Show data that takes into consideration consecutive weeks [closed] - sql

Question

Closed. This question needs details or clarity. It is not currently accepting answers.
Want to improve this question? Add details and clarify the problem by editing this post.
Closed 1 year ago.
Improve this question
Hello I have such a situation I basically need to write a SQL code for such a statement
select *,
case when 'Issue' IN ('Overforecasted', 'Underforecasted') AND 'Start Date' DISTINCT 3 dates THEN 'Issue exists for 3 weeks in a row'
FROM Merged;
I know this is not a proper SQL format but does someone know how it can be edited?
Per one DMDUNIT check if it has 3 issues in column "Issues" and later check if it has 3 different start dates. If it has 3 issues ('Overforecasted', "Underforecasted") and 3 different dates for the same DMDUNIT I need to return it in a new column (end as "3InARow")
The current edited draft
SET ARITHABORT OFF
SET ANSI_WARNINGS OFF;
;WITH Forecast AS (
SELECT LOC, DMDUNIT, STARTDATE, TOTFCST
FROM SCPOMGR.FCSTPERFSTATIC
WHERE STARTDATE >= '2021-11-24'
), Actuals AS (
SELECT LOC, DMDUNIT, DMDPostDate, HistoryQuantity
FROM SCPOMGR.HISTWIDE_CHAIN
WHERE DMDPostDate >= '2021-11-24'
), Merged as (
select
COALESCE(f.LOC, a.LOC) AS LOC,
COALESCE(f.DMDUNIT, a.DMDUNIT) AS DMDUNIT,
COALESCE(f.STARTDATE, a.DMDPostDate) AS "Start Date",
SUM(F.TOTFCST) AS "Forecast",
SUM(a.HistoryQuantity) AS "Actuals",
SUM(ABS(a.HistoryQuantity) - f.TOTFCST) AS "Abs Error",
(1 - HistoryQuantity - TOTFCST) / HistoryQuantity as "FA%",
SUM(a.HistoryQuantity) / SUM(f.TOTFCST) AS "Bias",
CASE
WHEN TOTFCST > HistoryQuantity THEN 'Overforecasted'
WHEN TOTFCST < HistoryQuantity THEN 'Underforecasted'
WHEN HistoryQuantity IS NULL AND TOTFCST > 0 THEN 'Overforecasted'
WHEN TOTFCST IS NULL AND HistoryQuantity > 0 THEN 'Underforecasted'
WHEN TOTFCST = 0.000 AND HistoryQuantity IS NULL THEN 'No issue'
END AS Issue
FROM Forecast f FULL OUTER JOIN Actuals a
ON f.LOC = a.LOC AND f.DMDUNIT = a.DMDUNIT AND f.STARTDATE = a.DMDPostDate
GROUP BY
COALESCE(f.LOC, a.LOC),
COALESCE(f.DMDUNIT, a.DMDUNIT),
COALESCE(f.STARTDATE, a.DMDPostDate),
a.HistoryQuantity, F.TOTFCST),
Transitions as (
select *,
case when indicator <> lag(indicator)
over (partition by DMDUNIT order by "Start Date")
then 1 end as tripped
from Merged cross apply (
select case when Issue in ('Overforecasted', 'Underforecasted')
then 1 else 0 end as indicator) v
), Bundles as (
select *, count(tripped) over (partition by DMDUNIT order by "Start Date") as grp
from Transitions
), Streaks as (
select *, count(*) over (partition by DMDUNIT, grp) as cnt
from Bundles
)
select *, case when indicator = 1 and cnt >= 3 then 'Yes' else 'No' end as InIssueStreak, cnt as StreakLength
from Streaks;

Answer 1

WITH Forecast AS (
SELECT LOC, DMDUNIT, STARTDATE, TOTFCST
FROM SCPOMGR.FCSTPERFSTATIC
WHERE STARTDATE >= '2021-11-24'
), Actuals AS (
SELECT LOC, DMDUNIT, DMDPostDate, HistoryQuantity
FROM SCPOMGR.HISTWIDE_CHAIN
WHERE DMDPostDate >= '2021-11-24'
), Merged AS (
SELECT
COALESCE(f.LOC, a.LOC) AS LOC,
COALESCE(f.DMDUNIT, a.DMDUNIT) AS DMDUNIT,
COALESCE(f.STARTDATE, a.DMDPostDate) AS "Start Date",
SUM(F.TOTFCST) AS "Forecast",
SUM(a.HistoryQuantity) AS "Actuals",
SUM(ABS(a.HistoryQuantity) - f.TOTFCST) AS "Abs Error"
(1 - SUM(a.HistoryQuantity - SUM(f.TOTFCST)) / SUM(a.HistoryQuantity) as "FA%",
SUM(a.HistoryQuantity) / SUM(f.TOTFCST) AS "Bias",
CASE
WHEN SUM(f.TOTFCST) > SUM(a.HistoryQuantity) THEN 'Overforecasted'
WHEN SUM(f.TOTFCST) < SUM(a.HistoryQuantity) THEN 'Underforecasted'
WHEN SUM(a.HistoryQuantity) IS NULL AND SUM(f.TOTFCST) > 0 THEN 'Overforecasted'
WHEN SUM(f.TOTFCST) IS NULL AND SUM(a.HistoryQuantity) > 0 THEN 'Underforecasted'
WHEN SUM(f.TOTFCST) = 0.000 AND SUM(a.HistoryQuantity) IS NULL THEN 'No issue'
END AS Issue
FROM Forecast f FULL OUTER JOIN Actuals a
ON f.LOC = a.LOC AND f.DMDUNIT = a.DMDUNIT AND f.STARTDATE = a.DMDPostDate
GROUP BY
COALESCE(f.LOC, a.LOC),
COALESCE(f.DMDUNIT, a.DMDUNIT),
COALESCE(f.STARTDATE, a.DMDPostDate)
ORDER BY
COALESCE(f.LOC, a.LOC),
COALESCE(f.DMDUNIT, a.DMDUNIT),
COALESCE(f.STARTDATE, a.DMDPostDate)
)
select *,
case when
min(Issue) over (
partition by DMDUNIT order by "Start Date"
rows between 2 preceding and current row) =
max(Issue) over (
partition by DMDUNIT order by "Start Date"
rows between 2 preceding and current row) and
count(Issue) over (
partition by DMDUNIT order by "Start Date"
rows between 2 preceding and current row) = 3
then 'Yes' else 'No' end as "3InARow"
from Merged;
If that doesn't work then try gaps and islands:
with (<copied from above...>), Transitions as (
select *,
case when indicator <> lag(indicator)
over (partition by DMDINIT order by "Start Date")
then 1 end as tripped
from Merged cross apply (
select case when Issue in ('Overforecasted', 'Underforecasted')
then 1 else 0 end as indicator) v
), Bundles as (
select *, sum(tripped) over (partition by DMDUNIT order by "Start Date") as grp
from Transitions
), Streaks as (
select *, count(*) over (partition by DMDUNIT, grp) as cnt
from Bundles
)
select *, case when cnt >= 3 then 'Yes' else 'No' end as InStreak, cnt as StreakLength
from Streaks;
https://dbfiddle.uk/?rdbms=sqlserver_2017&fiddle=9fcbab1d93b7297aebc340111aa3a448

Answer 2

The fiddle
I've created a working test case which demonstrates how to identify consecutive weeks of over or under forecast items.
All the detail about how a weekly value is calculated or how you identify items is immaterial to the basic question. I've ignored issues related to missing data for a week. That can be easily included and just obscures the fundamental (and only) question asked.
For this example, just assume we have only the weeks of interest and all weeks for all items exist.
The SQL, which also contains the data:
WITH forecast (item, prediction) AS (
SELECT 1, 1000 UNION
SELECT 2, 500
)
, actuals (weekno, item, actual) AS (
SELECT 1, 1, 500 UNION
SELECT 2, 1, 600 UNION
SELECT 3, 1, 1600 UNION
SELECT 4, 1, 1600 UNION
SELECT 5, 1, 600 UNION
SELECT 6, 1, 1100 UNION
SELECT 7, 1, 1200 UNION
SELECT 8, 1, 1150 UNION
SELECT 9, 1, 601 UNION
SELECT 10, 1, 602 UNION
SELECT 11, 1, 603 UNION
SELECT 1, 2, 1500 UNION
SELECT 2, 2, 600 UNION
SELECT 3, 2, 550 UNION
SELECT 4, 2, 500 UNION
SELECT 5, 2, 600 UNION
SELECT 6, 2, 491 UNION
SELECT 7, 2, 492 UNION
SELECT 8, 2, 493 UNION
SELECT 9, 2, 494 UNION
SELECT 10, 2, 620
)
, step1 AS (
SELECT a.*
, f.prediction
, CASE WHEN actual < prediction THEN -1
WHEN actual > prediction THEN +1
ELSE 0
END AS side
FROM forecast AS f
JOIN actuals AS a
ON f.item = a.item
)
, step2 AS (
SELECT *
, CASE WHEN side = LAG(side) OVER (PARTITION BY item ORDER BY weekno) THEN 0 ELSE 1 END AS edge
FROM step1
)
, step3 AS (
SELECT *
, SUM(edge) OVER (PARTITION BY item ORDER BY weekno) AS xgroup
FROM step2
)
, step4 AS (
SELECT *
, COUNT(*) OVER (PARTITION BY item, xgroup) AS xcount
FROM step3
)
SELECT *
FROM step4
WHERE xcount >= 3
ORDER BY item, weekno
;
The result:
weekno
item
actual
prediction
side
edge
xgroup
xcount
6
1
1100
1000
1
1
4
3
7
1
1200
1000
1
0
4
3
8
1
1150
1000
1
0
4
3
9
1
601
1000
-1
1
5
3
10
1
602
1000
-1
0
5
3
11
1
603
1000
-1
0
5
3
1
2
1500
500
1
1
1
3
2
2
600
500
1
0
1
3
3
2
550
500
1
0
1
3
6
2
491
500
-1
1
4
4
7
2
492
500
-1
0
4
4
8
2
493
500
-1
0
4
4
9
2
494
500
-1
0
4
4