如何在 BigQuery 中通过每个用户的线性插值填充不规则缺失的时间序列值？

Question

我的数据缺少时间序列值不规则每个用户，我想要使用 BigQuery 标准 SQL 以一定的间隔通过线性插值对其进行转换。

+------+---------------------+-------+
| name |        time         | value |
+------+---------------------+-------+
| Jane | 2020-11-14 09:01:00 |     3 |
| Jane | 2020-11-14 09:05:00 |     5 |
| Jane | 2020-11-14 09:07:00 |     1 |
| Jane | 2020-11-14 09:09:00 |     8 |
| Jane | 2020-11-14 09:10:00 |     4 |
| Kay  | 2020-11-14 09:01:00 |     7 |
| Kay  | 2020-11-14 09:04:00 |     1 |
| Kay  | 2020-11-14 09:05:00 |    10 |
| Kay  | 2020-11-14 09:09:00 |     6 |
| Kay  | 2020-11-14 09:10:00 |     7 |
+------+---------------------+-------+

我想将其转换如下：

+------+---------------------+-------+-----------------+
| name |        time         | value |                 |
+------+---------------------+-------+-----------------+
| Jane | 2020-11-14 09:01:00 | 3     |                 |
| Jane | 2020-11-14 09:02:00 | 3.5   | <= interpolaetd |
| Jane | 2020-11-14 09:03:00 | 4     | <= interpolaetd |
| Jane | 2020-11-14 09:04:00 | 4.5   | <= interpolaetd |
| Jane | 2020-11-14 09:05:00 | 5     |                 |
| Jane | 2020-11-14 09:06:00 | 3     | <= interpolaetd |
| Jane | 2020-11-14 09:07:00 | 1     |                 |
| Jane | 2020-11-14 09:08:00 | 4.5   | <= interpolaetd |
| Jane | 2020-11-14 09:09:00 | 8     |                 |
| Jane | 2020-11-14 09:10:00 | 4     |                 |
| Kay  | 2020-11-14 09:01:00 | 7     |                 |
| Kay  | 2020-11-14 09:02:00 | 5     | <= interpolaetd |
| Kay  | 2020-11-14 09:03:00 | 3     | <= interpolaetd |
| Kay  | 2020-11-14 09:04:00 | 1     |                 |
| Kay  | 2020-11-14 09:05:00 | 10    |                 |
| Kay  | 2020-11-14 09:06:00 | 9     | <= interpolaetd |
| Kay  | 2020-11-14 09:07:00 | 8     | <= interpolaetd |
| Kay  | 2020-11-14 09:08:00 | 7     | <= interpolaetd |
| Kay  | 2020-11-14 09:09:00 | 6     |                 |
| Kay  | 2020-11-14 09:10:00 | 7     |                 |
+------+---------------------+-------+-----------------+

我能问你一些聪明的解决方案吗？

补充：这是this stackoverflow question 的应用问题。它非常相似，但不同之处在于该数据是时间序列数据，它具有每个用户的名称。

谢谢。

Answer 1

以下是 BigQuery SQL

#standardSQL
select name, time,
    ifnull(value, start_value 
      + (end_value - start_value) / timestamp_diff(end_tick, start_tick, minute) * timestamp_diff(time, start_tick, minute)
    ) as value_interpolated
from (
    select name, time, value,
    first_value(tick ignore nulls ) over win1 as start_tick,
    first_value(value ignore nulls) over win1 as start_value,
    first_value(tick ignore nulls ) over win2 as end_tick,
    first_value(value ignore nulls) over win2 as end_value,
    from (
        select name, time, t.time as tick, value
        from (
            select name, generate_timestamp_array(min(time), max(time), interval 1 minute) times
            from `project.dataset.table`
            group by name
        )
        cross join unnest(times) time 
        left join `project.dataset.table` t 
        using(name, time)
    )
    window 
        win1 as (partition by name order by time desc rows between current row and unbounded following),
        win2 as (partition by name order by time rows between current row and unbounded following)
)     

如果应用于您问题的样本数据 - 输出是

Answer 2

这与您之前的问题没有太大不同。从接受的答案开始，您可以这样做：

select name, time,
    ifnull(value, start_value + (end_value - start_value) / (end_tick - start_tick) * (time - start_tick)) as value_interpolated
from (
    select name, time, value,
    first_value(tick ignore nulls ) over win1 as start_tick,
    first_value(value ignore nulls) over win1 as start_value,
    first_value(tick ignore nulls ) over win2 as end_tick,
    first_value(value ignore nulls) over win2 as end_value,
    from (
        select name, time, t.time as tick, value
        from (
            select name, generate_array(min(time), max(time)) times
            from `project.dataset.table`
            group by name
        )
        cross join unnest(times) time 
        left join `project.dataset.table` t using(name, time)
    )
    window 
        win1 as (partition by name order by time desc rows between current row and unbounded following),
        win2 as (partition by name order by time rows between current row and unbounded following)
)