更新

Question

我经常使用相同的聚合函数组合进行查询。 E.g。

SELECT
    my_id,
    sum(a * weight) / nullif(sum(CASE WHEN a IS NOT NULL THEN weight END), 0) AS a,
    sum(b * weight) / nullif(sum(CASE WHEN b IS NOT NULL THEN weight END), 0) AS b
FROM my_table
GROUP BY my_id

我想避免一遍又一遍地重复相同的表达。使用新函数weighted_avg得到相同的结果会很棒：

SELECT
    my_id,
    weighted_avg(a, weight) AS a,
    weighted_avg(b, weight) AS b
FROM my_table
GROUP BY my_id

我知道的唯一方法就是使用CREATE AGGREGATE中间状态和SFUNC来调用每一行。不幸的是，这比原始查询慢得多，这使得它在我的情况下无法使用。

我想我的理想解决方案看起来像

CREATE AGGREGATE FUNCTION weighted_avg(x float, weight float)
RETURNS float AS $$
    SELECT sum(x * weight) / nullif(sum(CASE WHEN x IS NOT NULL THEN weight END), 0) 
$$ language SQL IMMUTABLE;

并在执行查询时内联。但我找不到Postgres支持的类似内容。

Answer 1

您没有显示测试的聚合函数。这就是我创建它的方式：

create function weighted_avg_acumm (fa float[], x float, weight float)
returns float[] as $$
    select array[
        fa[1] + x * weight,
        fa[2] + weight
    ]::float[]
$$ language sql immutable strict;

create function weighted_avg_acumm_final (fa float[])
returns float as $$
    select fa[1] / fa[2]
$$ language sql immutable strict;

create aggregate weighted_avg (x float, weight float)(
    sfunc = weighted_avg_acumm,
    finalfunc = weighted_avg_acumm_final,
    stype = float[],
    initcond = '{0,0}'
);

更新

我测试了它对我来说也慢得多：

create table t (a int, weight int);
insert into t (a, weight)
select
    nullif(round(random() * 10), 0),
    trunc(random() * 10) + 1
from generate_series(1,1000000)
;

explain analyze
select weighted_avg(a, weight)
from t;
                                                    QUERY PLAN                                                     
-------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=269425.25..269425.26 rows=1 width=8) (actual time=7933.440..7933.440 rows=1 loops=1)
   ->  Seq Scan on t  (cost=0.00..14425.00 rows=1000000 width=8) (actual time=0.018..241.571 rows=1000000 loops=1)
 Planning time: 0.189 ms
 Execution time: 7933.508 ms

explain analyze
select 
    sum(a::numeric * weight) / 
    nullif(sum(case when a is not null then weight end), 0)
from t;
                                                    QUERY PLAN                                                     
-------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=26925.00..26925.02 rows=1 width=8) (actual time=904.852..904.852 rows=1 loops=1)
   ->  Seq Scan on t  (cost=0.00..14425.00 rows=1000000 width=8) (actual time=0.010..127.264 rows=1000000 loops=1)
 Planning time: 0.048 ms
 Execution time: 904.891 ms

如何在没有减速的情况下在PostgreSQL中重用具有聚合的表达式

1 个答案:

更新