
       DateGroup,
       -- Use the larger group number to ensure both conditions are met
       CASE 
           WHEN TransactionGroup >= DateGroup 
               THEN TransactionGroup
           ELSE DateGroup
        END AS FinalGroup
    FROM
        GroupsByTransactionCount 
    INNER JOIN 
        GroupsByDate ON GroupsByTransactionCount.ClientId = GroupsByDate.ClientId
                     AND GroupsByTransactionCount.TransactionId = GroupsByDate.TransactionId
)
SELECT 
    ClientId,
    FinalGroup,
    COUNT(*) AS TransactionsInGroup,
    MIN(TransactionDate) AS GroupStartDate,
    MAX(TransactionDate) AS GroupEndDate
FROM 
    FinalGroups
GROUP BY 
    ClientId, FinalGroup
ORDER BY 
    ClientId, FinalGroup;

Answer 1

这里尝试演示 2 个用例，两个客户端都有 6 个事务

• 客户 1 有一些太旧了，无法考虑
• 客户端 2 的行数过多。

下面的行数限制为 5（而不是 500,000，但在针对实际数据使用时可以更改该限制）。请注意，RowNum 按日期降序排列，以便保留“最新”行。

CREATE TABLE Transactions (
    transaction_id  INT,
    transaction_date    Date,
    client_id   VARCHAR(512)
);

INSERT INTO Transactions (transaction_id, transaction_date, client_id) VALUES
  -- only 3 rows within the 365 days
    ('1', '2022-12-12', '1'), -- too old
    ('2', '2023-02-02', '1'), -- too old
    ('3', '2023-04-04', '1'),  -- too old
    ('4', '2023-10-13', '1'),
    ('5', '2023-11-11', '1'),
    ('6', '2024-10-12', '1'), -- most recent

  -- all rows within 365 days, but limit to 5 rows
    ('7', '2024-01-12', '2'), 
    ('8', '2024-02-02', '2'),
    ('9', '2024-04-04', '2'),
    ('10', '2024-07-12', '2'),
    ('11', '2024-09-01', '2'),
    ('12', '2024-10-12', '2'); -- most recent


SELECT
    client_id
  , dateadd(day,-365,Max(transaction_date)) as min_date
  , Max(transaction_date) as most_recent_date
  , count(*) as rows_in_group
FROM Transactions
GROUP BY client_id

client_id	最小日期	最近日期	组中的行数
1	2023-10-13	2024-10-12	6
2	2023-10-13	2024-10-12	6

select
  d.*
from (
    select
         t.client_id, t.transaction_date, g.min_date
       , ROW_NUMBER() OVER (PARTITION BY t.Client_Id ORDER BY t.Transaction_Date DESC) AS RowNum
    from Transactions as t
    inner join (
           SELECT
               client_id
             , dateadd(day,-365,Max(transaction_date)) as min_date
           FROM Transactions
           GROUP BY client_id
           ) as g on t.client_id = g.client_id
      ) as d
where d.RowNum <= 5 and transaction_date >= min_date

client_id	交易日期	最小日期	行数
1	2024-10-12	2023-10-13	1
1	2023-11-11	2023-10-13	2
1	2023-10-13	2023-10-13	3
2	2024-10-12	2023-10-13	1
2	2024-09-01	2023-10-13	2
2	2024-07-12	2023-10-13	3
2	2024-04-04	2023-10-13	4
2	2024-02-02	2023-10-13	5

小提琴