对一列求和的有效方法，其中另一列的 count(value_1) / count(value_2) 大于 x

Question

I have a table of the following structure:

| id | bool | amt |
-------------------
| 1  | 0    | 4   |
| 1  | 1    | 3   |
| 1  | 1    | 5   |
| 2  | 0    | 8   |
| 2  | 1    | 4   |
| 2  | 0    | 4   |

I want to get the sum of the amt but only when the the ratio of bool = 1 / bool = 0 per id is greater than 0.6.

I have successfully done this like this:

SELECT SUM(amt) as total_amt,
FROM table
WHERE id IN (
    SELECT id 
    FROM table 
    GROUP BY id 
    HAVING CAST(SUM(bool) AS DOUBLE) / CAST(COUNT(bool) AS DOUBLE) > 0.6
)

However, my problem is that this is a toy simulation of my actual tables and data, and in reality it is a very large amount of data. When I run this query on all my data, I get errors either saying that the memory limit of the cluster has been reached, or that the execution time has reached the limit. If I remove the WHERE statement which finds the id s satisfying the ratio, then it runs without errors.

Before resorting to having these limits increased, is there any way I can achieve this more efficiently, either in terms of memory, execution time, or both?

Answer 1

You can use two levels of aggregation:

select sum(id_amount)
from (select id, sum(amount) as id_amount,
             avg(case when bool then 1.0 else 0 end) as ratio
      from t
      group by id
     ) t
where ratio > 0.6;

Note: I don't have much experience with Presto. I think you can use:

avg(bool)

or:

avg(bool::int)

instead of the above expression.