BigQuery - 在给定的时间点获取不同的值

Question

我有一个包含用户操作的表格（例如查看页面、点击按钮等）。每行包含一个 user_id、一个日期 (created_on) 和操作的名称。我想创建一个查询，它会为每个日期创建一个嵌套字段，其中包含在该日期之前并包括该日期所采取的不同操作。例如，我有一个名为user_actions 的表：

-------------------------------------
| user_id |    date    |   action   |
-------------------------------------
|    1    | 2018-04-01 |    click   |
|    2    | 2018-04-01 |     view   |
|    1    | 2018-04-02 |     view   |
|    2    | 2018-04-02 |     view   |
|    2    | 2018-04-03 |      buy   |
-------------------------------------

would result in

-------------------------------------
| user_id |    date    |   actions  |
-------------------------------------
|    1    | 2018-04-01 |     click  |
|    2    | 2018-04-01 |     view   |
|    1    | 2018-04-02 |     click  |
|    2    | 2018-04-02 |     view   |
|         |            |     view   |
|    2    | 2018-04-03 |     view   |
|    2    |            |      buy   |
-------------------------------------

在第二个表中，actions 是一个嵌套的重复字段。我知道在某个时间点我可以使用类似于以下内容的内容：

SELECT
    user_id,
    date,
    ARRAY(action)
FROM
    user_actions
GROUP BY
    1,2

但是我不确定如何扩展它以为原始表中的每个日期提供相同的计算，并且只查看 date 字段之前的时间。

任何帮助将不胜感激。谢谢！

Answer 1

创建一个包含在该日期之前（包括该日期）所采取的不同操作的嵌套字段

以下是 BigQuery Standrad SQL

#standardSQL
SELECT user_id, date, 
  ARRAY(
    SELECT DISTINCT action FROM UNNEST(actions) action
  ) actions
FROM (
  SELECT user_id, date, ARRAY_AGG(action) OVER(win) actions
  FROM `project.dataset.table`
  WINDOW win AS (
    PARTITION BY user_id ORDER BY date 
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  )
)

您可以使用您问题中的示例数据进行测试，如以下示例所示

#standardSQL
WITH `project.dataset.table` AS (
  SELECT 1 user_id, '2018-04-01' date, 'click' action UNION ALL
  SELECT 2, '2018-04-01', 'view' UNION ALL
  SELECT 1, '2018-04-02', 'view' UNION ALL
  SELECT 2, '2018-04-02', 'view' UNION ALL
  SELECT 2, '2018-04-03', 'buy' 
)
SELECT user_id, date, 
  ARRAY(
    SELECT DISTINCT action FROM UNNEST(actions) action
  ) actions
FROM (
  SELECT user_id, date, ARRAY_AGG(action) OVER(win) actions
  FROM `project.dataset.table`
  WINDOW win AS (
    PARTITION BY user_id ORDER BY date 
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  )
)
-- ORDER BY date, user_id    

结果

更新

以下版本支持更通用的情况，同一用户在同一天执行多项操作（我意识到这不是我最初回答的情况）

#standardSQL
SELECT user_id, date, 
  ARRAY(
    SELECT DISTINCT action FROM UNNEST(SPLIT(actions)) action
  ) actions
FROM (
  SELECT user_id, date , STRING_AGG(actions) OVER(win) actions
  FROM (
    SELECT user_id, date, STRING_AGG(DISTINCT action) actions
    FROM `project.dataset.table`
    GROUP BY user_id, date
  )
  WINDOW win AS (
    PARTITION BY user_id ORDER BY date 
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  )
)

您可以使用以下示例数据对其进行测试（注意 extyra 行与 activity = 'play' ）

#standardSQL
WITH `project.dataset.table` AS (
  SELECT 1 user_id, DATE '2018-04-01' date, 'click' action UNION ALL
  SELECT 2, '2018-04-01', 'view' UNION ALL
  SELECT 1, '2018-04-02', 'view' UNION ALL
  SELECT 1, '2018-04-02', 'play' UNION ALL
  SELECT 2, '2018-04-02', 'view' UNION ALL
  SELECT 2, '2018-04-03', 'buy' 
)
SELECT user_id, date, 
  ARRAY(
    SELECT DISTINCT action FROM UNNEST(SPLIT(actions)) action
  ) actions
FROM (
  SELECT user_id, date , STRING_AGG(actions) OVER(win) actions
  FROM (
    SELECT user_id, date, STRING_AGG(DISTINCT action) actions
    FROM `project.dataset.table`
    GROUP BY user_id, date
  )
  WINDOW win AS (
    PARTITION BY user_id ORDER BY date 
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  )
)
-- ORDER BY date, user_id

结果