grouping, splitting, and counting rows - sql-server

Question

given this table and data:
DECLARE #Table table (RowID int, RowCode char(1), RowValue int);set nocount on
INSERT #Table VALUES ( 6,'A',3757 )
INSERT #Table VALUES ( 5,'A',37827)
INSERT #Table VALUES (14,'A',48411)
INSERT #Table VALUES ( 1,'A',48386)
INSERT #Table VALUES (20,'A',48450)
INSERT #Table VALUES ( 7,'A',46155)
INSERT #Table VALUES (13,'A',721 )
INSERT #Table VALUES ( 2,'A',49335)
INSERT #Table VALUES (15,'A',4700 )
INSERT #Table VALUES (19,'A',64416)
INSERT #Table VALUES ( 8,'A',27246)
INSERT #Table VALUES (12,'B',54929)
INSERT #Table VALUES (16,'B',3872 )
INSERT #Table VALUES ( 3,'C',728 )
INSERT #Table VALUES (11,'C',1050 )
INSERT #Table VALUES ( 9,'C',3191 )
INSERT #Table VALUES (17,'C',866 )
INSERT #Table VALUES ( 4,'C',838 )
INSERT #Table VALUES (10,'D',550 )
INSERT #Table VALUES (18,'D',1434 );set nocount off
I need this:
VVVVVVVV
RowID RowCode RowValue RowChunk
----- ------- -------- --------
1 A 48386 1
2 A 49335 1
5 A 37827 1
6 A 3757 1
7 A 46155 1
8 A 27246 2
13 A 721 2
14 A 48411 2
15 A 4700 2
19 A 64416 2
20 A 48450 3
12 B 54929 4
16 B 3872 4
3 C 728 5
4 C 838 5
9 C 3191 5
11 C 1050 5
17 C 866 5
10 D 550 6
18 D 1434 6
RowChunk starts at 1 and is incremented by 1 for each RowCode change and/or when there have been 5 of the same RowCode values.

Answer 1

Basically my solution uses the same approach as yours, only with slightly different devices employed.
WITH NumberedRows AS (
SELECT
RowID,
RowCode,
RowValue,
CodeChunk = (ROW_NUMBER() OVER (PARTITION BY RowCode ORDER BY RowID) - 1) / 5
FROM #Table
)
SELECT
RowID,
RowCode,
RowValue,
RowChunk = DENSE_RANK() OVER (ORDER BY RowCode, CodeChunk)
FROM NumberedRows

Answer 2

I don't think there's an analysis function, or any reasonable combination of such, which will address this. You'll have to do it RBAR with a cursor or, slightly faster in my experience, a loop.
This example of looping assumes that RowID is unique. If RowID is not the clustered PK, this will be very slow, so if that's the case you'll want to create a temp table.
DECLARE #RowID INT = (SELECT MIN(RowID) FROM #Table)
DECLARE #MaxRowID INT = (SELECT MAX(RowID) FROM #Table)
DECLARE #RowCode CHAR(1)
DECLARE #LastRowCode CHAR(1)
DECLARE #RowValue INT
DECLARE #Chunk INT = 0
DECLARE #RecsThisChunk INT
DECLARE #Results TABLE (RowID INT NOT NULL PRIMARY KEY, RowCode CHAR(1) NOT NULL, RowValue INT NOT NULL, Chunk INT NOT NULL)
WHILE #RowID <= #MaxRowID
BEGIN
-- Handle gaps in RowID
IF NOT EXISTS (SELECT * FROM #Table WHERE RowID = #RowID) GOTO EndOfLoop
-- Load values for this record
SELECT #RowCode = RowCode, #RowValue = RowValue FROM #Table WHERE RowID = #RowID
IF #LastRowCode IS NULL OR #RowCode <> #LastRowCode OR #RecsThisChunk = 5
BEGIN
-- Start a new chunk
SET #Chunk = #Chunk + 1
SET #RecsThisChunk = 1
END
ELSE
BEGIN
-- Same chunk
SET #RecsThisChunk = #RecsThisChunk + 1
END
SET #LastRowCode = #RowCode
INSERT INTO #Results (RowID, RowCode, RowValue, Chunk) VALUES (#RowID, #RowCode, #RowValue, #Chunk)
EndOfLoop:
SET #RowID = #RowID + 1
END
SELECT * FROM #Results
You may have tweak this a bit for 2005, I use 2008 routinely and don't recall all the little differences.
FYI, the results you show don't quite match the sample data.
Hope this helps! The only alternative I see is a cursor, or handling this in the application layer.

Answer 3

this does the trick without a loop:
;WITH NumberedRows AS (
SELECT
r.RowID, r.RowCode, r.RowValue, CEILING(ROW_NUMBER() OVER(PARTITION BY r.RowCode ORDER BY r.RowCode,r.RowID)/5.0) AS CodeRowChunk
FROM #Table r
)
, AllChunks AS (
SELECT r.*,ROW_NUMBER() OVER(ORDER BY RowCode,CodeRowChunk) AS ChunkRowNumber
FROM (SELECT DISTINCT
RowCode, CodeRowChunk
FROM NumberedRows) r
)
SELECT
a.RowID, RowCode, a.RowValue,
(SELECT ChunkRowNumber FROM AllChunks c WHERE c.RowCode=a.RowCode and c.CodeRowChunk=a.CodeRowChunk) AS RowChunk
FROM NumberedRows a

Answer 4

This is the answer you are looking for :
create Table [table] (RowID int, RowCode char(1), RowValue int)
INSERT [Table] VALUES ( 6,'A',3757 )
INSERT [Table] VALUES ( 5,'A',37827)
INSERT [Table] VALUES (14,'A',48411)
INSERT [Table] VALUES ( 1,'A',48386)
INSERT [Table] VALUES (20,'A',48450)
INSERT [Table] VALUES ( 7,'A',46155)
INSERT [Table] VALUES (13,'A',721 )
INSERT [Table] VALUES ( 2,'A',49335)
INSERT [Table] VALUES (15,'A',4700 )
INSERT [Table] VALUES (19,'A',64416)
INSERT [Table] VALUES ( 8,'A',27246)
INSERT [Table] VALUES (12,'B',54929)
INSERT [Table] VALUES (16,'B',3872 )
INSERT [Table] VALUES ( 3,'C',728 )
INSERT [Table] VALUES (11,'C',1050 )
INSERT [Table] VALUES ( 9,'C',3191 )
INSERT [Table] VALUES (17,'C',866 )
INSERT [Table] VALUES ( 4,'C',838 )
INSERT [Table] VALUES (10,'D',550 )
INSERT [Table] VALUES (18,'D',1434 )
IF object_id('tempdb..#tempTable') IS NOT NULL
BEGIN
DROP TABLE #tempTable
END
CREATE TABLE #tempTable
(RowID int, RowCode char(1), RowValue int,RowChunk int)
INSERT INTO #tempTable
select RowID,RowCode,RowValue,null from [table]
declare #RowId int
declare #RowCode char(1)
declare #Count int
declare #CurrentCode char(1)
declare #CountCurrent int
set #Count=1
set #CurrentCode=1
set #CountCurrent=0
DECLARE contact_cursor CURSOR FOR
SELECT RowID,RowCode FROM [table]
OPEN contact_cursor
FETCH NEXT FROM contact_cursor into #RowId,#RowCode
set #CurrentCode=#RowCode
WHILE ##FETCH_STATUS = 0
BEGIN
if(#CurrentCode=#RowCode)
begin
if(#CountCurrent=5)
begin
set #CountCurrent=1
set #Count=#Count+1
update #tempTable set RowChunk=#Count where RowID=#RowID
end
else
begin
set #CountCurrent=#CountCurrent+1
update #tempTable set RowChunk=#Count where RowID=#RowID
end
end
else
begin
set #CurrentCode=#RowCode
set #CountCurrent=1
set #Count=#Count+1
update #tempTable set RowChunk=#Count where RowID=#RowID
end
FETCH NEXT FROM contact_cursor into #RowId,#RowCode
END
CLOSE contact_cursor
DEALLOCATE contact_cursor
select * from #tempTable
GO