Finding the MIN or a group of values not used in a previous group

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 1, 2018 at 12:13 pm

  #409244

  I have been having trouble finding anything on google because I'm not sure how to phrase the search. What I need to do is to select the MIN R from each Id group that hasn't already been selected by a previous group. This is the trimmed down version with an example of a row that has no R value.

  I have a table simplified

  CREATE TABLE Data(Id int, R int)
  INSERT INTO Data VALUES(1,1),(2,1),(3,2),(3,3),(4,2),(4,3),(4,4),(5,4),(5,5),(6,5),(6,6),(7,6)

  
  I think I should be able to do it using a tally table and window functions, but the grouping is giving me a headache. I want the min in the group that has not been previously used. I know I can create a cursor to do what I want, but I think there should be a way to do this set based. Any help is appreciated.

• Lynn Pettis

  SSC Guru

  Points: 442467

  More actions

  November 1, 2018 at 12:15 pm

  #2012092

  Chris Souchik - Thursday, November 1, 2018 12:13 PM

  I have been having trouble finding anything on google because I'm not sure how to phrase the search. What I need to do is to select the MIN R from each Id group that hasn't already been selected by a previous group. This is the trimmed down version with an example of a row that has no R value.

  I have a table simplified

  CREATE TABLE Data(Id int, R int)
  INSERT INTO Data VALUES(1,1),(2,1),(3,2),(3,3),(4,2),(4,3),(4,4),(5,4),(5,5),(6,5),(6,6),(7,6)

  
  I think I should be able to do it using a tally table and window functions, but the grouping is giving me a headache. I want the min in the group that has not been previously used. I know I can create a cursor to do what I want, but I think there should be a way to do this set based. Any help is appreciated.

  So expected results based on the above data.

• ScottPletcher

  SSC Guru

  Points: 100948

  More actions

  November 1, 2018 at 12:24 pm

  #2012095

  This may not be the most efficient, but I believe it works, and shouldn't be too much overhead anyway unless you have lots of unqiue Ids:

  
;WITH cte_group AS (
    SELECT Id, MIN(R) AS R
    FROM Data
    GROUP BY id
)
SELECT cg.Id, cg_unique.R
FROM cte_group cg
LEFT OUTER JOIN (
    SELECT *, ROW_NUMBER() OVER(PARTITION BY R ORDER BY Id) AS row_num
    FROM cte_group
) AS cg_unique ON cg_unique.row_num = 1 AND cg_unique.Id = cg.Id

  SQL DBA,SQL Server MVP(07, 08, 09) A socialist is someone who will give you the shirt off *someone else's* back.

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 1, 2018 at 12:24 pm

  #2012096

  Correct

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 1, 2018 at 12:40 pm

  #2012098

  Scott,

  Very close, but I am seeing this:

  
  Id 4 should not be null it should be 3. Thanks

• andycadley

  SSCertifiable

  Points: 5296

  More actions

  November 1, 2018 at 1:14 pm

  #2012099

  I think this will do it:

  with X as
(
select id, max(R) over (order by id, R) as maxR
from #Data
),
Y as
(
select id, min(maxR) as R
from X
group by id
)
select
 id, nullif(R, lag(R,1) over (order by id))
from Y

• Jason A. Long

  SSC-Insane

  Points: 23780

  More actions

  November 1, 2018 at 2:32 pm

  #2012114

  My entry...

  IF OBJECT_ID('tempdb..#TestData', 'U') IS NOT NULL
BEGIN DROP TABLE #TestData; END;

  CREATE TABLE #TestData (
    Id INT NOT NULL,
    R INT NOT NULL
    );
INSERT #TestData(Id, R) VALUES
    (1,1),(2,1),(3,2),(3,3),(4,2),(4,3),(4,4),(5,4),(5,5),(6,5),(6,6),(7,6);

  --================================================================================

  SELECT
    x1    .Id,
    R = CASE WHEN x1.R > ISNULL(MAX(x1.R) OVER (ORDER BY x1.R, x1.Id ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING), 0) THEN x1.R END
FROM (
        SELECT
            td    .Id,
            td.R,
            r1 = ISNULL(MIN(td.R) OVER (PARTITION BY td.Id ORDER BY td.Id, td.R ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING), td.R),
            r2 = ISNULL(MAX(td.R) OVER (ORDER BY td.Id, td.R ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING), 0)
        FROM
            #TestData td
        ) x1
WHERE    
    x1.R = CASE WHEN x1.r1 >= x1.r2 THEN x1.r1 ELSE x1.r2 END;

  Results...
  Id    R
----------- -----------
1    1
2    NULL
3    2
4    3
5    4
6    5
  7    6
  

• Jason A. Long

  SSC-Insane

  Points: 23780

  More actions

  November 1, 2018 at 2:35 pm

  #2012115

  andycadley - Thursday, November 1, 2018 1:14 PM

  I think this will do it:

  with X as
(
select id, max(R) over (order by id, R) as maxR
from #Data
),
Y as
(
select id, min(maxR) as R
from X
group by id
)
select
 id, nullif(R, lag(R,1) over (order by id))
from Y

  Very nice!

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 2, 2018 at 5:24 am

  #2012166

  Andy and Jason,

  Thanks for your solution. It works great on this data set, but unfortunately it doesn't work for all data sets. For example,
  

  I'm starting to think it may be easier to use a cursor, even if it will be slower. The recursive nature of the query may not allow a set based solution, at least it has me stumped.

• george_at_sql

  SSCommitted

  Points: 1967

  More actions

  November 2, 2018 at 11:15 am

  #2012239

  I also think, it would need a recursive solution to since lead / lag values based solution looks at the values when the query was kicked off and doesnt get influenced by the values that changes when the query progresses.

  I could do this using recursive cte as follows

  
CREATE TABLE Data(Id int, R int)
INSERT INTO Data VALUES (1,1);
INSERT INTO Data VALUES (2,1);
INSERT INTO Data VALUES (3,2);
INSERT INTO Data VALUES (3,3);
INSERT INTO Data VALUES (4,2);
INSERT INTO Data VALUES (4,3);
INSERT INTO Data VALUES (4,4);
INSERT INTO Data VALUES (5,4);
INSERT INTO Data VALUES (5,5);
INSERT INTO Data VALUES (6,5);
INSERT INTO Data VALUES (6,6);
INSERT INTO Data VALUES (7,6);

  /*
truncate table data

  INSERT INTO Data VALUES (1,1);
INSERT INTO Data VALUES (2,1);
INSERT INTO Data VALUES (2,2);
INSERT INTO Data VALUES (2,3);
INSERT INTO Data VALUES (3,3);*/

  with data1
as (select id /*This block only to generate row_numbers based on the records in the table..*/
     ,r
     ,row_number() over(order by id,r) as rnk
   from data
   )
,cte(id,r,rnk,grp_flag,concat_val)
  as (select a.id,a.r,a.rnk,1 as grp_flag,cast(a.r as varchar2(100)) as concat_val
    from data1 a
   where rnk=1 /*Start with first record and construct the concat_val only if the "r" is not previously used and "r" value is not already populated for the id*/
   union all
   select b.id,b.r,b.rnk,case when a.id=b.id and a.grp_flag=1 then 0
              else case when a.concat_val like '%'||b.r||'%' then 0
                 else 1
                end
             end as grp_flag
            ,case when a.id=b.id and a.grp_flag=1 then a.concat_val
              else case when a.concat_val like '%'||b.r||'%' then a.concat_val
                 else b.r||'/'||a.concat_val
                end
            end as concat_val
    from cte a
    join data1 b
     on a.rnk+1=b.rnk
   )
select x.id,y.r
  from ( select distinct id
    from data1
    ) x
left join cte y
   on x.id=y.id
  and y.grp_flag=1
order by x.id  

• george_at_sql

  SSCommitted

  Points: 1967

  More actions

  November 4, 2018 at 10:10 pm

  #2012364

  Did the above query work out?

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 5, 2018 at 7:46 am

  #2012394

  George,

  I will try and test this today. At a quick glance, it looks promising.

  Thanks,
  Chris

• Chris Souchik

  SSC-Addicted

  Points: 419

  More actions

  November 9, 2018 at 12:47 pm

  #2013020

  Unfortunately really busy week. I haven't had a chance to review yet. Will update when I get a chance. Thanks

• sgmunson

  SSC Guru

  Points: 110639

  More actions

  November 15, 2018 at 12:08 pm

  #2013730

  Here's an actual cursor.   I thought about this one for a long time, and realized I could not use a recursive CTE and still use any form of aggregate, or a 2nd reference to the CTE itself.   That was really crimping my style, so I gave up and went this way instead.   Run times were about 0.01563 seconds, +/-, before caching took over and both queries would run instantaneously.   Fortunately, it works for both sets of input data:
  SET NOCOUNT ON;
IF OBJECT_ID(N'tempdb..#Data', N'U') IS NOT NULL
    BEGIN
    DROP TABLE #Data;
    END;

  IF OBJECT_ID(N'tempdb..#Results', N'U') IS NOT NULL
    BEGIN
    DROP TABLE #Results;
    END;

  CREATE TABLE #Data(
    Id int,
    R int
);
INSERT INTO #Data (Id, R)
    VALUES    (1,1),
            (2,1),
            (3,2),
            (3,3),
            (4,2),
            (4,3),
            (4,4),
            (5,4),
            (5,5),
            (6,5),
            (6,6),
            (7,6);

  CREATE TABLE #Results (
    Id int NOT NULL PRIMARY KEY CLUSTERED,
    MIN_R int NULL
);

  DECLARE @Id AS int, @r AS int, @First AS int = 1, @BegDateTime AS datetime2 = sysdatetime(),
        @EndDateTime AS datetime2;

  DECLARE MYCURSOR CURSOR FAST_FORWARD READ_ONLY FOR
    SELECT
        D.Id,
        MIN(D.R) AS R
    FROM #Data AS D
    GROUP BY D.Id;

  OPEN MYCURSOR;
FETCH NEXT FROM MYCURSOR INTO @Id, @r;

  WHILE @@FETCH_STATUS = 0
    BEGIN;
    IF @First = 1
        BEGIN;
        BEGIN TRAN;
        INSERT INTO #Results (Id, MIN_R)
        VALUES (@Id, @r);
        COMMIT TRAN;
        END;
     ELSE
        BEGIN;
        BEGIN TRAN;
        INSERT INTO #Results (Id, MIN_R)
        VALUES (@Id,
                    CASE
                        WHEN (
                                SELECT MIN(D.R)
                                FROM #Data AS D
                                WHERE    D.R NOT IN (SELECT R.MIN_R FROM #Results AS R WHERE R.MIN_R IS NOT NULL AND R.Id < @Id)
                                    AND D.Id = @Id
                            ) IS NULL
                            AND (SELECT COUNT(1) FROM #Data AS D WHERE D.Id = @Id) = 1
                            THEN NULL
                        ELSE (
                                SELECT MIN(D.R)
                                FROM #Data AS D
                                WHERE    D.R NOT IN (SELECT R.MIN_R FROM #Results AS R WHERE R.MIN_R IS NOT NULL AND R.Id < @Id)
                                    AND D.Id = @Id
                            )
                    END)
        COMMIT TRAN;
        END;

      SET @First = @First + 1;
    FETCH NEXT FROM MYCURSOR INTO @Id, @r;
    END;

  CLOSE MYCURSOR;
DEALLOCATE MYCURSOR;

  SET @EndDateTime = sysdatetime();
PRINT 'Total Run Time, Query Only: ' + CONVERT(varchar(11), CONVERT(decimal(10,9), DATEDIFF(microsecond, @BegDateTime, @EndDateTime)/1E6)) + ' Seconds';

  SELECT *
FROM #Data
ORDER BY Id;

  SELECT *
FROM #Results AS R
ORDER BY R.Id;
GO

  SET NOCOUNT ON;
IF OBJECT_ID(N'tempdb..#Data', N'U') IS NOT NULL
    BEGIN
    DROP TABLE #Data;
    END;

  IF OBJECT_ID(N'tempdb..#Results', N'U') IS NOT NULL
    BEGIN
    DROP TABLE #Results;
    END;

  CREATE TABLE #Data(
    Id int,
    R int
);
INSERT INTO #Data (Id, R)
    VALUES    (1,1),
            (2,1),
            (2,2),
            (2,3),
            (3,3);

  CREATE TABLE #Results (
    Id int NOT NULL PRIMARY KEY CLUSTERED,
    MIN_R int NULL
);

  DECLARE @Id AS int, @r AS int, @First AS int = 1, @BegDateTime AS datetime2 = sysdatetime(),
        @EndDateTime AS datetime2;

  DECLARE MYCURSOR CURSOR FAST_FORWARD READ_ONLY FOR
    SELECT
        D.Id,
        MIN(D.R) AS R
    FROM #Data AS D
    GROUP BY D.Id;

  OPEN MYCURSOR;
FETCH NEXT FROM MYCURSOR INTO @Id, @r;

  WHILE @@FETCH_STATUS = 0
    BEGIN;
    IF @First = 1
        BEGIN;
        BEGIN TRAN;
        INSERT INTO #Results (Id, MIN_R)
        VALUES (@Id, @r);
        COMMIT TRAN;
        END;
     ELSE
        BEGIN;
        BEGIN TRAN;
        INSERT INTO #Results (Id, MIN_R)
        VALUES (@Id,
                    CASE
                        WHEN (
                                SELECT MIN(D.R)
                                FROM #Data AS D
                                WHERE    D.R NOT IN (SELECT R.MIN_R FROM #Results AS R WHERE R.MIN_R IS NOT NULL AND R.Id < @Id)
                                    AND D.Id = @Id
                            ) IS NULL
                            --AND (SELECT COUNT(1) FROM #Data AS D WHERE D.Id = @Id) = 1
                            THEN NULL
                        ELSE (
                                SELECT MIN(D.R)
                                FROM #Data AS D
                                WHERE    D.R NOT IN (SELECT R.MIN_R FROM #Results AS R WHERE R.MIN_R IS NOT NULL AND R.Id < @Id)
                                    AND D.Id = @Id
                            )
                    END)
        COMMIT TRAN;
        END;

      SET @First = @First + 1;
    FETCH NEXT FROM MYCURSOR INTO @Id, @r;
    END;

  CLOSE MYCURSOR;
DEALLOCATE MYCURSOR;

  SET @EndDateTime = sysdatetime();
PRINT 'Total Run Time, Query Only: ' + CONVERT(varchar(11), CONVERT(decimal(10,9), DATEDIFF(microsecond, @BegDateTime, @EndDateTime)/1E6)) + ' Seconds';

  SELECT *
FROM #Data
ORDER BY Id;

  SELECT *
FROM #Results AS R
ORDER BY R.Id;

  Steve (aka sgmunson) 🙂 🙂 🙂
  Rent Servers for Income (picks and shovels strategy)