Conditional Aggregation

Question

Post reply

Conditional Aggregation

Chrissy321

SSCoach

Points: 15789

March 12, 2024 at 10:34 pm

#4394228

Thanks if you can help.

I would like to SUM by Color but only if SampleType = 1.

I could do this with UNION ALL but I am wondering if I can eliminate the UNION ALL. If I can eliminate the UNION ALL I would then do some performance testing between the two methods.

DROP TABLE IF EXISTS #SampleData
CREATE TABLE #SampleData (Color VARCHAR(10), SampleType INT, SampleValue INT)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Blue',1,100)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Blue',1,101)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Blue',1,102)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Green',1,200)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Green',1,201)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Red',2,300)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Red',2,301)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Yellow',3,400)
INSERT INTO #SampleData (Color,SampleType,SampleValue)VALUES ('Yellow',3,401)

SELECT * FROM #SampleData

SELECT * FROM
(
SELECT DISTINCT Color,SampleType,
SUM(SampleValue)OVER(PARTITION BY Color) AS SampleValue
FROM #SampleData WHERE SampleType = 1

UNION ALL 

SELECT * FROM #SampleData WHERE SampleType <> 1
)SD

Viewing 11 posts - 1 through 11 (of 11 total)

You must be logged in to reply to this topic. Login to reply

Phil Parkin SSC Guru Points: 247159 More actions · Answer 1

Phil Parkin

SSC Guru

Points: 247159

March 12, 2024 at 10:50 pm

#4394229

Try

SUM(IIF(SampleType = 1, SampleValue,0))

Chrissy321 SSCoach Points: 15789 More actions · Answer 2

Phil Parkin wrote:

Try
SUM(IIF(SampleType = 1, SampleValue,0))

Like this?

SELECT Color,SampleType,

SUM(IIF(SampleType = 1, SampleValue,0))

FROM #SampleData

GROUP BY

Color,SampleType

Chrissy321 SSCoach Points: 15789 More actions · Answer 3

This seems to work but I somehow doubt it will perform any better.

SELECT
DISTINCT
Color,SampleType,
SUM(SampleValue)OVER(PARTITION BY CASE WHEN SampleType = 1 THEN Color ELSE CONVERT(VARCHAR(MAX),RowNum) END) AS SampleValue
FROM
(
SELECT  
ROW_NUMBER()OVER (ORDER BY Color,SampleType,SampleValue ) AS RowNum,
Color,SampleType,SampleValue
FROM #SampleData 
)SD

Phil Parkin SSC Guru Points: 247159 More actions · Answer 4

It's a bit hacky, but this avoids the UNION ALL:

WITH Uniqued
AS (SELECT Grouper = IIF(sd.SampleType = 1, 1, CHECKSUM (*))
          ,sd.Color
          ,sd.SampleType
          ,sd.SampleValue
    FROM #SampleData sd)
SELECT Uniqued.Color
      ,Uniqued.SampleType
      ,SUM (Uniqued.SampleValue)
FROM Uniqued
GROUP BY Uniqued.Grouper
        ,Uniqued.Color
        ,Uniqued.SampleType
ORDER BY Uniqued.SampleType
        ,Uniqued.Color;

Jonathan AC Roberts SSCoach Points: 18407 More actions · Answer 5

SELECT Color,
       SampleType,
       SUM(SampleValue) SampleValue
  FROM #SampleData
 GROUP BY Color, 
          SampleType, 
          CASE WHEN SampleType = 1 THEN '1' ELSE CONVERT(char(36), NewId()) END 
;

Jonathan AC Roberts SSCoach Points: 18407 More actions · Answer 6

This might be more efficient:

;WITH CTE AS
(
    SELECT *,
           ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) rn
      FROM #SampleData
)
SELECT Color,SampleType,SUM(SampleValue) SampleValue
  FROM CTE
 GROUP BY Color,
          SampleType,
          CASE WHEN SampleType = 1 THEN 0 ELSE rn END 
;

Chrissy321 SSCoach Points: 15789 More actions · Answer 7

Thanks for all the solutions. I'll adapt to my actual DDL and report back If I uncover anything notable.

Phil Parkin SSC Guru Points: 247159 More actions · Answer 8

If your actual tables include a PK not mentioned here, you may be able to use that to make your row unique, which would obviate the need for a derived 'row uniquer'.

CASE WHEN SampleType = 1 THEN 0 ELSE PK END

(From Jonathan's latest code)

This would remove the need for a CTE.

Chrissy321 SSCoach Points: 15789 More actions · Answer 9

Good point. I have a composite primary key comprised of 3 INT columns.

Maybe.

CASE WHEN SampleType = 1 THEN 0 ELSE PK1 + PK2 + PK3 END

Although I guess the 3 numbers summed could produce duplicates. Casting to CHAR and concatenating would create a unique value but the cast could be a performance killer.

Phil Parkin SSC Guru Points: 247159 More actions · Answer 10

You are right to consider the potential for duplicates. I can't think of an elegant (and fool-proof) way of handling the composite key – so I'd stick with Jonathan's suggestion.

Conditional Aggregation

Cookies on SQLServerCentral