我有一个包含大约10亿行的数据库，我们需要它来确定集合中的年龄中位数。对十亿行进行排序是困难的，但如果你将可以找到的不同值(年龄范围从0到100)聚合在一起，你可以对这个列表进行排序，并使用一些算术魔术来找到你想要的任何百分位数，如下所示:

with rawData(count_value) as
(
    select p.YEAR_OF_BIRTH
        from dbo.PERSON p
),
overallStats (avg_value, stdev_value, min_value, max_value, total) as
(
  select avg(1.0 * count_value) as avg_value,
    stdev(count_value) as stdev_value,
    min(count_value) as min_value,
    max(count_value) as max_value,
    count(*) as total
  from rawData
),
aggData (count_value, total, accumulated) as
(
  select count_value, 
    count(*) as total, 
        SUM(count(*)) OVER (ORDER BY count_value ROWS UNBOUNDED PRECEDING) as accumulated
  FROM rawData
  group by count_value
)
select o.total as count_value,
  o.min_value,
    o.max_value,
    o.avg_value,
    o.stdev_value,
    MIN(case when d.accumulated >= .50 * o.total then count_value else o.max_value end) as median_value,
    MIN(case when d.accumulated >= .10 * o.total then count_value else o.max_value end) as p10_value,
    MIN(case when d.accumulated >= .25 * o.total then count_value else o.max_value end) as p25_value,
    MIN(case when d.accumulated >= .75 * o.total then count_value else o.max_value end) as p75_value,
    MIN(case when d.accumulated >= .90 * o.total then count_value else o.max_value end) as p90_value
from aggData d
cross apply overallStats o
GROUP BY o.total, o.min_value, o.max_value, o.avg_value, o.stdev_value
;

这个查询取决于你的db支持窗口函数(包括ROWS UNBOUNDED precede)，但如果你没有，这是一个简单的事情，将aggData CTE与自身连接，并将所有先前的总数聚合到' cumulative '列，用于确定哪个值包含指定的预分词。上面的示例计算p10、p25、p50(中位数)、p75和p90。

屁股的

对于一个表站和列lat_n，下面是MySQL代码来获得中位数:

set @rows := (select count(1) from station);
set @v1 := 0;
set @sql1 := concat('select lat_n into @v1 from station order by lat_n asc limit 1 offset ', ceil(@rows/2) - 1);
prepare statement1 from @sql1;
execute statement1;
set @v2 := 0;
set @sql2 := concat('select lat_n into @v2 from station order by lat_n asc limit 1 offset ', ceil((@rows + 1)/2) - 1);
prepare statement2 from @sql2;
execute statement2;
select (@v1 + @v2)/2;

按维度分组的中位数:

SELECT your_dimension, avg(t1.val) as median_val FROM (
SELECT @rownum:=@rownum+1 AS `row_number`,
   IF(@dim <> d.your_dimension, @rownum := 0, NULL),
   @dim := d.your_dimension AS your_dimension,
   d.val
   FROM data d,  (SELECT @rownum:=0) r, (SELECT @dim := 'something_unreal') d
  WHERE 1
  -- put some where clause here
  ORDER BY d.your_dimension, d.val
) as t1
INNER JOIN  
(
  SELECT d.your_dimension,
    count(*) as total_rows
  FROM data d
  WHERE 1
  -- put same where clause here
  GROUP BY d.your_dimension
) as t2 USING(your_dimension)
WHERE 1
AND t1.row_number in ( floor((total_rows+1)/2), floor((total_rows+2)/2) )

GROUP BY your_dimension;

我的代码，高效，没有表或额外的变量:

SELECT
((SUBSTRING_INDEX(SUBSTRING_INDEX(group_concat(val order by val), ',', floor(1+((count(val)-1) / 2))), ',', -1))
+
(SUBSTRING_INDEX(SUBSTRING_INDEX(group_concat(val order by val), ',', ceiling(1+((count(val)-1) / 2))), ',', -1)))/2
as median
FROM table;

这些方法从同一个表中选择两次。如果源数据来自一个昂贵的查询，这是一种避免运行两次的方法:

select KEY_FIELD, AVG(VALUE_FIELD) MEDIAN_VALUE
from (
    select KEY_FIELD, VALUE_FIELD, RANKF
    , @rownumr := IF(@prevrowidr=KEY_FIELD,@rownumr+1,1) RANKR
    , @prevrowidr := KEY_FIELD
    FROM (
        SELECT KEY_FIELD, VALUE_FIELD, RANKF
        FROM (
            SELECT KEY_FIELD, VALUE_FIELD 
            , @rownumf := IF(@prevrowidf=KEY_FIELD,@rownumf+1,1) RANKF
            , @prevrowidf := KEY_FIELD     
            FROM (
                SELECT KEY_FIELD, VALUE_FIELD 
                FROM (
                    -- some expensive query
                )   B
                ORDER BY  KEY_FIELD, VALUE_FIELD
            ) C
            , (SELECT @rownumf := 1) t_rownum
            , (SELECT @prevrowidf := '*') t_previd
        ) D
        ORDER BY  KEY_FIELD, RANKF DESC
    ) E
    , (SELECT @rownumr := 1) t_rownum
    , (SELECT @prevrowidr := '*') t_previd
) F
WHERE RANKF-RANKR BETWEEN -1 and 1
GROUP BY KEY_FIELD

MySQL文档中这一页的注释有以下建议:

-- (mostly) High Performance scaling MEDIAN function per group
-- Median defined in http://en.wikipedia.org/wiki/Median
--
-- by Peter Hlavac
-- 06.11.2008
--
-- Example Table:

DROP table if exists table_median;
CREATE TABLE table_median (id INTEGER(11),val INTEGER(11));
COMMIT;


INSERT INTO table_median (id, val) VALUES
(1, 7), (1, 4), (1, 5), (1, 1), (1, 8), (1, 3), (1, 6),
(2, 4),
(3, 5), (3, 2),
(4, 5), (4, 12), (4, 1), (4, 7);



-- Calculating the MEDIAN
SELECT @a := 0;
SELECT
id,
AVG(val) AS MEDIAN
FROM (
SELECT
id,
val
FROM (
SELECT
-- Create an index n for every id
@a := (@a + 1) mod o.c AS shifted_n,
IF(@a mod o.c=0, o.c, @a) AS n,
o.id,
o.val,
-- the number of elements for every id
o.c
FROM (
SELECT
t_o.id,
val,
c
FROM
table_median t_o INNER JOIN
(SELECT
id,
COUNT(1) AS c
FROM
table_median
GROUP BY
id
) t2
ON (t2.id = t_o.id)
ORDER BY
t_o.id,val
) o
) a
WHERE
IF(
-- if there is an even number of elements
-- take the lower and the upper median
-- and use AVG(lower,upper)
c MOD 2 = 0,
n = c DIV 2 OR n = (c DIV 2)+1,

-- if its an odd number of elements
-- take the first if its only one element
-- or take the one in the middle
IF(
c = 1,
n = 1,
n = c DIV 2 + 1
)
)
) a
GROUP BY
id;

-- Explanation:
-- The Statement creates a helper table like
--
-- n id val count
-- ----------------
-- 1, 1, 1, 7
-- 2, 1, 3, 7
-- 3, 1, 4, 7
-- 4, 1, 5, 7
-- 5, 1, 6, 7
-- 6, 1, 7, 7
-- 7, 1, 8, 7
--
-- 1, 2, 4, 1

-- 1, 3, 2, 2
-- 2, 3, 5, 2
--
-- 1, 4, 1, 4
-- 2, 4, 5, 4
-- 3, 4, 7, 4
-- 4, 4, 12, 4


-- from there we can select the n-th element on the position: count div 2 + 1