SQL 查询中的四分位数

Question

我有一个非常简单的table这样的：

CREATE TABLE IF NOT EXISTS LuxLog (
  Sensor TINYINT,
  Lux INT,
  PRIMARY KEY(Sensor)
)

它包含来自不同传感器的数千条日志。

我想为所有传感器配备 Q1 和 Q3。

我可以对每个数据进行一次查询，但对我来说最好对所有传感器进行一次查询（从一次查询中获取 Q1 和 Q3）

我认为这将是一个相当简单的操作，因为四分位数被广泛使用并且是频率计算中的主要统计变量之一。事实上，我发现了大量过于复杂的解决方案，而我希望找到一些简洁明了的解决方案。

谁能给我提示？

编辑：这是我在网上找到的一段代码，但它对我不起作用：

SELECT  SUBSTRING_INDEX(
        SUBSTRING_INDEX(
            GROUP_CONCAT(                 -- 1) make a sorted list of values
                Lux
                ORDER BY Lux
                SEPARATOR ','
            )
        ,   ','                           -- 2) cut at the comma
        ,   75/100 * COUNT(*)        --    at the position beyond the 90% portion
        )
    ,   ','                               -- 3) cut at the comma
    ,   -1                                --    right after the desired list entry
    )                 AS `75th Percentile`
    FROM    LuxLog
    WHERE   Sensor=12
    AND     Lux<>0

我得到 1 作为 return 值，而它应该是一个可以被 10 整除的数字 (10,20,30.....1000)

Answer 1

好吧，使用 NTILE 非常简单，但它是一个 Postgres 函数。你基本上只是做这样的事情：

SELECT value_you_are_NTILING,
    NTILE(4) OVER (ORDER BY value_you_are_NTILING DESC) AS tiles
FROM
(SELECT math_that_gives_you_the_value_you_are_NTILING_here AS value_you_are_NTILING FROM tablename);

这是我在 SQLFiddle 上为您制作的一个简单示例：http://sqlfiddle.com/#!15/7f05a/1

在 MySQL 中，您将使用 RANK...这是用于此的 SQLFiddle：http://www.sqlfiddle.com/#!2/d5587/1（这来自下面链接的问题）

MySQL RANK() 的使用来自此处回答的 Whosebug：Rank function in MySQL

寻找 Salman A 的答案。

Answer 2

应该这样做：

select
    ll.*,
    if (a.position is not null, 1,
        if (b.position is not null, 2, 
        if (c.position is not null, 3, 
        if (d.position is not null, 4, 0)))
    ) as quartile
from
    luxlog ll
    left outer join luxlog a on ll.position = a.position and a.lux > (select count(*)*0.00 from luxlog) and a.lux <= (select count(*)*0.25 from luxlog)
    left outer join luxlog b on ll.position = b.position and b.lux > (select count(*)*0.25 from luxlog) and b.lux <= (select count(*)*0.50 from luxlog)
    left outer join luxlog c on ll.position = c.position and c.lux > (select count(*)*0.50 from luxlog) and c.lux <= (select count(*)*0.75 from luxlog)
    left outer join luxlog d on ll.position = d.position and d.lux > (select count(*)*0.75 from luxlog)
;

完整示例如下：

use example;

drop table if exists luxlog;

CREATE TABLE LuxLog (
  Sensor TINYINT,
  Lux INT,
  position int,
  PRIMARY KEY(Position)
);

insert into luxlog values (0, 1, 10);
insert into luxlog values (0, 2, 20);
insert into luxlog values (0, 3, 30);
insert into luxlog values (0, 4, 40);
insert into luxlog values (0, 5, 50);
insert into luxlog values (0, 6, 60);
insert into luxlog values (0, 7, 70);
insert into luxlog values (0, 8, 80);

select count(*)*.25 from luxlog;
select count(*)*.50 from luxlog;

select
    ll.*,
    a.position,
    b.position,
    if(
        a.position is not null, 1,
        if (b.position is not null, 2, 0)
    ) as quartile
from
    luxlog ll
    left outer join luxlog a on ll.position = a.position and a.lux >= (select count(*)*0.00 from luxlog) and a.lux < (select count(*)*0.25 from luxlog)
    left outer join luxlog b on ll.position = b.position and b.lux >= (select count(*)*0.25 from luxlog) and b.lux < (select count(*)*0.50 from luxlog)
    left outer join luxlog c on ll.position = c.position and c.lux >= (select count(*)*0.50 from luxlog) and c.lux < (select count(*)*0.75 from luxlog)
    left outer join luxlog d on ll.position = d.position and d.lux >= (select count(*)*0.75 from luxlog) and d.lux < (select count(*)*1.00 from luxlog)
;    


select
    ll.*,
    if (a.position is not null, 1,
        if (b.position is not null, 2, 
        if (c.position is not null, 3, 
        if (d.position is not null, 4, 0)))
    ) as quartile
from
    luxlog ll
    left outer join luxlog a on ll.position = a.position and a.lux > (select count(*)*0.00 from luxlog) and a.lux <= (select count(*)*0.25 from luxlog)
    left outer join luxlog b on ll.position = b.position and b.lux > (select count(*)*0.25 from luxlog) and b.lux <= (select count(*)*0.50 from luxlog)
    left outer join luxlog c on ll.position = c.position and c.lux > (select count(*)*0.50 from luxlog) and c.lux <= (select count(*)*0.75 from luxlog)
    left outer join luxlog d on ll.position = d.position and d.lux > (select count(*)*0.75 from luxlog)
;

Answer 3

或者您可以这样使用排名：

select
    ll.*,
    @curRank := @curRank + 1 as rank,
    if (@curRank <= (select count(*)*0.25 from luxlog), 1,
        if (@curRank <= (select count(*)*0.50 from luxlog), 2, 
        if (@curRank <= (select count(*)*0.75 from luxlog), 3, 4))
    ) as quartile
from
    luxlog ll,
    (SELECT @curRank := 0) r
;

这将只为每个四分位数提供一条记录：

select
    x.quartile, group_concat(position)
from (
    select
        ll.*,
        @curRank := @curRank + 1 as rank,
        if (@curRank > 0 and @curRank <= (select count(*)*0.25 from luxlog), 1,
            if (@curRank > 0 and @curRank <= (select count(*)*0.50 from luxlog), 2, 
            if (@curRank > 0 and @curRank <= (select count(*)*0.75 from luxlog), 3, 4))
        ) as quartile
    from
        luxlog ll,
        (SELECT @curRank := 0) r
) x
group by quartile

+ ------------- + --------------------------- +
| quartile      | group_concat(position)      |
+ ------------- + --------------------------- +
| 1             | 10,20                       |
| 2             | 30,40                       |
| 3             | 50,60                       |
| 4             | 70,80                       |
+ ------------- + --------------------------- +
4 rows

编辑：删除后的 sqlFiddle 示例 (http://sqlfiddle.com/#!9/a14a4/17) 如下所示

/*SET @number_of_rows := (SELECT COUNT(*) FROM LuxLog);
SET @quartile := (ROUND(@number_of_rows*0.25));
SET @sql_q1 := (CONCAT('(SELECT "Q1" AS quartile_name , Lux, Sensor FROM LuxLog WHERE Sensor=101 ORDER BY Lux DESC LIMIT 1 OFFSET ', @quartile,')'));
SET @sql_q3 := (CONCAT('( SELECT "Q3" AS quartile_name , Lux, Sensor FROM LuxLog WHERE Sensor=101 ORDER BY Lux ASC LIMIT 1 OFFSET ', @quartile,');'));
SET @sql := (CONCAT(@sql_q1,' UNION ',@sql_q3));
PREPARE stmt1 FROM @sql;
EXECUTE stmt1;*/

Answer 4

参见 SqlFiddle：http://sqlfiddle.com/#!9/accca6/2/6 注意：对于 sqlfiddle，我生成了 100 行，1 到 100 之间的每个整数都有一行，但它是随机顺序（在 excel 中完成）。

这是代码：

SET @number_of_rows := (SELECT COUNT(*) FROM LuxLog);
SET @quartile := (ROUND(@number_of_rows*0.25));
SET @sql_q1 := (CONCAT('(SELECT "Q1" AS quartile_name , Lux, Sensor FROM LuxLog ORDER BY Lux DESC LIMIT 1 OFFSET ', @quartile,')'));
SET @sql_q3 := (CONCAT('( SELECT "Q3" AS quartile_name , Lux, Sensor FROM LuxLog ORDER BY Lux ASC LIMIT 1 OFFSET ', @quartile,');'));
SET @sql := (CONCAT(@sql_q1,' UNION ',@sql_q3));
PREPARE stmt1 FROM @sql;
EXECUTE stmt1;

编辑：

SET @current_sensor := 101;
SET @quartile := (ROUND((SELECT COUNT(*) FROM LuxLog WHERE Sensor = @current_sensor)*0.25));
SET @sql_q1 := (CONCAT('(SELECT "Q1" AS quartile_name , Lux, Sensor FROM LuxLog WHERE Sensor=', @current_sensor,' ORDER BY Lux DESC LIMIT 1 OFFSET ', @quartile,')'));
SET @sql_q3 := (CONCAT('( SELECT "Q3" AS quartile_name , Lux, Sensor FROM LuxLog WHERE Sensor=', @current_sensor,' ORDER BY Lux ASC LIMIT 1 OFFSET ', @quartile,');'));
SET @sql := (CONCAT(@sql_q1,' UNION ',@sql_q3));
PREPARE stmt1 FROM @sql;
EXECUTE stmt1;

基本推理如下：对于四分位数 1，我们希望从顶部获得 25%，因此我们想知道有多少行，即：

SET @number_of_rows := (SELECT COUNT(*) FROM LuxLog);

现在我们知道了行数，我们想知道它的 25% 是多少，就是这一行：

SET @quartile := (ROUND(@number_of_rows*0.25));

然后要找到一个四分位数，我们要按 Lux 对 LuxLog table 进行排序，然后获取行号“@quartile”，为此我们将 OFFSET 设置为 @quartile 表示我们想从行号@quartile 开始我们的 select 并且我们说 limit 1 表示我们只想检索一行。那是：

SET @sql_q1 := (CONCAT('(SELECT "Q1" AS quartile_name , Lux, Sensor FROM LuxLog ORDER BY Lux DESC LIMIT 1 OFFSET ', @quartile,')'));

我们对另一个四分位数做（几乎）相同的事情，但我们不是从顶部开始（从高值到低值），而是从底部开始（它解释了 ASC）。

但是现在我们只是将字符串存储在变量@sql_q1 和@sql_q3 中，所以连接它们，我们合并查询的结果，我们准备查询并执行它.

Answer 5

这是我提出的用于计算四分位数的查询；它在 ~0.04s w/ ~5000 table 行中运行。我包含了 min/max 值，因为我最终使用这些数据来构建四个四分位数范围：

   SELECT percentile_table.percentile, avg(ColumnName) AS percentile_values
    FROM   
        (SELECT @rownum := @rownum + 1 AS `row_number`, 
                   d.ColumnName 
            FROM   PercentileTestTable d, 
                   (SELECT @rownum := 0) r 
            WHERE  ColumnName IS NOT NULL 
            ORDER  BY d.ColumnName
        ) AS t1, 
        (SELECT count(*) AS total_rows 
            FROM   PercentileTestTable d 
            WHERE  ColumnName IS NOT NULL 
        ) AS t2, 
        (SELECT 0 AS percentile 
            UNION ALL 
            SELECT 0.25
            UNION ALL 
            SELECT 0.5
            UNION ALL 
            SELECT 0.75
            UNION ALL 
            SELECT 1
        ) AS percentile_table  
    WHERE  
        (percentile_table.percentile != 0 
            AND percentile_table.percentile != 1 
            AND t1.row_number IN 
            ( 
                floor(( total_rows + 1 ) * percentile_table.percentile), 
                floor(( total_rows + 2 ) * percentile_table.percentile)
            ) 
        ) OR (
            percentile_table.percentile = 0 
            AND t1.row_number = 1
        ) OR (
            percentile_table.percentile = 1 
            AND t1.row_number = total_rows
        )
    GROUP BY percentile_table.percentile;

Fiddle 这里：http://sqlfiddle.com/#!9/58c0e2/1

肯定存在性能问题；如果有人对如何改进这一点有任何反馈，我会很高兴。

示例数据列表：

 3, 4, 4, 4, 7, 10, 11, 12, 14, 16, 17, 18

示例查询输出：

| percentile | percentile_values |
|------------|-------------------|
|          0 |                 3 |
|       0.25 |                 4 |
|        0.5 |              10.5 |
|       0.75 |                15 |
|          1 |                18 |

Answer 6

我将此解决方案与 MYSQL 函数一起使用：

x就是你要的百分位数

array_values 您的 group_concat 值排序并由 ,

分隔

DROP FUNCTION IF EXISTS centile;

delimiter $$
CREATE FUNCTION `centile`(x Text, array_values TEXT) RETURNS text
BEGIN

Declare DIFF_RANK TEXT;
Declare RANG_FLOOR INT;
Declare COUNT INT;
Declare VALEUR_SUP TEXT;
Declare VALEUR_INF TEXT;

SET COUNT = LENGTH(array_values) - LENGTH(REPLACE(array_values, ',', '')) + 1;
SET RANG_FLOOR = FLOOR(ROUND((x) * (COUNT-1),2));
SET DIFF_RANK = ((x) * (COUNT-1)) - FLOOR(ROUND((x) * (COUNT-1),2));

SET VALEUR_SUP = CAST(SUBSTRING_INDEX(SUBSTRING_INDEX(array_values,',', RANG_FLOOR+2),',',-1) AS DECIMAL);
SET VALEUR_INF = CAST(SUBSTRING_INDEX(SUBSTRING_INDEX(array_values,',', RANG_FLOOR+1),',',-1) AS DECIMAL);

/****
    https://fr.wikipedia.org/wiki/Quantile
    x_j+1 + g (x_j+2 - x_j+1)       
***/
RETURN  Round((VALEUR_INF + (DIFF_RANK* (VALEUR_SUP-VALEUR_INF) ) ),2);

END$$

示例：

Select centile(3/4,GROUP_CONCAT(lux ORDER BY lux SEPARATOR ',')) as quartile_3
FROM LuxLog
WHERE Sensor=12 AND Lux<>0

SQL 查询中的四分位数

Quartiles in SQL query

mysql

sql

percentile

quantile