如何在 SQL Bigquery 中的另一个事件之前计算特定事件的数量？

Question

我有一个包含日期、事件和用户的表格。有一个名为“A”的事件。我想找出特定事件在 Sql Bigquery 中事件“A”之前和之后发生了多少次。例如，

   User           Date             Events
    123          2018-02-13            X.Y.A
    123          2018-02-12            X.Y.B
    134          2018-02-10            Y.Z.A
    123          2018-02-11            A
    123          2018-02-01            X.Y.Z
    134          2018-02-05            X.Y.B
    134          2018-02-04            A

输出会是这样的

User       Event    Before   After
123          A      1        3
134          A      0        1

我必须计算的事件包含一个特定的前缀。意味着我必须检查以（ X.Y.then 一些事件名称）开头的事件。所以，X.Y.SomeEvent 是我必须设置计数器的事件。有什么建议吗？

Answer 1

以下是 BigQuery SQL

#standardSQL
SELECT user, event, before, after 
FROM (
  SELECT user, event, 
    COUNTIF(event LIKE 'X.Y.%') OVER(PARTITION BY user ORDER BY dt ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING) before,
    COUNTIF(event LIKE 'X.Y.%') OVER(PARTITION BY user ORDER BY dt ROWS BETWEEN 1 FOLLOWING AND UNBOUNDED FOLLOWING ) after
  FROM `project.dataset.events`
)
WHERE event = 'A'
-- ORDER BY user

您可以在问题中使用虚拟数据对其进行测试

#standardSQL
WITH `project.dataset.events` AS (
  SELECT 123 user, '2018-02-13' dt, 'X.Y.A' event UNION ALL
  SELECT 123, '2018-02-12', 'X.Y.B' UNION ALL
  SELECT 123, '2018-02-11', 'A' UNION ALL
  SELECT 134, '2018-02-10', 'Y.Z.A' UNION ALL
  SELECT 134, '2018-02-05', 'X.Y.B' UNION ALL
  SELECT 134, '2018-02-04', 'A' UNION ALL
  SELECT 123, '2018-02-01', 'X.Y.Z' 
)
SELECT user, event, before, after 
FROM (
  SELECT user, event, 
    COUNTIF(event LIKE 'X.Y.%') OVER(PARTITION BY user ORDER BY dt ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING) before,
    COUNTIF(event LIKE 'X.Y.%') OVER(PARTITION BY user ORDER BY dt ROWS BETWEEN 1 FOLLOWING AND UNBOUNDED FOLLOWING ) after
  FROM `project.dataset.events`
)
WHERE event = 'A'
ORDER BY user

Answer 2

用户窗口函数用于查找“A”出现的日期。然后使用条件聚合来统计前后的事件：

select userid,
       sum(case when date < a_date and event like 'X.Y%' then 1 else 0 end) as before,
       sum(case when date > a_date and event like 'X.Y%' then 1 else 0 end) as before
from (select t.*,
             min(case when event = 'A' then date end) over (partition by userid) as a_date
      from t
     ) t
group by userid