MEDIAN CUME_DIST PERCENT_RANK PERCENTILE_CONT PERCENTILE_DISC

Connected to Oracle Database 11g Enterprise Edition Release 11.2.0.1.0 
Connected as [email protected]:1521/orcl

MEDIAN

中位数

以0到1之间的分数形式返回某个值在数据分区中的排名
CUME_DIST	(0,1]	取不到0
PERCENT_RANK	[0,1]	0各1都可以取到

接收一个0到1之间的几率值
PERCENTILE_CONT	连续分布模型	按percent_rank返回值计算对应值的内插值百分比
PERCENTILE_DISC	离散分布模型	升序时，取比所传递的参数cume_dist值更大的值
		降序时，取比所传递的参数cume_dist值更小的值

MEDIAN

SQL> SELECT MEDIAN(SAL) FROM EMP WHERE DEPTNO = 20;
MEDIAN(SAL)
-----------
       2975

PERCENTILE_CONT

SQL> SELECT PERCENTILE_CONT(0.5) WITHIN GROUP(ORDER BY SAL)  FROM EMP WHERE DEPTNO = 20;


PERCENTILE_CONT(0.5)WITHINGROU
------------------------------
                          2975

对比查看CUME_DIST 与 PERCENT_RANK

SELECT '1' AS u,
       emp.ename,
       emp.sal,
       cume_dist() over(ORDER BY sal) AS cume_dist,
       percent_rank() over(ORDER BY sal) AS percent_rank1,
       percent_rank() over(ORDER BY sal DESC) AS percent_rank2
  FROM (SELECT deptno,
               CASE WHEN ename = 'SCOTT' THEN 4000 ELSE sal END AS sal,
               ename
          FROM emp
         WHERE deptno = 20
         ORDER BY sal, ename) emp
 WHERE deptno = 20
UNION ALL
SELECT '2' AS union_1,
       emp.ename,
       emp.sal,
       cume_dist() over(ORDER BY sal) AS cume_dist,
       percent_rank() over(ORDER BY sal) AS percent_rank1,
       percent_rank() over(ORDER BY sal DESC) AS percent_rank2
  FROM emp
 WHERE deptno = 20
 ORDER BY 1, 3, 2;

对比查看 PERCENTILE_CONT PERCENTILE_DISC

CREATE OR REPLACE VIEW v AS 
SELECT deptno,
       CASE
         WHEN ename = 'SCOTT' THEN
          4000
         ELSE
          sal
       END AS sal,
       ename,
       row_number() over(ORDER BY sal, ename) AS seq
  FROM emp
 WHERE deptno = 20;

SELECT 0.2 AS pct,
       percentile_cont(0.2) within GROUP(ORDER BY sal) AS percentile_cont,
       percentile_disc(0.2) within GROUP(ORDER BY sal) AS percentile_disc
  FROM v
 WHERE deptno = 20
UNION ALL
SELECT 0.4 AS pct,
       percentile_cont(0.4) within GROUP(ORDER BY sal) AS percentile_cont,
       percentile_disc(0.4) within GROUP(ORDER BY sal) AS percentile_disc
  FROM v
 WHERE deptno = 20
UNION ALL
SELECT 0.45 AS pct,
       percentile_cont(0.45) within GROUP(ORDER BY sal) AS percentile_cont,
       percentile_disc(0.45) within GROUP(ORDER BY sal) AS percentile_disc
  FROM v
 WHERE deptno = 20
UNION ALL
SELECT 0.5 AS pct,
       percentile_cont(0.5) within GROUP(ORDER BY sal) AS percentile_cont,
       percentile_disc(0.5) within GROUP(ORDER BY sal) AS percentile_disc
  FROM v
 WHERE deptno = 20
UNION ALL
SELECT 0.55 AS pct,
       percentile_cont(0.55) within GROUP(ORDER BY sal) AS percentile_cont,
       percentile_disc(0.55) within GROUP(ORDER BY sal) AS percentile_disc
  FROM v
 WHERE deptno = 20

为了便于分析我们把上面的数据放在一起查看

	SELECT (0.4 - 0.25) / (0.5 - 0.25) * (2975 - 1100) + 1100 from dual
ename	sal	cume_dist	percent_rank asc	percent_rank desc		PERCENTILE_CONT	PERCENTILE_DISC
SMITH	800	0.2	0	1	0.2	1040	800
ADAMS	1100	0.4	0.25	0.75	0.4	2225	1100
JONES	2975	0.6	0.5	0.5	0.45	2600	2975
FORD	3000	0.8	0.75	0.25	0.5	2975	2975
SCOTT	4000	1	1	0	0.55	2980	2975
SMITH	800	0.2	0	1
ADAMS	1100	0.4	0.25	0.75
JONES	2975	0.6	0.5	0.5
FORD	3000	1	0.75	0
SCOTT	3000	1	0.75	0

注意上面的PERCENTILE_CONT当参数为0.4时，因percent_rank 无对应的值。所以是按内插百分比取值。

计算方法为 SELECT (0.4 - 0.25) / (0.5 - 0.25) * (2975 - 1100) + 1100 from dual

而PERCENTILE_DISC取值在0.45 0.5 0.55时因cume_dist没有对应的值。所以取的是0.6对应的值2975

CSDN表格居然会换行。大家可以复制到word里再看。