我需要按日期计算库存中存在的产品数量。但是，在数据库中，仅当消费者查看产品时才记录该产品。
例如考虑以下基本表结构：
date | productId | views
July 1 | A | 8
July 2 | A | 6
July 2 | B | 4
July 3 | A | 2
July 4 | A | 8
July 4 | B | 6
July 4 | C | 4
July 5 | C | 2
July 10 | A | 17
使用以下查询，我尝试确定给定日期的库存产品数量。
select date, count(distinct productId) as Inventory, sum(views) as views
from (
select date, productId, count(*) as views
from SomeTable
group by date, productID
order by date asc, productID asc
)
group by date
这是输出
date | Inventory | views
July 1 | 1 | 8
July 2 | 2 | 10
July 3 | 1 | 2
July 4 | 3 | 18
July 5 | 1 | 2
July 10 | 1 | 17
由于缺少行，我的输出不能准确反映库存中有多少产品。
对库存的正确理解如下：
-产品A从7月1日至7月10日在库存中。
-产品B从7月2日至7月4日在库存中。
-产品C的库存从7月4日到7月5日。
正确的SQL输出应为：
date | Inventory | views
July 1 | 1 | 8
July 2 | 2 | 10
July 3 | 2 | 2
July 4 | 3 | 18
July 5 | 2 | 2
July 6 | 1 | 0
July 7 | 1 | 0
July 8 | 1 | 0
July 9 | 1 | 0
July 10 | 1 | 17
如果您继续关注，请允许我确认我很乐意将“广告资源中”定义为第一个视图和最后一个视图之间的日期差。
我遵循了以下错误过程：
首先，我创建了一个表，该表是每个productID和每个日期的笛卡尔乘积。 ''' 日期为（ 选择日期 来自SomeTable 按日期分组 ）， 产品为（ 选择productId 来自SomeTable 按productId分组 ） 选择Dates.date，Products.productId from Dates交叉加入产品 '''
然后我尝试进行正确的外部联接，以将其减少为丢失的记录：
with Records as (
select date, productId, count(*) as views
from SomeTable
group by date, productId
),
Cartesian as (
{See query above}
)
Select Cartesian.date, Cartesian.productId, 0 as views #for upcoming union
from Cartesian right outer join Records
on Cartesian.date = Records.date
where Records.productId is null
然后在手边缺少行，将它们合并回记录。 这样做会产生一个新问题：多余的行。
date | productId | views
July 1 | A | 8
July 1 | B | 0
July 1 | C | 0
July 2 | A | 6
July 2 | B | 4
July 2 | C | 0
July 3 | A | 2
July 3 | B | 0
July 3 | C | 0
July 4 | A | 8
July 4 | B | 6
July 4 | C | 4
July 5 | A | 2
July 5 | B | 0
July 5 | C | 0
July 6 | A | 0
July 6 | B | 0
July 6 | C | 0
July 7 | A | 0
July 7 | B | 0
July 7 | C | 0
July 8 | A | 0
July 8 | B | 0
July 8 | C | 0
July 9 | A | 0
July 9 | B | 0
July 9 | C | 0
July 10 | A | 17
July 10 | B | 0
July 10 | C | 0
当我运行我的简单查询时
select date, count(distinct productId) as Inventory, sum(views) as views
在那张桌子上，我又得到了错误的输出：
date | Inventory | views
July 1 | 3 | 8
July 2 | 3 | 10
July 3 | 3 | 2
July 4 | 3 | 18
July 5 | 3 | 2
July 6 | 3 | 0
July 7 | 3 | 0
July 8 | 3 | 0
July 9 | 3 | 0
July 10 | 3 | 17
我的下一个想法是遍历每个productId，确定它的第一个和最后一个日期，然后与Cartesian表结合起来，条件是Cartesian.date介于每个特定产品的第一个和最后一个日期之间。
必须有一种更简单的方法来执行此操作。谢谢。
答案 0 :(得分：1)
以下是用于BigQuery标准SQL
#standardSQL
WITH dates AS (
SELECT day FROM (
SELECT MIN(day) min_day, MAX(day) max_day
FROM `project.dataset.table`
), UNNEST(GENERATE_DATE_ARRAY(min_day, max_day, INTERVAL 1 DAY)) day
), ranges AS (
SELECT productId, MIN(day) min_day, MAX(day) max_day
FROM `project.dataset.table` t
GROUP BY productId
)
SELECT day, COUNT(DISTINCT productId) Inventory, SUM(IFNULL(views, 0)) views
FROM dates d, ranges r
LEFT JOIN `project.dataset.table` USING(day, productId)
WHERE day BETWEEN min_day AND max_day
GROUP BY day
如果要应用到您的问题的示例数据中，如下面的示例
#standardSQL
WITH `project.dataset.table` AS (
SELECT DATE '2019-07-01' day, 'A' productId, 8 views UNION ALL
SELECT '2019-07-02', 'A', 6 UNION ALL
SELECT '2019-07-02', 'B', 4 UNION ALL
SELECT '2019-07-03', 'A', 2 UNION ALL
SELECT '2019-07-04', 'A', 8 UNION ALL
SELECT '2019-07-04', 'B', 6 UNION ALL
SELECT '2019-07-04', 'C', 4 UNION ALL
SELECT '2019-07-05', 'C', 2 UNION ALL
SELECT '2019-07-10', 'A', 17
), dates AS (
SELECT day FROM (
SELECT MIN(day) min_day, MAX(day) max_day
FROM `project.dataset.table`
), UNNEST(GENERATE_DATE_ARRAY(min_day, max_day, INTERVAL 1 DAY)) day
), ranges AS (
SELECT productId, MIN(day) min_day, MAX(day) max_day
FROM `project.dataset.table` t
GROUP BY productId
)
SELECT day, COUNT(DISTINCT productId) Inventory, SUM(IFNULL(views, 0)) views
FROM dates d, ranges r
LEFT JOIN `project.dataset.table` USING(day, productId)
WHERE day BETWEEN min_day AND max_day
GROUP BY day
-- ORDER BY day
结果是
Row day Inventory views
1 2019-07-01 1 8
2 2019-07-02 2 10
3 2019-07-03 2 2
4 2019-07-04 3 18
5 2019-07-05 2 2
6 2019-07-06 1 0
7 2019-07-07 1 0
8 2019-07-08 1 0
9 2019-07-09 1 0
10 2019-07-10 1 17