SQL：快速累积频率查询（postgres）

Question

我希望从我们的数据库中获取累积频率数据。我创建了一个简单的临时表，其中包含我们所看到的所有唯一状态更新计数，以及具有该数量状态更新的用户数量。

     Table "pg_temp_4.statuses_count_tmp"
     Column     |  Type   | Modifiers 
----------------+---------+-----------
 statuses_count | integer | 
 frequency      | bigint  | 
Indexes:
    "statuses_count_idx" UNIQUE, btree (statuses_count)

我当前的查询是：

select statuses_count, frequency/(select * from total_statuses)::float, (select sum(frequency)/(select * from total_statuses)::float AS percentage from statuses_count_tmp WHERE statuses_count <= SCT.statuses_count) AS cumulative_percent  FROM statuses_count_tmp AS SCT ORDER BY statuses_count DESC;

但这需要相当长的时间，并且查询数量增长得很快。因此，对于大约 50,000 行，我正在查看要读取的 50k 阶乘行。坐在这里看着查询逐渐消失，我希望有一个我还没有完成的更好的解决方案。

希望能得到这样的东西：

0       0.26975161      0.26975161
1       0.15306534      0.42281695
2       0.05513516      0.47795211
3       0.03050646      0.50845857
4       0.02064444      0.52910301

Answer 1

假设您有 PostgreSQL 8.4 或更高版本，应该可以使用窗口函数来解决。我猜测 total_statuses 是一个视图或临时表，类似于 select sum(Frequency) from statuses_count_tmp？我在这里将其编写为 CTE，这应该使其在语句持续时间内只计算一次结果：

with total_statuses as (select sum(frequency) from statuses_count_tmp)
select statuses_count,
       frequency / (select * from total_statuses) as frequency,
       sum(frequency) over(order by statuses_count)
           / (select * from total_statuses) as cumulative_frequency
from statuses_count_tmp

如果没有 8.4 的窗口函数，您最好的选择就是迭代地处理数据：

create type cumulative_sum_type as ( statuses_count int, frequency numeric, cumulative_frequency numeric );
create or replace function cumulative_sum() returns setof cumulative_sum_type strict stable language plpgsql as $
declare
  running_total bigint := 0;
  total bigint;
  data_in record;
  data_out cumulative_sum_type;
begin
  select sum(frequency) into total from statuses_count_tmp;
  for data_in in select statuses_count, frequency from statuses_count_tmp order by statuses_count
  loop
    data_out.statuses_count := data_in.statuses_count;
    running_total := running_total + data_in.frequency;
    data_out.frequency = data_in.frequency::numeric / total;
    data_out.cumulative_frequency = running_total::numeric / total;
    return next data_out;
  end loop;
end;
$;
select * from cumulative_sum();