查询以计算多列中的通过值中值

Question

我正在将 SAS 项目转换为 T-SQL，需要计算中位数。

在SAS中可以使用中值函数（例如：SELECT MEDIAN(col1, col2, col3, col4) FROM myTable），但是在SQL服务器中不存在。

我已经研究并找到了一些很好的例子来计算 T-SQL 中的中位数，但我不知道如何在我的查询中使用它，如下所示：

SELECT 
    'test' AS colTest,
    CASE 
        WHEN c1 < 0 
           THEN (10) 
           ELSE 0 
    END AS myMedian --this 10 value is just for example, it should be get from a median function
FROM 
    #tb_test

我一直在尝试下面的例子，得到了这个：

CREATE TABLE #tb_test 
(
     id INT, 
     c1 INT, 
     c2 INT, 
     c3 INT, 
     c4 INT
)

INSERT INTO #tb_test VALUES(1, 2, 4, 6, 8)
INSERT INTO #tb_test VALUES(2, -1, 3, 5, 7)

SELECT * FROM #tb_test;

table是：

|id| c1| c2| c3| c4|
+--+---+---+---+---+
|1 | 2 | 4 | 6 | 8 |
|2 |-1 | 3 | 5 | 7 |

据我计算这些列的中位数是：

--my getMedian query:
WITH cte AS
(
    SELECT 
        id, Names, Results
    FROM
        (SELECT id, c1, c2, c3, c4
         FROM #tb_test) p
    UNPIVOT
        (Results FOR Names IN (c1, c2, c3, c4)
        ) AS UNPIVOT_tb_test
)
SELECT DISTINCT PERCENTILE_CONT(0.5) 
       WITHIN GROUP (ORDER BY Results) 
       OVER (partition BY id) AS Median
FROM cte

这导致：

|Median|
+------+
|4     |
|5     |

我试图在最终查询中包含此 getMedian，但没有成功。所需的最终查询将是：

SELECT 
    'test' AS colTest,
    CASE 
        WHEN c1 < 0 
           THEN ([is it possible to use the getMedian query here?]) 
        ELSE 0 
    END AS myMedian
FROM 
    #tb_test

及其各自期望的结果：

|colTest | myMedian |
+--------+----------+
|test    | 4        |
|test    | 5        |

有人知道我该怎么做吗？

提前致谢。

Answer 1

看来你挺接近的，我简单加了个join就得到了下面的，是你想要的吗？

WITH cte AS
(
    SELECT 
        id, Names, Results
    FROM
        (SELECT id, c1, c2, c3, c4
         FROM tb_test) p
    UNPIVOT
        (Results FOR Names IN (c1, c2, c3, c4)
        ) AS UNPIVOT_tb_test
), medians as (
  SELECT distinct id, PERCENTILE_CONT(0.5) 
         WITHIN GROUP (ORDER BY Results) 
         OVER (partition BY id) AS Median
  FROM cte
)
select t1.*
  , t2.Median -- add here whatever condition you wish, e.g. case c1<0, etc.
from tb_test as t1 inner join medians as t2 on (t1.id=t2.id)

Answer 2

该功能确实存在。它只需要 apply:

select t.*, m.median
from #test t cross apply
     (select top (1) percentile_cont(0.5) within group (order by c) over () as median
      from (values (t.c1), (t.c2), (t.c3), (t.c4)) v(c)
     ) m;

这种有点作弊。该函数实际上是一个 window 函数，而不是聚合函数。 select top (1) 只是 returns 一行。

Answer 3

According to @AaronBertrand，计算中位数最快的方法是使用 OFFSET/FETCH。 它的作用是 select 一个或通过对值的总数进行计算并取平均值来获得两个中间值。

您可以在 SELECT:

的子查询中执行此操作

SELECT
    t.*,
    Median = CASE WHEN c1 < 0 THEN (
      SELECT AVG(m.n * 1.0) FROM (
        SELECT n
        FROM (VALUES (t.c1),(t.c2),(t.c3),(t.c4)) v(n)
        ORDER BY v.n
        OFFSET (4 - 1) / 2 ROWS    -- 4 is the number of values
        FETCH NEXT 1 + (1 - 4 % 2) ROWS ONLY  -- again 4
      ) m
    )
    ELSE 0 END
FROM #tb_test t

或者您可以将它放在 APPLY 中，然后再 select


SELECT
    t.*,
    Median = CASE WHEN c1 < 0 THEN m.Median ELSE 0 END
FROM #tb_test t
CROSS APPLY (
    SELECT AVG(m.n * 1.0) FROM (
        SELECT Median = n
        FROM (VALUES (t.c1),(t.c2),(t.c3),(t.c4)) v(n)
        ORDER BY v.n
        OFFSET (4 - 1) / 2 ROWS    -- 4 is the number of values
        FETCH NEXT 1 + (1 - 4 % 2) ROWS ONLY  -- again 4
    ) m
) m

查询以计算多列中的通过值中值

Query to calculate median passing values in many columns

sql

sql-server

tsql

median