如何根据百分位数过滤 table 然后在 HQL 中随机抽样？

Question

我正在尝试从 table 中随机抽样 200 行，但首先我想对其进行过滤以仅从变量中选取前 1% 的值。

我收到以下错误 -

Error while compiling statement: FAILED: ParseException line 3:31 cannot recognize input near 'select' 'percentile_approx' '(' in expression specification

下面是我的查询-

> with sample_pop as (select * from
> mytable a where
> a.transaction_amount > (select
> percentile_approx(transaction_amount, 0.99) as top1
>                             from mytable) )
> 
> select * from sample_pop  distribute by rand(1) sort by rand(1) limit
> 200;

Answer 1

我不认为 Hive 支持标量子查询的方式与您使用它们的方式相同（仅适用于 IN/EXISTS）。所以将逻辑移动到 FROM 子句：

with sample_pop as (
      select *
      from mytable a cross join
           (select percentile_approx(transaction_amount, 0.99) as top1
            from mytable
           ) aa
      where a.transaction_amount > aa.top1
     )
select * 
from sample_pop distribute by rand(1) 
order by rand(1)
limit 200;

Answer 2

通过以下查询解决了我的问题 -

with sample_pop as (select a.* from 
          (
          select *, cum_dist() over (order by transaction_amount asc) pct
          from mytable
          ) a
where pct >= 0.99
)
select * 
from sample_pop distribute by rand(1) 
order by rand(1)
limit 200;

如何根据百分位数过滤 table 然后在 HQL 中随机抽样？

How to filter table based on percentile and then random sample in HQL?

sql

hive

hql

hiveql

hue