SQL：选择top proportion的列和其他列对应的属性

Question

假设我有这样的数据

group,      city,  user,
nyc_frisbee, nyc,   joe,
nyc_frisbee, nyc,   ben,
nyc_frisbee,  sf,   tim,
sf_cooking,   sf,   tim,
sf_cooking,  atl,   jon,

我的目标是

group,       top_city, prob
nyc_firsbee, nyc,      .66
sf_cooking,  sf,       .5
sf_ccoking,  atl,      .5

边缘情况：在下面这个 table 中，nyc_frisbee 只出现一次，因为只有一个排名靠前的城市。但是，sf_cooking 出现了两次，因为 sf 和 atl 是并列的。

如何在 SQL 中完成此操作？

我目前拥有的：

SELECT group, city, top/total AS prob
FROM (
     SELECT 
         group, 
         city, 
         freq,
         MAX(freq) OVER(PARTITION BY group) AS top,
         SUM(freq) OVER(PARTITION BY group) AS total
     FROM
         (SELECT group, city, COUNT(city) AS freq
          FROM mytable
          GROUP BY group, city) inner_query
     ) outer_query
WHERE outer.freq = outer.top

这不符合我的预期...

group,       top_city, prob
nyc_firsbee, nyc,      .66
nyc_firsbee, sf,       .66
sf_cooking,  sf,       .5
sf_ccoking,  atl,      .5

所以我为每个具有相同概率值的唯一 group/city 组合得到一行。

Answer 1

怎么样：

WITH x as (
  SELECT group, city, CAST(COUNT(*) AS FLOAT)/SUM(COUNT(*)) OVER(PARTITION BY group) AS prob
  FROM mytable
  GROUP BY group, city
)

SELECT x.*
FROM 
  x 
  INNER JOIN 
  (SELECT group, MAX(prob) maxprob FROM x GROUP BY group) y 
  ON 
    x.group = y.group AND
    x.prob = y.maxprob

我们将 group/cities 归结为

group,      city,  prob,
nyc_frisbee, nyc,   0.66,
nyc_frisbee,  sf,   0.33,
sf_cooking,   sf,   0.5,
sf_cooking,  atl,   0.5,

并别名为 X，然后我们将其加入到 group, max(prob) 上的一组自身中。因为 sf_cooking, 0.5 是最大值，join 匹配两次，保持平局，但是 nyc 的最大值是 0.66，匹配一次，不包括 0.33

Answer 2

看看下面的方法是否适合你。先计算概率，然后按概率对组进行排序，最后得到不同的行。

with x as (
select *,
    Count(*) over(partition by group, City) * 1.0 / Count(*) over(partition by group) prob
from t
), r as (
    select *, dense_rank() over(partition by group order by prob desc) rn
    from x
)
select distinct group, city, prob
from r
where rn=1

SQL：选择top proportion的列和其他列对应的属性

SQL: Selecting top proportion of column and corresponding attributes in other columns

sql

presto