问题
I need to write a query that shows the result broken down by FormID
that have a value greater than 5 based on the most recent LogDate
.
Based on the most recent LogDate
, if there was a value that was less than 5, it should display values from that point that are greater than 5 as the values under 5 is a 'reset' if you will.
I am essentially looking at recent consecutive LogDate
records that are greater than 5.
Say we have the following record set:
FormID Value LogDate
--------------------------
Form2 6 10/12/19
Form2 7 10/13/19
Form1 8 10/12/19
Form1 12 10/12/19
Form1 3 10/14/19
Form1 8 10/15/19
Form1 6 10/21/19
The following would return the following (please note I like to show the row_num as well:
FormID Value LogDate row_num
----------------------------------
Form2 6 10/12/19 1
Form2 7 10/13/19 2
Form1 8 10/15/19 1
Form1 6 10/21/19 2
Note in the example above, since the following record had a recent value under 5 (value of 3), we need to get the records that are above 5.
Another example:
FormID Value LogDate
Form1 8 10/15/19
Form1 3 10/21/19
RESULT: No result would be shown as there are in recent record that is greater than 5
Another example:
FormID Value LogDate
Form2 4 10/12/19
Form2 3 10/13/19
Form1 16 10/12/19
Form1 3 10/12/19
Form1 3 10/14/19
Form1 8 10/15/19
Form1 12 10/21/19
Result here would be:
FormID Value LogDate row_num
Form1 8 10/15/19 1
Form1 12 10/21/19 2
Another example:
FormID Value LogDate
Form1 12 10/12/19
Form2 13 10/13/19
Result:
FormID Value LogDate row_num
Form1 12 10/12/19 1
Form2 13 10/13/19 2
From my understanding, this can be done with the LAG function but not sure how to put it altogether.
We can do something like the following:
DECLARE @mytable TABLE
(
FormID VARCHAR(50),
[Value] INT,
LogDate DATETIME
)
select t.*,
lag(value) over(partition by formid order by logdate) lag_value
from @mytablet
But not sure how to pull it all together.
回答1:
If I follow you correctly, you can do this with window functions like this:
select
from (
select t.*,
row_number() over(partition by formid order by logdate desc) rn,
sum(case when value > 5 then 1 else 0 end) over(partition by formid order by logdate desc) grp
from mytable t
) t
where rn = grp
The idea is to compare the number of values above 5
to a row number, counting from the most recent value. Rows where the two values are equal can be retained.
回答2:
Find an indicative answer in fiddle.
The reset_calendar
is the dates that a reset happened and is used to filter out the data.
SELECT temp.*,
ROW_NUMBER() OVER (PARTITION BY temp.FormID ORDER BY temp.LogDate) AS Sequence
FROM (
SELECT t.*
FROM t
LEFT JOIN (
SELECT FormID, MAX(LogDate) AS recent_reset
FROM t
WHERE Value<6
GROUP BY FormID) AS reset_calendar
ON t.FormID = reset_calendar.FormID
WHERE t.LogDate > reset_calendar.recent_reset OR reset_calendar.recent_reset IS NULL)temp
回答3:
One method is:
select t.*,
row_number() over (partition by formid order by logdate)
from t
where t.logdate > (select coalesce(max(t.logdate), '2000-01-01')
from t t2
where t2.formid = t.formid and t.value <= 5
);
You can also use window functions:
select t.*,
row_number() over (partition by formid order by logdate)
from (select t.*,
max(case when value <= 5 then logdate end) over (partition by formid) as logdate_5
from t
) t
where logdate_5 is null or
date > logdate_5
order by formid, logdate;
来源:https://stackoverflow.com/questions/64753746/sql-server-find-recent-consecutive-records-that-are-greater-than-5