如何从一列中的不同值中采样，但只返回另一列中唯一的记录？

Question

我正在努力解决使用 Teradata 的采样问题

下面是数据的格式

ID    Group     Rank
1     dog       1 
1     cat       1 
1     lion      1  
1     elephant  2 
2     dog       1 
2     cat       1 
2     lion      1 
2     elephant  1 
3     dog       1
3     cat       2 
3     lion      1 
3     elephant  1 
4     dog       2 
4     cat       1 
4     lion      1 
4     elephant  1 
... 

理想情况下，我希望为 Group 中的每个条目返回一个样本编号，但只有 ID 中的唯一值。

下面是我生成的当前查询，但这会返回 ID 的重复项

SELECT ID, Group FROM Table 
WHERE rank = 1 
SAMPLE 
 WHEN group = 'dog' then 10
 WHEN group = 'cat' then 10
 WHEN group = 'elephant' then 5
 WHEN group = 'lion' then 5
END

Answer 1

with cte as
 (
   SELECT ID, Group,
      random(1,10000) as rnd -- RANDOM can't be directly used in OLAP-functions
   FROM Table 
   WHERE rank = 1 
 )
SELECT ID, Group
FROM cte
QUALIFY 
   ROW_NUMBER() -- get one random row per ID
   OVER (PARTITION BY ID 
         ORDER BY rnd) = 1
SAMPLE 
 WHEN group = 'dog' then 10
 WHEN group = 'cat' then 10
 WHEN group = 'elephant' then 5
 WHEN group = 'lion' then 5
END

Answer 2

假设您有足够的记录，为每个 id 选择一个随机行，然后从中选择适当的数字：

select t.*
from (select t.*,
             row_number() over (partition by group order by seqnum) as sequm_g
      from (select t.*,
                   row_number() over (partition by id order by random(1, 1000000))
            from t
           ) t
      where seqnum = 1
     ) t
where (group in ('dog', 'cat') and seqnum_g <= 10) or
      (group in ('elephant', 'lion') and seqnum_g <= 5) ;

这并不能保证这些组在结果集中足够大。 但是，如果您有足够的与组大小相关的数据，那么它应该可以工作。