Question

我有一个看起来像这样的表：

customer_id  purchase_date  category
1            9/1/2018       Electronics
1            9/2/2018       Art
1            9/3/2018       Books
1            9/5/2018       CPG
2            9/2/2018       Books
2            9/4/2018       Electronics
2            9/20/2018      CPG

在这里，我试图得出CPG购买之前的最近购买。为了进一步说明，这是我的步骤：

步骤1.创建一个不在CPG类别中的购买表：

WITH OTHERS AS(
    SELECT customer_id,
           category as others_category,
           purchase_date
    FROM orders o
    WHERE category IN ('Electronics', 'Books', 'Art')
),

第2步。在CPG类别中创建购买表：

CPG AS( 
    SELECT customer_id,
           category as cpg_category,
           purchase_date
    FROM orders o
    WHERE category = 'CPG'
)

第3步：左加入：这就是我卡住的地方。我想生成一个在CPG购买之前具有最近OTHER购买的表格。也就是说，输出应如下所示：

others_category  count_distinct_customers
Electronics            1
Books                  1

理想情况下，我不想使用CTE。 SQL类型是SQL Server 2017。

Answer 1

这就是我在SQL Server 2017中的处理方式，但是，我不确定这是否可以在2005年使用（不幸的是，就像我说的那样，我已经没有2005测试环境了）。我在SQL Server 2008中添加了 think CREATE FUNCTION [dbo].[CantidadDeContratos](@IdNegocio tinyint) RETURNS @tmpCantidadDeContratos TABLE ( CantidadContratos int , NumeroDocumento varchar(20) , CUIT varchar(20) ) AS BEGIN INSERT INTO @tmpCantidadDeContratos SELECT COUNT(con.idAutomovil) CantidadContratos , cli.NumeroDocumento, cli.CUIT FROM Contrato con INNER JOIN Cliente cli ON cli.idAutomovil = con.idAutomovil WHERE cli.idNegocio = @IdNegocio AND con.EstadoContrato4NH = 1 AND con.TipoContrato4NH != 3 GROUP BY cli.NumeroDocumento , cli.CUIT RETURN END GO。当然，“ VTE”在2005年将无法运行，因为APPLY构造函数子句是在2008年添加的（如果我记得，正确），但是，您至少要有一张表进行测试：

VALUES

Answer 2

换句话说，您只想要紧随其后的“ CPG购买”（由同一位客户购买？）的购买。

借助分析功能LEAD()，您可以查找“后续”行中的内容，而不必将数据重新加入自身。

WITH
  orders_with_lookup AS
(
  SELECT
    *,
    LEAD(category) OVER (PARTITION BY customer_id ORDER BY purchase_date)   AS customers_next_purchase_category
  FROM
    orders
)
SELECT
  category,
  COUNT(DISTINCT customer_id)   AS count_distinct_customers
FROM
  orders_with_lookup
WHERE
  customers_next_purchase_category = 'CPG'
GROUP BY
  category
ORDER BY
  category

Answer 3

您可以使用OUTER APPLY（从SQL Server 2005开始可用）将每个CPG订单与其之前购买的OTHERS一起加入，然后可以仅返回该OTHERS订单数据。

我添加了一个DISTINCT，因此，如果两个CPG订单的其他前一个订单相同（因为在它们之间没有输入其他订单），则仅返回一次。

   SELECT DISTINCT others.*
   FROM orders cpg
        OUTER APPLY (SELECT top 1 others.*
                     FROM orders as others
                     WHERE category <> 'CPG' and
                           others.purchase_date < cpg.purchase_date
                           order by others.purchase_date desc) as others
   WHERE category = 'CPG'

Answer 4

尝试一下

;WITH CTE(customer_id , purchase_date , category)
AS
(
SELECT 1,'9/1/2018' ,'Electronics' UNION ALL
SELECT 1,'9/2/2018' ,'Art'         UNION ALL
SELECT 1,'9/3/2018' ,'Books'       UNION ALL
SELECT 1,'9/5/2018' ,'CPG'         UNION ALL
SELECT 2,'9/2/2018' ,'Books'       UNION ALL
SELECT 2,'9/4/2018' ,'Electronics' UNION ALL
SELECT 2,'9/20/2018','CPG'     
)
,CTE2 
AS
(
SELECT customer_id,purchase_date,category, 
       ROW_NUMBER()OVER(PARTITION BY customer_id ORDER BY purchase_date DESC) AS MostRecentPurchase
FROM
(
SELECT customer_id , 
       CAST( purchase_date AS DATE) purchase_date, 
       category
FROM CTE
)dt
)
SELECT Category AS Others_category,
       COUNT(DISTINCT customer_id) AS Count_distinct_customers
FROM  CTE2
WHERE  MostRecentPurchase = 2
GROUP BY category

结果

Others_category     Count_distinct_customers
-----------------------------------------
Books                   1
Electronics             1

Answer 5

这是另一种方法...有很多方法可以给这只猫剥皮。

declare @Something table
(
    customer_id int
    , purchase_date date
    , category varchar(20)
)
insert @Something values
(1, '9/1/2018 ', 'Electronics')
, (1, '9/2/2018 ', 'Art')
, (1, '9/3/2018 ', 'Books')
, (1, '9/5/2018 ', 'CPG')
, (2, '9/2/2018 ', 'Books')
, (2, '9/4/2018 ', 'Electronics')
, (2, '9/20/2018', 'CPG')

, (3, '9/2/2018 ', 'Books') --added customer 3 
, (3, '9/4/2018 ', 'Electronics')
, (3, '9/20/2018', 'CPG')

select category
    , DistinctCustomerCount = count(*)
from
(
    select *
        , RowNum = row_number()over(partition by customer_id, case when category = 'CPG' then 1 else 0 end order by purchase_date desc)
    from @Something
) x
where x.category <> 'CPG'
    and x.RowNum = 1
group by x.category

在最新购买之前选择最新购买

5 个答案: