SQL Server Ranking issue - sql-server

Question

I am trying to apply ranking to my data set the logic is as follows:
For each ID , Order by ID2 ASC and Order by IsMaster Desc rank the row 1 and only change it when the ID4 value changes
My dataset and desired output looks like:
Test data
CREATE TABLE Test_Table
(ID INT ,ID2 INT, IsMaster INT, ID4 VARCHAR(10))
GO
INSERT INTO Test_Table (ID ,ID2 , IsMaster , ID4 )
VALUES
(1, 101, 1 ,'AAA') -- 1 <-- Desired output for rank
,(1, 102, 0 ,'AAA') -- 1
,(1, 103, 0 ,'AAB') -- 2
,(1, 104, 0 ,'AAB') -- 2
,(1, 105, 0 ,'CCC') -- 3
,(2, 101, 1 ,'AAA') -- 1
,(2, 102, 0 ,'AAA') -- 1
,(2, 103, 0 ,'AAA') -- 1
,(2, 104, 0 ,'AAB') -- 2
,(2, 105, 0 ,'CCC') -- 3
this is what I have tried so far:
SELECT *
,DENSE_RANK() OVER (PARTITION BY ID ORDER BY ID2 ASC, IsMaster DESC ) rn
FROM Test_Table
please please please help me thank you.

Answer 1

This is a island/gap problem.
First you use LAG() to see if you have a different ID4 on the same partition.
Is important you also need partition by IsMaster
Then you create the islands when ID4 changes.
Finally use comulative SUM() to get the proper rank.
Sql Demo
WITH id4_change as (
SELECT *,
LAG(ID4) OVER (PARTITION BY ID, IsMaster ORDER BY ID2) as prev
FROM Test_Table
), islands as (
SELECT *,
CASE WHEN ID4 = PREV
THEN 0
ELSE 1
END as island
FROM id4_change
)
SELECT *,
SUM(island) OVER (PARTITION BY ID, IsMaster ORDER BY ID2) rank
FROM islands
ORDER BY ID, ID2, IsMaster DESC
;
OUTPUT: You can see when ID4 = PREV doesnt create a new "Island" so have same rank.
EDIT: You can simplify first two querys
WITH id4_change as (
SELECT *,
CASE WHEN ID4 = LAG(ID4) OVER (PARTITION BY ID, IsMaster ORDER BY ID2)
THEN 0
ELSE 1
END as island
FROM Test_Table
)
SELECT *,
SUM(island) OVER (PARTITION BY ID, IsMaster ORDER BY ID2) rank
FROM id4_change
ORDER BY ID, ID2, IsMaster DESC
;

Answer 2

Another way probably less efficient but it will work.
WITH X AS
(
SELECT *
,ROW_NUMBER() OVER (PARTITION BY ID ORDER BY ID2) RowNum
FROM dbo.Test_Table
)
, CTE_VehicleNumber
as
(
SELECT T.ID , T.ID2, t.IsMaster ,T.ID4 , t.RowNum , 1 as [Rank]
FROM X as T
WHERE T.IsMaster = 1
UNION ALL
SELECT T.ID, T.ID2, t.IsMaster ,T.ID4 , t.RowNum , CASE WHEN t.ID4 <> c.ID4 THEN 1+ C.[Rank]
ELSE 0+ C.[Rank]
END as [Rank]
FROM CTE_VehicleNumber as C
inner join X as T ON T.RowNum = C.RowNum + 1
AND t.ID = c.ID
)
SELECT ID , ID2, IsMaster ,ID4 , [Rank]
FROM CTE_VehicleNumber
ORDER BY ID , ID2, IsMaster ,ID4 , [Rank]
OPTION (MAXRECURSION 0);

Answer 3

Are you sure that your orders of ID2 and IsMaster affect the desired result, considering the rest of the data in ID and ID4?
I just tried to use the following code:
; WITH CTE AS (
SELECT DISTINCT ID, ID4, DENSE_RANK() OVER (ORDER BY ID4) Rnk
FROM #Test_Table
)
SELECT t.*, c.Rnk
FROM #Test_Table t
INNER JOIN CTE c ON t.ID = c.ID AND t.ID4 = c.ID4;
... and even with changing the order of ID2 and IsMaster I can't get it to "misbehave" - IF there's only one IsMaster = 1 per a group of ID4's and no duplicates in ID2.