Thanks.

We rented data from a data broker. In this data, you have a few hundred thousand names, and these names are generally grouped together at buildings (a B-to-B type environment) and have job titles. There are thousands of buildings, meaning that it’s not quite as simple as the fruit example, where there are only about five of them. This list doesn’t get run all the time, so performance is not as much an issue for me as it may be for others - therefore, I used the “elegant” solution listed above (just above USE UNION), as it was the clearest and easiest to modify (even if it does have to make multiple passes with a correlated subquery).

What I did was:
-First, rank the job titles numerically to create a hierarchy (I did this on a separate table which gets added to the FROM list in the SELECT statement described in the next step with an INNER JOIN). It’s more important for administrators to get the marketing piece than their peons, so they get a higher ranking. However: one may have the situation where many of the peons have the same job title (and the same rank), meaning that it’s more difficult to pull the top 5 ranks as 1, 2, 3, 4, 5 because you may have one 1 and six or more 4s, for instance.

-SELECT the entire list into a separate table (can be temporary) that has an int IDENTITY column, sorting this list first by building ID and then by the job title rank. The third sort criteria (unspecified) seems based on the ID of the record which is fairly random (so if there are two fours, the one with the lower random ID number gets selected first) - but this could be specified with even more precision if desired.

-Then use the “elegant” query to select the top x# of records at each building based on the IDENTITY column, where the lowest identity number is always the highest-ranked job-title-holders, and on down the list until you hit the x# limit.

This ensures that no more than x# of people per building will get the mailing piece, and that at the very least the top dog(s) will get it and a few others. Other criteria can obviously be added to the sorting if desired based on whatever criteria one would like - whether they’d responded to prior offers, for instance - when it goes into the sub-table to get an IDENTITY value.

If I want to make sure that a certain group of buildings (say, ones with more than 20 names at the location) get y# (let’s say five) of the pieces but ones with less than 10 get a lower number (x#, say three), then it is easy to SELECT these into two separate IDENTITY-set tables and then select from both of those tables with a UNION clause to get a final list.

Hope this practical example helps someone just like this article helped me.

create a temporary table (by using the #)

create a cursor to select only the id of each group.

loop until end of cursor and select top 1 record for each group
and insert the result to the temporary.

select all records of temporary and return the data

that worked for me! following is the store procedure i used…

CREATE PROCEDURE dbo.sp_GetExamsLastStatus (@FromDate datetime,
@ToDate datetime)
/*
(
@parameter1 int = 5,
@parameter2 datatype OUTPUT
)
*/
AS
– Declare the variables to store the values returned by FETCH.
DECLARE @ExamID int, @StatusDateTime datetime, @Status smallint;
DECLARE @PrvExamID int;

– Create temporary table
CREATE TABLE #ExamStatus(ExamID int, StatusDateTime datetime,Status smallint );

– create cursor
DECLARE Exam_Cursor CURSOR FOR
SELECT examstatus.ExamID, StatusDateTime, Status
FROM ExamStatus inner join Exam on examstatus.examid=exam.examid
where exam.examdate between @FromDate and @ToDate
ORDER BY ExamID, StatusDateTime DESC;

– open curson
OPEN Exam_Cursor;

FETCH NEXT FROM Exam_Cursor
INTO @ExamID, @StatusDateTime, @Status;
SET @PrvExamID=0;
WHILE @@FETCH_STATUS = 0 –loop until end of file
BEGIN
if @ExamID@PrvExamID –if record changed
begin
SET @PrvExamID=@ExamID –store new ID

–select the top 1 ordered by date record and
–save it to the temporary file
insert into #ExamStatus (ExamID, StatusDateTime, Status)
SELECT TOP (1) ExamID, StatusDateTime, Status
FROM ExamStatus
WHERE (ExamID = @ExamID)
ORDER BY StatusDateTime DESC;

END

FETCH NEXT FROM Exam_Cursor
INTO @ExamID, @StatusDateTime, @Status;
END;
CLOSE Exam_Cursor;

DEALLOCATE Exam_Cursor;

–show results
–select ExamID, StatusDateTime, Status from #ExamStatus ;
SELECT #ExamStatus.ExamID, Exam.ExamDate, Exam.StartTime, Exam.ExamHours, #ExamStatus.StatusDateTime, #ExamStatus.Status,
ExamClass.Description AS ClassDescription
FROM #ExamStatus INNER JOIN
Exam ON #ExamStatus.ExamID = Exam.ExamID LEFT OUTER JOIN
ExamClass ON Exam.SiteCenterID = ExamClass.SiteCenterID AND Exam.ClassID = ExamClass.ClassID
ORDER BY Exam.ExamDate,Exam.StartTime;

RETURN

select type, variety, price
from fruits
where price = (select min(price) from fruits as f where f.type = fruits.type

just wrong?

It more or less assumes that price will be unique. If the same price was repeated for 2 types of fruits and it happend to be the minumum price for one of them, the other fruit would also be returned.
Am I right?