如何在每一列中找到最常见的值

Question

我有一个 table 看起来像这样：

category    name1    name2    name3    name4    name5

  1         John      Sam     John     Katy      Cat
  1         John      Ivan    Bob      Andrew    Tom
  1         Sam       Ivan    George   Bob       Tom
  2         Jack      Siri    Elsa     Noah      Anna
  2         Jack      Bob     Tomas    Noah      Tom

我需要做的是在每个类别的每个列中找到最常见的值。也就是说，我需要以下结果：

category    name1    name2    name3    name4    name5

  1         John      Ivan    John     Katy      Tom
  2         Jack      Siri    Elsa     Noah      Anna

如果有多个值具有相同的频率，则可以选择其中的任何一个。

到目前为止，我只用这个脚本为一个专栏做到了这一点：

SELECT top(1) category, name1, COUNT(name1) AS freq
FROM data
GROUP BY category, name1
ORDER BY freq DESC

但是我如何为 SQL 服务器中的多个列执行此操作？

Answer 1

首先创建一个CTE，使用COUNT() window 函数return 每个名称在每个类别中出现的次数，然后使用FIRST_VALUE() window 函数为每列获取出现次数最多的名称：

WITH cte AS (
  SELECT *,
         COUNT(*) OVER (PARTITION BY category, name1) count1,
         COUNT(*) OVER (PARTITION BY category, name2) count2,
         COUNT(*) OVER (PARTITION BY category, name3) count3,
         COUNT(*) OVER (PARTITION BY category, name4) count4,
         COUNT(*) OVER (PARTITION BY category, name5) count5
  FROM tablename
)
SELECT DISTINCT category,
       FIRST_VALUE(name1) OVER (PARTITION BY category ORDER BY count1 DESC) name1,
       FIRST_VALUE(name2) OVER (PARTITION BY category ORDER BY count2 DESC) name2,
       FIRST_VALUE(name3) OVER (PARTITION BY category ORDER BY count3 DESC) name3,
       FIRST_VALUE(name4) OVER (PARTITION BY category ORDER BY count4 DESC) name4,
       FIRST_VALUE(name5) OVER (PARTITION BY category ORDER BY count5 DESC) name5
FROM cte

参见demo。

Answer 2

一个选项，有很多重复，但适合您当前的结构...

（尽管在处理同样频繁的名称时它仍然具有相同的 non-deterministic/arbitrary 行为）

WITH
  counted AS
(
  SELECT
    category,
    name1,
    COUNT(*) OVER (PARTITION BY category, name1)  AS name1_freq,
    name2,
    COUNT(*) OVER (PARTITION BY category, name2)  AS name2_freq
  FROM
    yourTable
),
  ranked AS
(
  SELECT
    category,
    name1,
    ROW_NUMBER() OVER (PARTITION BY category ORDER BY name1_freq DESC)  AS name1_rank,
    name2,
    ROW_NUMBER() OVER (PARTITION BY category ORDER BY name2_freq DESC)  AS name2_rank
  FROM
    counted
)
SELECT
  category,
  MAX(CASE WHEN name1_rank = 1 THEN name1 END)  AS name1_most_common,
  MAX(CASE WHEN name2_rank = 1 THEN name2 END)  AS name2_most_common
FROM
  ranked
GROUP BY
  category

如您所见，有很多重复。这就是为什么 SQL 以规范化数据结构为基础的原因。如此之多以至于规范化然后反规范化你的结构可能是一个有效的选择......

WITH
  normalised(category, col, name) AS
(
            SELECT category, 1, name1 FROM yourTable 
  UNION ALL SELECT category, 2, name2 FROM yourTable
),
  counted AS
(
  SELECT
    category, col, name, COUNT(*) AS freq
  FROM
    normalised
  GROUP BY
    category, col, name
),
  ranked AS
(
  SELECT
    category, col, name,
    ROW_NUMBER() OVER (PARTITION BY category, col ORDER BY freq DESC)  AS rank
  FROM
    counted
)
SELECT
  category,
  MAX(CASE WHEN col = 1 THEN name END)  AS name1_most_common,
  MAX(CASE WHEN col = 2 THEN name END)  AS name2_most_common
FROM
  ranked
WHERE
  rank = 1
GROUP BY
  category

Answer 3

如果您不介意行中的结果，您可以逆透视，这样会更简单：

select category, which, name
from (select t.category, v.which, v.name, count(*) as cnt,
             row_number() over (partition by t.category, v.which order by count(*) desc) as seqnum
      from t cross apply
           (values (1, name1), (2, name2), (3, name3), (4, name4), (5, name4)
           ) v(which, name)
      from t
      group by t.category, v.which, v.name
      ) cwn
where seqnum = 1;

如果你想在列中重新透视，你可以重新透视：

with cwn as (
      select t.category, v.which, v.name, count(*) as cnt,
             row_number() over (partition by t.category, v.which order by count(*) desc) as seqnum
      from t cross apply
           (values (1, name1), (2, name2), (3, name3), (4, name4), (5, name4)
           ) v(which, name)
      from t
      group by t.category, v.which, v.name
     )
select category,
       max(case when which = 1 then name end) as name1,
       max(case when which = 2 then name end) as name2,
       max(case when which = 3 then name end) as name3,
       max(case when which = 4 then name end) as name4,
       max(case when which = 5 then name end) as name5
from cwn
where seqnum = 1
group by category

如何在每一列中找到最常见的值

How to find the most common value in each column

sql

tsql

sql-server

window-functions