如何解决嵌套聚合函数错误？答案

【问题标题】：How to resolve nested aggregate function error?如何解决嵌套聚合函数错误？
【发布时间】：2021-12-30 16:46:30
【问题描述】：

我使用 case 函数对一个项目的聚合计数进行存储，并希望将所有计数的总和除以每个存储桶的值（希望每个存储桶显示为总数的百分比）。但是，我收到一个错误，即我无法嵌套聚合函数，这是我理解的，但需要一些帮助才能找到替代解决方案来实现我的目标。

错误：

Aggregate functions cannot be nested: [COUNT("values".CASE_AGE_CATEGORY)] nested in [SUM(COUNT("values".CASE_AGE_CATEGORY))]

代码：

SELECT Case_Age_Category, COUNT(Case_Age_Category)/sum(count(Case_Age_Category)) as Volume
FROM
(
SELECT DISTINCT(c.CASE_ID),c.CLOSED_AT,
    CASE
        WHEN TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) >0
        AND TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) <24 then '0-24 HOURS'
    
        WHEN TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) >24
        AND TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) <48 then '24-48 HOURS'
    
        WHEN TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) >48
        AND TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) <72 then '48-72 HOURS'
        
        WHEN TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) >72
        AND TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) <96 then '72-96 HOURS'
        
        WHEN TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) >96
        AND TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) <120 then '96-120 HOURS'
        
        else '>5 DAYS'
    End as Case_Age_Category
FROM TEST_DB.STAGING.DW_DECISIV_CASES c inner join DB.Seed.DEALER_MAPPING d on c.DEALER_ID = d.DECISIVDEALERID
WHERE d.DIVISION = 'K' 
    and RO_NUMBER is not NULL 
    and (d.DEALERCATEGORY ILIKE 'DEALER' OR d.DEALERCATEGORY ILIKE 'RTC') 
    and d.DEALERUSAGE ILIKE 'PRODUCTION' 
    and d.OWNERGROUPCODE !='S040'    
)

WHERE CLOSED_AT >= '2021-01-01 00:00:00.000'
GROUP BY Case_Age_Category
ORDER BY Case_Age_Category ASC

当前输出截图：

【问题讨论】：

标签： sql snowflake-cloud-data-platform aggregate-functions case-statement

【解决方案1】：

希望每个桶显示为总数的百分比

将 COUNT(...) 与加窗的 SUM() OVER() 结合以获得所有组的总数：

SELECT Case_Age_Category, 
      DIV0(COUNT(Case_Age_Category), SUM(COUNT(Case_Age_Category)) OVER()) as Volume
FROM
(
  -- ...
) sub
WHERE CLOSED_AT >= '2021-01-01 00:00:00.000'
GROUP BY Case_Age_Category
ORDER BY Case_Age_Category ASC

db<>fiddle demo

【讨论】：

【解决方案2】：

您的时间分段存在一些问题。

您没有捕捉到负数或零小时，因此它们将被归类为“>5”
您与任何地方的 24、48、72、96 小时都不匹配
2秒的时差，被TIMEDIFF标记为1小时可能没问题

SELECT 
    ASSET_CHECKED_IN_AT, 
    STATUS_CHANGED_TO_COMPLETE_HERE_AT,
    TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) as hour_diff,
    CASE
        WHEN hour_diff >0 AND hour_diff <24 then '0-24 HOURS'
        WHEN hour_diff >24 AND hour_diff <48 then '24-48 HOURS'
        WHEN hour_diff >48 AND hour_diff <72 then '48-72 HOURS'
        WHEN hour_diff >72 AND hour_diff <96 then '72-96 HOURS'
        WHEN hour_diff >96 AND hour_diff <120 then '96-120 HOURS'
        else '>5 DAYS'
    End as Case_Age_Category,
    (date_part(epoch_second, STATUS_CHANGED_TO_COMPLETE_HERE_AT::timestamp_ntz)-date_part(epoch_second, ASSET_CHECKED_IN_AT::timestamp_ntz))/3600 as hour_diff_2
FROM VALUES 
    ('2021-01-23 13:45:00','2021-01-23 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:40:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:44:00'),
    ('2021-01-23 13:59:59','2021-01-23 14:00:01'),
    ('2021-01-23 13:45:00','2021-01-24 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-27 13:45:00'),
    ('2021-01-24 13:45:00','2021-01-23 13:45:00')
    v(ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT);

给予：

ASSET_CHECKED_IN_AT	STATUS_CHANGED_TO_COMPLETE_HERE_AT	HOUR_DIFF	CASE_AGE_CATEGORY	HOUR_DIFF_2
2021-01-23 13:45:00	2021-01-23 13:45:00	0	>5 DAYS	0
2021-01-23 13:45:00	2021-01-23 14:40:00	1	0-24 HOURS	0.916667
2021-01-23 13:45:00	2021-01-23 14:44:00	1	0-24 HOURS	0.983333
2021-01-23 13:59:59	2021-01-23 14:00:01	1	0-24 HOURS	0.000556
2021-01-23 13:45:00	2021-01-24 13:45:00	24	>5 DAYS	24
2021-01-23 13:45:00	2021-01-27 13:45:00	96	>5 DAYS	96
2021-01-24 13:45:00	2021-01-23 13:45:00	-24	>5 DAYS	-24

接下来，在 CASE 中评估的 WHEN 子句的顺序是呈现的顺序，因此您可以使用它来删除一半的评估，并且如果您将时间差作为顶级列，您的 SQL 更具可读性。

鉴于这将进入另一层 SQL，额外的列是正常工作的，是使 SQL 更具可读性的一种相当不错的方法。

SELECT 
    ASSET_CHECKED_IN_AT, 
    STATUS_CHANGED_TO_COMPLETE_HERE_AT,
    TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) as hour_diff,
    CASE
        WHEN hour_diff <0 then '-ve HOURS'
        WHEN hour_diff <24 then '0-24 HOURS'
        WHEN hour_diff <48 then '24-48 HOURS'
        WHEN hour_diff <72 then '48-72 HOURS'
        WHEN hour_diff <96 then '72-96 HOURS'
        WHEN  hour_diff <120 then '96-120 HOURS'
        else '>=5 DAYS'
    End as Case_Age_Category
FROM VALUES 
    ('2021-01-23 13:45:00','2021-01-23 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:40:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:44:00'),
    ('2021-01-23 13:59:59','2021-01-23 14:00:01'),
    ('2021-01-23 13:45:00','2021-01-24 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-27 13:45:00'),
    ('2021-01-24 13:45:00','2021-01-23 13:45:00')
    v(ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT);

ASSET_CHECKED_IN_AT	STATUS_CHANGED_TO_COMPLETE_HERE_AT	HOUR_DIFF	CASE_AGE_CATEGORY
2021-01-23 13:45:00	2021-01-23 13:45:00	0	0-24 HOURS
2021-01-23 13:45:00	2021-01-23 14:40:00	1	0-24 HOURS
2021-01-23 13:45:00	2021-01-23 14:44:00	1	0-24 HOURS
2021-01-23 13:59:59	2021-01-23 14:00:01	1	0-24 HOURS
2021-01-23 13:45:00	2021-01-24 13:45:00	24	24-48 HOURS
2021-01-23 13:45:00	2021-01-27 13:45:00	96	96-120 HOURS
2021-01-24 13:45:00	2021-01-23 13:45:00	-24	-ve HOURS

所以现在介绍另一种计算百分比的方法，RATIO_TO_REPORT

所以你可以使用 Lukasz 展示的 DIV0 方法，我在下面解构了它，但 RETIO_TO_REPORT 完成了提升，我以多种形式展示它以表明它对解构/组合版本感到满意：

With data AS (
    SELECT * 
    FROM VALUES 
    ('2021-01-23 13:45:00','2021-01-23 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:40:00'),
    ('2021-01-23 13:45:00','2021-01-23 14:44:00'),
    ('2021-01-23 13:59:59','2021-01-23 14:00:01'),
    ('2021-01-23 13:45:00','2021-01-24 13:45:00'),
    ('2021-01-23 13:45:00','2021-01-27 13:45:00'),
    ('2021-01-24 13:45:00','2021-01-23 13:45:00')
    v(ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT)
), cat_data AS (
    SELECT 
        ASSET_CHECKED_IN_AT, 
        STATUS_CHANGED_TO_COMPLETE_HERE_AT,
        TIMEDIFF('HOUR', ASSET_CHECKED_IN_AT, STATUS_CHANGED_TO_COMPLETE_HERE_AT) as hour_diff,
        CASE
            WHEN hour_diff <0 then '-ve HOURS'
            WHEN hour_diff <24 then '0-24 HOURS'
            WHEN hour_diff <48 then '24-48 HOURS'
            WHEN hour_diff <72 then '48-72 HOURS'
            WHEN hour_diff <96 then '72-96 HOURS'
            WHEN  hour_diff <120 then '96-120 HOURS'
            else '>=5 DAYS'
        End as Case_Age_Category
    FROM data
)
SELECT Case_Age_Category
    ,COUNT(1) as cat_count
    ,SUM(cat_count) OVER() as total_count_a
    ,DIV0(cat_count, total_count_a) as percentage
    ,RATIO_TO_REPORT(cat_count) over () as percentage_2
    ,RATIO_TO_REPORT(COUNT(1)) over () as percentage_3
    ,RATIO_TO_REPORT(COUNT(Case_Age_Category)) over () as percentage_4
FROM cat_data
GROUP BY 1;

给予：

CASE_AGE_CATEGORY	CAT_COUNT	TOTAL_COUNT_A	PERCENTAGE	PERCENTAGE_2	PERCENTAGE_3	PERCENTAGE_4
0-24 HOURS	4	7	0.571428	0.571429	0.571429	0.571429
24-48 HOURS	1	7	0.142857	0.142857	0.142857	0.142857
-ve HOURS	1	7	0.142857	0.142857	0.142857	0.142857
96-120 HOURS	1	7	0.142857	0.142857	0.142857	0.142857

【讨论】：