Conditional summing in SQL - sql

Question

I'm having trouble getting a query to work. What I'm trying to do is sum a counter by user id but with conditions.
Currently my query gives the following output.
User ID EndDate Date Index
123 5/1/12 1/1/12 -1
123 5/1/12 1/25/12 1
123 5/1/12 2/13/12 -1
456 4/1/12 1/18/12 -1
456 4/1/12 2/15/12 -1
456 4/1/12 2/18/12 1
What I want to do with this list is sum the Index by User Id but with a catch. The Index must be summed in date order, also the min value of the index is -1 and max is 1, so the values can be -1, 0, 1 only. So with user 123, the process would be -1 then you add 1 then you add -1 for a final sum of -1. But for user 456 you start with -1 then you have -1 again but the sum must remain -1 then you have 1, so the final sum is 0. Below is what I'm been trying to do but I can't figure it out. I would really appreciate some help.
DECLARE #Period char(6)
DECLARE #StatusCount int
SET #Period = '201201'
SET #StatusCount = 0
SELECT Q1.UserID, Q1.End_Date,
Sum(Case
When Index = -1 Then Case When #IndexCount >=0 Then #IndexCount - 1 Else #IndexCount + 0 End
When Index = 1 Then Case When #IndexCount <=0 Then #IndexCount + 1 Else #IndexCount + 0 End
END) as FinalIndex
FROM
(
(SELECT UserID, End_Date, Enter_Dt, 1 as Index
FROM UserTable
WHERE (Code in ('A', 'B') and PRD = #Period)
GROUP BY UserID, End_Date, Enter_Dt)
UNION
(SELECT UserID, End_Date, Enter_Dt, -1 as Index
FROM UserTable
WHERE (Code in ('C', 'D') and PRD = #Period)
GROUP BY UserID, End_Date, Enter_Dt)
) as Q1
GROUP BY Q1.UserID, Q1.End_Date
ORDER BY Q1.UserID ASC, Q1.End_Date ASC
I think my main problem is I can't figure out how to accumulate the Index properly. I can't get IndexCount to remember the the previous value and then start again from 0 with the next User ID
The result I get with this query is
User ID EndDate Index
123 5/01/12 -1
456 4/01/12 -1
Which is just summing the Index

Answer 1

I'll illustrate a running total by userID solution here, leaving out the other details of your query for clarity. Basically, add a IndexRT column populate it with a running total that resets for each new userID.
EDIT: constrain running total to -1,0,1
create table #temp(userID int, [Index] int, IndexRT int)
insert into #temp (userID,[Index]) values (123,-1) , (123,1) , (123,-1) , (456,-1) , (456,-1) , (456,1)
declare #rt int; set #rt=0;
with a as (
select select TOP 100 percent *
,r=ROW_NUMBER()over(partition by userID order by userID)
from #temp order by userID,ROW_NUMBER()over(partition by userID order by userID)
)
update a set #rt = [IndexRT] = case when case when r=1 then [Index] else #rt + [Index] end between -1 and 1
then case when r=1 then [Index] else #rt + [Index] end
else #rt
end
select * from #temp;
go
drop table #temp;
Result:
userID Index IndexRT
----------- ----------- -----------
123 -1 -1
123 1 0
123 -1 -1
456 -1 -1
456 -1 -1
456 1 0

Answer 2

Sounds like you need to define a cycling sequence:
http://www.postgresql.org/docs/8.1/static/sql-createsequence.html
This will create the effect you've described in your problem, cycling -1,0,1 as rows iterate. You can achieve this in a select or upon inserting the record in your table.

Answer 3

This uses cursors, and seemed to work in SQL 2005 (I would advise checking your results thoroughly in case I missed something obvious here).
As a disclaimer, most database experts will tell you that cursors are horrible, inefficient, and should never be used. And yes-- you should use straight SELECT statements when possible. However: (1) I don't know that there is a straightforward SELECT statement for this type of accumulation, and (2) this will be a much more readable solution for anyone who has to maintain your code. If your data set is not unreasonably large (and you don't expect it to grow too much), cursors should be OK.
Note that I have substituted "Table3" for your entire FROM statement; I imported your sample table to create this example. The key element to note here is that I am sorting it first by User ID, then by Date in the DECLARE csrOrdered CURSOR statement. So you should just need to substitute your own sorted SQL statement in the cursor definition.
CREATE TABLE #sumResult (
[User ID] int,
[Sum Result] int
)
DECLARE csrOrdered CURSOR FOR
SELECT [User Id], [Index] FROM Table3 -- <- your table name goes here!
ORDER BY [User ID], Date, EndDate
DECLARE #uid int
DECLARE #index int
DECLARE #lastUid int
DECLARE #curSum int
SET #uid=0
SET #index=0
SET #lastUid=0
SET #curSum=0
OPEN csrOrdered
FETCH NEXT FROM csrOrdered
INTO #uid, #index
WHILE ##FETCH_STATUS=0
BEGIN
--Are we working on a new User ID? Then reset the sum and insert the last group into the table.
IF #uid<>#lastUid BEGIN
--Don't do an insert if we just got into the loop
IF #lastUid <> 0 BEGIN
INSERT INTO #sumResult ([User ID], [Sum Result]) VALUES (#lastUid, #curSum)
END
SET #curSum=0
SET #lastUid=#uid
END
SET #curSum = #curSum + #index
IF #curSum < -1 SET #curSum=-1
IF #curSum > 1 SET #curSum=1
FETCH NEXT FROM csrOrdered
INTO #uid, #index
END
IF #lastUid <> 0 BEGIN
INSERT INTO #sumResult ([User ID], [Sum Result]) VALUES (#lastUid, #curSum)
END
CLOSE csrOrdered
DEALLOCATE csrOrdered
SELECT * FROM #sumResult
DROP TABLE #sumResult