在SQL

Question

在PostgreSQL中，我试图找到具有低于60的值序列，然后是随后出现的高于60的两个连续值的主题。我也对第一个记录的值低于60的时间和第二个记录的值高于60的时间之间的时间长度感兴趣。此事件对于每个主题可能发生多次。

我正在努力寻找如何搜索数量不限的<60，然后再搜索2个> = 60的值。

RowID    SubjectID    Value    TimeStamp    
1           1          65   2142-04-29 12:00:00 
2           1          58   2142-04-30 03:00:00 
3           1          55   2142-04-30 04:00:00
4           1          54   2142-04-30 05:00:00
5           1          55   2142-04-30 06:15:00  
6           1          56   2142-04-30 06:45:00
7           1          65   2142-04-30 07:00:00 
8           1          65   2142-04-30 08:00:00
9           2          48   2142-05-04 03:30:00 
10          2          48   2142-05-04 04:00:00
11          2          50   2142-05-04 05:00:00
12          2          69   2142-05-04 06:00:00
13          2          68   2142-05-04 07:00:00
14          2          69   2142-05-04 08:00:00
15          2          50   2142-05-04 09:00:00
16          2          55   2142-05-04 10:00:00
17          2          50   2142-05-04 10:30:00
18          2          67   2142-05-04 11:00:00
19          2          67   2142-05-04 12:00:00

我目前的尝试使用滞后和超前功能，但是当我不确定需要向前走多远时，我不确定如何使用这些功能。这是一个前瞻一个价值而后一个价值的例子。我的问题是我不知道如何按subjectID进行划分，以期提前查看“ t”个时间点，其中每个主题的“ t”可能不同。

select t.subjectId, t.didEventOccur,
   (next_timestamp - timestamp) as duration
from (select t.*,
         lag(t.value) over (partition by t.subjectid order by t.timestamp) 
as prev_value,
         lead(t.value) over (partition by t.subjectid order by 
t.timestamp) as next_value,
         lead(t.timestamp) over (partition by t.subjectid order by 
 t.timestamp) as next_timestamp
  from t
 ) t
where value < 60 and next_value < 60 and
  (prev_value is null or prev_value >= 60);

我希望得到如下输出：

SubjectID  DidEventOccur Duration 
 1          1             05:00:00
 2          1             03:30:00
 2          1             03:00:00

Answer 1

一个纯粹的 SQL 解决方案，就像您一直在要求的那样：

SELECT subjectid, start_at, next_end_at - start_at AS duration
FROM  (
   SELECT *
        , lead(end_at) OVER (PARTITION BY subjectid ORDER BY start_at) AS next_end_at
   FROM  (
      SELECT subjectid, grp, big
           , min(ts) AS start_at
           , max(ts) FILTER (WHERE big AND big_rn = 2) AS end_at  -- 2nd timestamp
      FROM  (
         SELECT subjectid, ts, grp, big
              , row_number() OVER (PARTITION BY subjectid, grp, big ORDER BY ts) AS big_rn
         FROM  (
            SELECT subjectid, ts
                 , row_number() OVER (PARTITION BY subjectid ORDER BY ts)
                 - row_number() OVER (PARTITION BY subjectid, (value > 60) ORDER BY ts) AS grp
                 , (value > 60) AS big
            FROM   tbl
            ) sub1
         ) sub2
      GROUP  BY subjectid, grp, big
      ) sub3
   ) sub4
WHERE  NOT big                 -- identifies block of values <= 60 ...
AND    next_end_at IS NOT NULL -- ...followed by at least 2 values > 60
ORDER  BY subjectid, start_at;

我省略了无用的列DidEventOccur，而添加了start_at。否则，完全您想要的结果。

db <>提琴here

请改用plpgsql（或任何PL）中的过程解决方案，应该更快。更简单？我会说是的，但这取决于谁来判断。请参阅（有说明以了解该技术和更多链接）：

• How to number consecutive records per island?