从 postgres 窗口函数中排除空值

Question

对于每一行，我想取最后 20 个非空值的平均值。

窗口函数取下面的 20 行，包括空行并计算平均值，而我想要最后 20 个非空行的平均值。

我尝试了什么？

WITH adv_calculated AS (
    SELECT security_id                                                                               AS sec_id,
           date                                                                                      AS pd,
           AVG(volume)
           OVER (PARTITION BY (security_id) ORDER BY date ROWS BETWEEN 20 PRECEDING AND CURRENT ROW) AS adv
    FROM tbl_financial_index_data
    WHERE volume IS NOT NULL
    GROUP BY security_id, date
)
UPDATE tbl_financial_index_data
SET adv                  = amc.adv
FROM adv_calculated amc
WHERE amc.sec_id = security_id
  AND amc.pd = date

此解决方案适用于 volume 不为 null 的所有行，但它不会计算 volume 为 NULL 的行的 adv/average。 然后对于 null adv 和 volume 行，我必须运行这个非常慢的查询

UPDATE tbl_financial_index_data
SET average_daily_volume =
        (SELECT avg(t.volume)
         FROM (
                  SELECT a.volume
                  FROM tbl_financial_index_data a
                  WHERE a.security_id = tbl_financial_index_data.security_id
                    AND a.date::date <= tbl_financial_index_data.date::date
                    AND a.volume IS NOT NULL
                  ORDER BY a.date DESC
                  LIMIT 21
              ) t)
WHERE volume IS NULL;

我想避免使用第二个查询并使用第一个查询计算所有行的 ADV（因为它要快得多）。

Answer 1

只需省略WHERE 条件WHERE volume IS NOT NULL，那么你应该得到你想要的。

您可以将查询嵌套在外部查询中，以便稍后删除不需要的值：

WITH adv_calculated AS (
   SELECT ...
   FROM (SELECT AVG(volume) OVER (... ROWS BETWEEN ... AND ...),
                ...
         FROM tbl_financial_index_data
         GROUP BY security_id, date
        ) AS subq
   WHERE volume IS NOT NULL
)
SELECT ...

Answer 2

这只是一个变通方法

好的，所以我找到了解决方案。 如果某行的卷为空，则该行的广告将等于前一个非空卷行的广告，因此我必须找到某种方法来为它为空的行结转先前的非空广告.

我能够从 answer 找到一种方法。

下面是继承非空值的代码：

WITH temp_adv_filled_values AS (
    SELECT security_id,
           date,
           FIRST_VALUE(adv) OVER W                  AS adv
    FROM (
             SELECT security_id,
                    date,
                    adv,
                    SUM(CASE WHEN volume IS NULL THEN 0 ELSE 1 END)
                    OVER (PARTITION BY security_id ORDER BY date ) AS value_partition
             FROM tbl_financial_index_data
         ) AS q
        WINDOW W AS (PARTITION BY security_id, value_partition ORDER BY date)
)
UPDATE tbl_financial_index_data tfid
SET adv                  = tmcfv.adv
FROM temp_adv_filled_values tmcfv
WHERE tmcfv.security_id = tfid.security_id
  AND tmcfv.date = tfid.date;