Big Query 中的窗口函数和时差答案

【问题标题】：Window function and time difference in Big QueryBig Query 中的窗口函数和时差
【发布时间】：2016-02-01 18:29:16
【问题描述】：

我有一个大查询表定义为：

+----+----------------+------------+ |编号 |时间 |活动 | +----+----------------+------------+ | 1 | 2015-10-01 16:31:48.000000 |注册 | | 1 | 2015-10-01 16:41:48.000000 | 1_购买 | | 1 | 2015-10-01 16:51:48.000000 | 2_购买 | | 2 | 2015-10-01 16:31:48.000000 |注册 | | 2 | 2015-10-01 16:41:48.000000 | 1_购买 | | 3 | 2015-10-01 16:31:48.000000 |注册 | +----+----------------+------------+

我想计算每个 id 组 (1,2,3) 内的时间差，得到的结果为：

+----+----------------+------------+-- ---------------+---+ |编号 |时间 |活动 |时差 | | +----+----------------+------------+-- ---------------+---+ | 1 | 2015-10-01 16:31:48.000000 |注册 | - | | | 1 | 2015-10-01 16:41:48.000000 | 1_购买 | 00:10:00.000000 | | | 1 | 2015-10-01 16:61:48.000000 | 2_购买 | 00:20:00.000000 | | | 2 | 2015-10-01 16:31:48.000000 |注册 | - | | | 2 | 2015-10-01 16:41:48.000000 | 1_购买 | 00:10:00.000000 | | | 3 | 2015-10-01 16:31:48.000000 |注册 | no_purchase | | +----+----------------+------------+-- ---------------+---+

经过一番研究，我想我需要使用窗口功能......但我想不出任何解决方案。非常感谢任何帮助！最好的，五、

【问题讨论】：

标签： sql time google-bigquery window-functions

【解决方案1】：

是的，您可以为此使用分析窗口函数 - 这是使用 FIRST_VALUE 分析函数的一种方法：

SELECT id, time, event, (time - firsttime) / 60000000 FROM (
SELECT id, time, event, 
       FIRST_VALUE(time) OVER(PARTITION BY id ORDER BY time) AS firsttime FROM
(SELECT 1 id, TIMESTAMP('2015-10-01 16:31:48.000000') time, 'signup' event),
(SELECT 1 id, TIMESTAMP('2015-10-01 16:41:48.000000') time, '1_purchase' event),
(SELECT 1 id, TIMESTAMP('2015-10-01 16:51:48.000000') time, '2_purchase' event),
(SELECT 2 id, TIMESTAMP('2015-10-01 16:31:48.000000') time, 'signup' event),
(SELECT 2 id, TIMESTAMP('2015-10-01 16:41:48.000000') time, '1_purchase' event),
(SELECT 3 id, TIMESTAMP('2015-10-01 16:31:48.000000') time, 'signup' event)
)

【讨论】：

【解决方案2】：

select 
  id, time, event, 
  time(sec_to_timestamp((timestamp_to_sec(timestamp(time)) -     
    timestamp_to_sec(timestamp(prev_time))))) as timedifference,
  (timestamp_to_sec(timestamp(time)) -     
    timestamp_to_sec(timestamp(prev_time)))/60 as timefifference_in_min,

  right('0' + string(datediff(timestamp(time),timestamp(prev_time))),2) + ' ' +
  time(sec_to_timestamp((timestamp_to_sec(timestamp(time)) -     
    timestamp_to_sec(timestamp(prev_time))))) as timedifference_as_dd_hh_mm_ss

from (
  select 
    id, time, event,
    lag(time) over(partition by id order by time) as prev_time
  from (
  select f0_ as id, f1_ as time, f2_ as event from
    (select 1, '2015-10-01 16:31:48.000000', 'signup'),
    (select 1, '2015-10-01 16:41:48.000000', '1_purchase'),
    (select 1, '2015-10-01 16:51:48.000000', '2_purchase'),
    (select 2, '2015-10-01 16:31:48.000000', 'signup'),
    (select 2, '2015-10-01 16:41:48.000000', '1_purchase'),
    (select 3, '2015-10-01 16:31:48.000000', 'signup')
  )
)
order by id, time

【讨论】：

谢谢，还没用过滞后功能:)
但问题是......我只是在几个小时内得到了差异，如果差异超过 1 天，它不会考虑......
您的示例中的时差格式让我认为您不希望 diff 大于 24 小时。看起来情况并非如此，所以我在答案中添加了“timefifference_in_min”。如果您期望/需要某种特定格式 - 请提供示例
在我的回答中添加了 timedifference_as_dd_hh_mm_ss。产生“dd hh:mm:ss”
如果你觉得你得到了你正在寻找的答案 - 你可以投票/接受它:o)