How to get Cumulative data from the same table using SQL? - sql

Question

I have this table
table1
eventid entityid eventdate
----------------------------------------
123 xyz Jan-02-2019
541 xyz Jan-02-2019
234 xyz Jan-03-2019
432 xyz Jan-04-2019
111 xyz Jan-05-2019
124 xyz Jan-06-2019
123 xyz Jan-07-2019
234 xyz Jan-08-2019
432 xyz Jan-09-2019
111 xyz Jan-12-2019
I want to show final result as
entityid interval1 interval2
------------------------------
xyz 2 4
here intervals are in days.
Logic to calculate intervals are :
Ex - event 123 and 234 happens multiple time so date difference between each occurrence as shown below would be added finally into interval1.
Please note - its not necessary 234 would be always in a next row of 123. there could be other events in between.
Formula is
interval1 = datediff(day,eventdate of 123,eventdate of 234) + datediff(day,eventdate of 123,eventdate of 234) + and so on
Same for interval2 but for event 432 & 111.
entityid eventid1 eventid2 event_date_diff
--------------------------------------------
xyz 123 234 1
xyz 123 234 1
xyz 432 111 1
xyz 432 111 3
The challenge here is to find out if event 123 has 234 event or not in upcoming rows (not necessarily in immediate next row) and if its there then find the date difference. If there are any other events between 123-234 then we need to ignore those in between events. Also if 123 appears twice then need latest eventdate for 123.

Answer 1

Let's go over this in terms of your requirements, and build up the necessary pieces. This won't be approached in the order you stated them in, but in an order that makes them easier to understand.
Also if 123 appears twice then need latest eventdate for 123.
This means we need to create a range bounds. This is pretty easy:
NextOccurence AS (SELECT eventId, entityId, eventDate,
LEAD(eventDate) OVER(PARTITION BY eventId, entityId ORDER BY eventDate) AS nextOccurenceDate
FROM Table1)
... this will give us every occurrence of an event, with the next one, if present (these can be limited to just your "source" events, but I'm not bothering with that here).
The challenge here is to find out if event 123 has 234 event or not in upcoming rows (not necessarily in immediate next row) and if its there then find the date difference. If there are any other events between 123-234 then we need to ignore those in between events.
(and you previously mentioned it should be the minimum following date, if there were multiple following events).
For this we need to first map events:
EventMap AS (SELECT 123 AS original, 234 AS follow
UNION ALL
SELECT 432, 111)
... and use this to get the "next" following event in range, in what is partially a greatest-n-per-group query:
SELECT NextOccurence.entityId, NextOccurence.eventId, DATEDIFF(day, NextOccurence.eventDate, Table1.eventDate) AS diff
FROM NextOccurence
JOIN EventMap
ON EventMap.original = NextOccurence.eventId
CROSS APPLY (SELECT TOP 1 Table1.eventDate
FROM Table1
WHERE Table1.entityId = NextOccurence.entityId
AND Table1.eventId = EventMap.follow
AND Table1.eventDate >= NextOccurence.eventDate
AND (Table1.eventDate < NextOccurence.nextOccurenceDate OR NextOccurence.nextOccurenceDate IS NULL)
ORDER BY Table1.eventDate) AS Table1
... at this point, we have something close to your intermediate results table:
| entityId | eventId | diff |
|----------|---------|------|
| xyz | 123 | 1 |
| xyz | 123 | 1 |
| xyz | 432 | 1 |
| xyz | 432 | 3 |
... and what follows afterwards would be a standard PIVOT query to aggregate the results.
The final query ends up looking like this:
WITH NextOccurence AS (SELECT eventId, entityId, eventDate,
LEAD(eventDate) OVER(PARTITION BY eventId, entityId ORDER BY eventDate) AS nextOccurenceDate
FROM Table1),
EventMap AS (SELECT 123 AS original, 234 AS follow
UNION ALL
SELECT 432, 111)
SELECT entityId, [123] AS '123-234', [432] AS '432-111'
FROM (SELECT NextOccurence.entityId, NextOccurence.eventId, DATEDIFF(day, NextOccurence.eventDate, Table1.eventDate) AS diff
FROM NextOccurence
JOIN EventMap
ON EventMap.original = NextOccurence.eventId
CROSS APPLY (SELECT TOP 1 Table1.eventDate
FROM Table1
WHERE Table1.entityId = NextOccurence.entityId
AND Table1.eventId = EventMap.follow
AND Table1.eventDate >= NextOccurence.eventDate
AND (Table1.eventDate < NextOccurence.nextOccurenceDate OR NextOccurence.nextOccurenceDate IS NULL)
ORDER BY Table1.eventDate) AS Table1) AS d
PIVOT (SUM(diff)
FOR eventId IN ([123], [432])
) AS pvt
Fiddle example
...which generates the expected results:
| entityId | 123-234 | 432-111 |
|----------|---------|---------|
| xyz | 2 | 4 |

Answer 2

From what I understood of the question, we are asked to provide occurrences of each eventid per date. However these are to be represented in columns rather than rows.
My approach to this problem is firstly to pivot the data within a cte and then to select the unique value from each column in to the cross apply operator of a query. There may be better ways of doing it but this made the most sense to me.
DECLARE #T TABLE
(
EventId INT,
EntityId NVARCHAR(3),
EventDate DATETIME
);
INSERT INTO #T (EventId, EntityId, EventDate)
SELECT * FROM (VALUES
(123, 'xyz', '2019-01-02'),
(234, 'xyz', '2019-01-03'),
(432, 'xyz', '2019-01-04'),
(111, 'xyz', '2019-01-05'),
(124, 'xyz', '2019-01-06'),
(123, 'xyz', '2019-01-07'),
(234, 'xyz', '2019-01-08'),
(432, 'xyz', '2019-01-09'),
(111, 'xyz', '2019-01-12')
) X (EVENTID, ENTITYID, EVENTDATE);
with cte as (
select EntityId, [123] as Interval1, [234] as Interval2, [432] as Interval3, [111] as
Interval4, [124] as Interval5
from
(
select top 5 EntityId, EventId, min(eventdate) as ordering, count(distinct EventDate)
as
vol from #T
group by EntityId, EventId
order by ordering
) src
PIVOT
(
max(vol)
for EventId in ([123], [234], [432], [111], [124])
) as pvt)
select distinct EntityId, Interval1, Interval2, Interval3, Interval4, Interval5
from (select EntityId from cte) a
cross apply
(select Interval1 from cte where Interval1 is not null) b
cross apply
(select Interval2 from cte where Interval2 is not null) c
cross apply
(select Interval3 from cte where Interval3 is not null) d
cross apply
(select Interval4 from cte where Interval4 is not null) e
cross apply
(select Interval5 from cte where Interval5 is not null) f;

Answer 3

You can use lead() and conditional aggregation for this:
select sum(case when eventid = 123 and next_eventid = 234
then datediff(day, eventdate, next_eventdate)
end) as interval1,
sum(case when eventid = 432 and next_eventid = 111
then datediff(day, eventdate, next_eventdate)
end) as interval2
from (select t.*,
lead(eventid) over (partition by entityid order by eventdate) as next_eventid,
lead(eventdate) over (partition by entityid order by eventdate) as next_eventdate
from t
) t;
Probably the simplest way to handle intervening events is conditional cumulative arithemtic:
select sum(case when eventid = 123 and
then datediff(day, eventdate, next_eventdate_234)
end) as interval1,
sum(case when eventid = 432 and
then datediff(day, eventdate, next_eventdate_111)
end) as interval2
from (select t.*,
min(case when eventid = 234 then eventdate end) over (order by eventdate desc) as next_eventdate_234,
min(case when eventid = 111 then eventdate end) over (order by eventdate desc) as next_eventdate_111
from t
where eventid in (123, 234)
) t
where eventid in (123, 432);