Is it possible to create counts by date on historic events table? - sql

Question

I have an events table which contains the date of status changes. What I'm trying to achieve is to produce summary counts for each date, however I'm struggling as it is not a straight count by date but instead a count based on the last time the status changed.
The data is as follows:
------------------------------------------
IT_ID NEW_STATUS OLD_STATUS TIMESTAMP
------------------------------------------
100 4 3 06/05/2019
100 3 2 04/05/2019
200 2 1 03/05/2019
100 2 1 02/05/2019
300 2 1 02/05/2019
200 1 - 01/05/2019
100 1 - 01/05/2019
300 1 - 01/05/2019
-------------------------------------------
I've tried grouping, but this hasn't worked due to the above, SQL below for the straight count.
select max(trunc(timestamp)), new_status ,count(new_status)
from status_hist
where trunc(timestamp) >= '01/01/2019'
group by trunc(timestamp), new_status
Ideally I would like the data in the following format, however the key here is to counts against each date. Note, as no status changes took place on the 05/05/19 then it shows the same of the 04/05/19:
---------------------------------------------------------
Date Status 1 Status 2 Status 3 Status 4
---------------------------------------------------------
06/05/2019 0 2 0 1
05/05/2019 0 2 1 0
04/05/2019 0 2 1 0
03/05/2019 0 3 0 0
02/05/2019 1 2 0 0
01/05/2019 3 0 0 0
--------------------------------------------------------
Any help would be gratefully received.
Thanks

Answer 1

I think about handling this problem by getting the status of each person on each date. That requires a cross join to get the person/dates combinations and then some aggregation:
WITH dates as (
SELECT min_dt + LEVEL - 1 AS dt
FROM (SELECT MIN(ts) AS min_dt, MAX(ts) AS max_dt
FROM test_data
)
CONNECT BY min_dt + LEVEL - 1 <= max_dt
)
SELECT d.dt, i.it_id, max(td.new_status) keep (dense_rank first order by td.ts desc) as status
FROM dates d CROSS JOIN
(SELECT DISTINCT IT_ID FROM test_data) i LEFT JOIN
test_data td
ON td.IT_ID = i.IT_ID AND td.ts <= d.dt
GROUP BY d.dt, i.it_id;
The dates CTE is just calculating all dates. The rest is bringing in the latest status.
This can then be expanded to aggregate (or pivot) the results:
WITH dates as (
SELECT min_dt + LEVEL - 1 AS dt
FROM (SELECT MIN(ts) AS min_dt, MAX(ts) AS max_dt
FROM test_data
)
CONNECT BY min_dt + LEVEL - 1 <= max_dt
),
di as (
SELECT d.dt, i.it_id, max(td.new_status) keep (dense_rank first order by td.ts desc) as status
FROM dates d CROSS JOIN
(SELECT DISTINCT IT_ID FROM test_data) i LEFT JOIN
test_data td
ON td.IT_ID = i.IT_ID AND td.ts <= d.dt
GROUP BY d.dt, i.it_id
)
select dt,
sum(case when status = 1 then 1 else 0 end) as num_1,
sum(case when status = 2 then 1 else 0 end) as num_2,
sum(case when status = 3 then 1 else 0 end) as num_3,
sum(case when status = 4 then 1 else 0 end) as num_4
from di
group by dt
order by dt desc;
Here is a db<>fiddle.

Answer 2

You can do it using windowed aggregation functions:
Oracle Setup:
CREATE TABLE test_data ( IT_ID, NEW_STATUS, OLD_STATUS, "TIMESTAMP" ) AS
SELECT 100, 4, 3, DATE '2019-05-06' FROM DUAL UNION ALL
SELECT 100, 3, 2, DATE '2019-05-04' FROM DUAL UNION ALL
SELECT 200, 2, 1, DATE '2019-05-03' FROM DUAL UNION ALL
SELECT 100, 2, 1, DATE '2019-05-02' FROM DUAL UNION ALL
SELECT 300, 2, 1, DATE '2019-05-02' FROM DUAL UNION ALL
SELECT 200, 1, NULL, DATE '2019-05-01' FROM DUAL UNION ALL
SELECT 100, 1, NULL, DATE '2019-05-01' FROM DUAL UNION ALL
SELECT 300, 1, NULL, DATE '2019-05-01' FROM DUAL;
Query:
SELECT DISTINCT
dt AS "TIMESTAMP",
COUNT( CASE new_status WHEN 1 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
- COUNT( CASE old_status WHEN 1 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
AS Status1,
COUNT( CASE new_status WHEN 2 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
- COUNT( CASE old_status WHEN 2 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
AS Status2,
COUNT( CASE new_status WHEN 3 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
- COUNT( CASE old_status WHEN 3 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
AS Status3,
COUNT( CASE new_status WHEN 4 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
- COUNT( CASE old_status WHEN 4 THEN IT_ID END ) OVER ( ORDER BY dt RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW )
AS Status4
FROM test_data t
RIGHT OUTER JOIN (
SELECT min_dt + LEVEL - 1 AS dt
FROM ( SELECT MIN("TIMESTAMP") AS min_dt,
MAX("TIMESTAMP") AS max_dt
FROM test_data
)
CONNECT BY min_dt + LEVEL - 1 <= max_dt
) c
ON ( c.dt = t."TIMESTAMP" )
ORDER BY "TIMESTAMP" DESC
Output:
TIMESTAMP | STATUS1 | STATUS2 | STATUS3 | STATUS4
:-------- | ------: | ------: | ------: | ------:
06-MAY-19 | 0 | 2 | 0 | 1
05-MAY-19 | 0 | 2 | 1 | 0
04-MAY-19 | 0 | 2 | 1 | 0
03-MAY-19 | 0 | 3 | 0 | 0
02-MAY-19 | 1 | 2 | 0 | 0
01-MAY-19 | 3 | 0 | 0 | 0
db<>fiddle here

Answer 3

You can use the pivot function of SQL.
I don't have an oracle DB to test this:
declare #dates table(Date timestamp(3), NEW_STATUS number(10))
v_StartDate DATE := (SELECT MIN(timestamp) FROM [test].dbo)
v_EndDate DATE := (SELECT MAX(timestamp) FROM [test].dbo)
insert into #dates
SELECT nbr * INTERVAL '1' DAY(5) - 1 + v_StartDate as 'Date', null as NEW_STATUS
FROM ( SELECT ROW_NUMBER() OVER ( ORDER BY c.object_id ) AS Nbr
FROM sys.columns c
) nbrs
WHERE nbr - 1 <= v_EndDate - v_StartDate
SELECT timestamp as 'Date', 1 AS 'Status 1', 2 AS 'Status 2', 3 AS 'Status 3', 4 AS 'Status 4'
FROM
(SELECT Date as 'timestamp', NVL(NVL(d.new_status, t.NEW_STATUS),t2.NEW_STATUS) as new_status
FROM #dates d
left outer join Table_test t on d.Date = t.TIMESTAMP
left outer join Table_test t2 on INTERVAL '-1' DAY(5) +d.Date = t2.TIMESTAMP and NVL(d.new_status, t.NEW_STATUS) is null ) p
PIVOT
(
COUNT (new_status)
FOR new_status IN
( 1, 2, 3, 4 )
) AS pvt
ORDER BY pvt.TIMESTAMP desc
My Microsoft SQL Syntax is:
declare #dates table([Date] datetime, [NEW_STATUS] int)
DECLARE #StartDate DATE = (SELECT MIN(timestamp) FROM [test].[dbo].[Table_test])
DECLARE #EndDate DATE = (SELECT MAX(timestamp) FROM [test].[dbo].[Table_test])
insert into #dates
SELECT DATEADD(DAY, nbr - 1, #StartDate) as 'Date', null as NEW_STATUS
FROM ( SELECT ROW_NUMBER() OVER ( ORDER BY c.object_id ) AS Nbr
FROM sys.columns c
) nbrs
WHERE nbr - 1 <= DATEDIFF(DAY, #StartDate, #EndDate)
SELECT timestamp as 'Date', [1] AS 'Status 1', [2] AS 'Status 2', [3] AS 'Status 3', [4] AS 'Status 4'
FROM
(SELECT Date as 'timestamp', ISNULL(ISNULL(d.new_status, t.NEW_STATUS),t2.NEW_STATUS) as new_status
FROM #dates d
left outer join Table_test t on d.Date = t.TIMESTAMP
left outer join Table_test t2 on DATEADD(DAY,-1,d.Date) = t2.TIMESTAMP and ISNULL(d.new_status, t.NEW_STATUS) is null ) p
PIVOT
(
COUNT (new_status)
FOR new_status IN
( [1], [2], [3], [4] )
) AS pvt
ORDER BY pvt.TIMESTAMP desc