October 23, 2008 at 6:30 pm
Please help me!
I have this values that I need to count based on condition.
- If the value is between 10 and 15 and the code is = a then start count once I find the first value that meets the criteria up to 10 and start all over once I find another value that meets the criteria again.
CREATE TABLE #Counter
(
StartDate datetime,
value float,
code char(1)
)
INSERT INTO #Counter
VALUES ('08/10/2008 01:36', 3, 'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:42', 2, 'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:48', 2, 'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:54', 12, 'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:00', 13, 'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:06', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:12', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:18', 15,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:24', 15,'b')
INSERT INTO #Counter
VALUES ('08/10/2008 02:30', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:36', 4,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:42', 5,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:48', 6,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 02:54', 6,'b')
INSERT INTO #Counter
VALUES ('08/10/2008 03:00', 12,'b')
INSERT INTO #Counter
VALUES ('08/10/2008 03:06', 12,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:12', 1,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:18', 12,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:24', 11,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:30', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:36', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:42', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:48', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 03:54', 3,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 04:00', 4,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 04:06', 4,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:48', 4,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:48', 11,'a')
INSERT INTO #Counter
VALUES ('08/10/2008 01:48', 3,'a')
select * from #Counter
StartDate value code Counter
2008-08-10 01:36:00.0003anull
2008-08-10 01:42:00.0002anull
2008-08-10 01:48:00.0002anull
2008-08-10 01:54:00.00012a1
2008-08-10 02:00:00.00013a2
2008-08-10 02:06:00.0003a3
2008-08-10 02:12:00.0003a4
2008-08-10 02:18:00.00015a5
2008-08-10 02:24:00.00015b6
2008-08-10 02:30:00.0003a7
2008-08-10 02:36:00.0004a8
2008-08-10 02:42:00.0005a9
2008-08-10 02:48:00.0006a10
2008-08-10 02:54:00.0006bnull
2008-08-10 03:00:00.00012bnull
2008-08-10 03:06:00.00012a1
2008-08-10 03:12:00.0001a2
2008-08-10 03:18:00.00012a3
2008-08-10 03:24:00.00011a4
2008-08-10 03:30:00.0003a5
2008-08-10 03:36:00.0003a6
2008-08-10 03:42:00.0003a7
2008-08-10 03:48:00.0003a8
2008-08-10 03:54:00.0003a9
2008-08-10 04:00:00.0004a10
2008-08-10 04:06:00.0004anull
2008-08-10 01:48:00.0004anull
2008-08-10 01:48:00.00011a1
2008-08-10 01:48:00.0003a2
October 23, 2008 at 6:38 pm
Hint: You will have to use ROW_NUMBER OVER ( PARTITION BY ) 🙂
October 23, 2008 at 6:40 pm
Nikhil,
I already tried that and it is more complicated than I thought.
October 23, 2008 at 8:00 pm
Nikhil Shikarkhane (10/23/2008)
Hint: You will have to use ROW_NUMBER OVER ( PARTITION BY ) 🙂
I'd really like to see that for this one. Got code? 😉
--Jeff Moden
Change is inevitable... Change for the better is not.
October 23, 2008 at 8:02 pm
JohnDBA (10/23/2008)
Please help me!I have this values that I need to count based on condition.
- If the value is between 10 and 15 and the code is = a then start count once I find the first value that meets the criteria up to 10 and start all over once I find another value that meets the criteria again.
Couple of questions...
1. How many rows are in the real table?
2. Do you want to update the original table with the new count?
--Jeff Moden
Change is inevitable... Change for the better is not.
October 24, 2008 at 12:33 am
Is there any sequential unique column that we can use for sorting? If none, can you insert one, an IDENTITY, preferably.
-- CK
October 24, 2008 at 3:16 am
The solution to this is very much like a solution I posted earlier, but not without a row identifier as mentioned before.
October 24, 2008 at 6:42 am
This should work for you:
[font="Courier New"]
-- create a table to store our results
-- create a dummy unique identifier on the table
CREATE TABLE #Counter2
(
RowNumber INT IDENTITY(1,1)
,StartDate DATETIME
,[Value] FLOAT
,Code CHAR(1)
,Counter INT
)
--insert all existing data from original table into our results table
INSERT INTO #Counter2(StartDate, [Value], Code)
SELECT StartDate
, [Value]
, Code
FROM #Counter
-- create a counter variable
-- this will be used to count our rows
DECLARE @Counter INT
SET @Counter = 0
-- this is a dummy variable
-- we need it as an anchor to keep our running total
DECLARE @RowNumber INT
-- now we will set the counter
-- we need to start the counter when the value is between 10 and 15
-- accumulate the counter until we get to 10
-- start again when we get to the next value between 10 and 15
UPDATE #Counter2
SET @Counter = Counter = CASE WHEN (@Counter >= 1 AND @Counter < 10) OR ([Value] >= 10 AND [Value] <=15)
THEN @Counter + 1
ELSE 0 -- restart the counter
END
,@RowNumber = RowNumber
FROM #Counter2
-- take a look at our results
SELECT *
FROM #Counter2
-- clean up
DROP TABLE #Counter2
[/font]
October 24, 2008 at 6:45 am
P.S. You may want to read Jeff's cool article on Running Totals which explains how this approach works http://www.sqlservercentral.com/articles/Advanced+Querying/61716/
October 24, 2008 at 7:29 am
AnzioBake (10/24/2008)
The solution to this is very much like a solution I posted earlier, but not without a row identifier as mentioned before.
So, you shouldn't have a problem posting the code or at least a URL here, huh? 😉
--Jeff Moden
Change is inevitable... Change for the better is not.
October 24, 2008 at 7:31 am
ggraber (10/24/2008)
P.S. You may want to read Jeff's cool article on Running Totals which explains how this approach works http://www.sqlservercentral.com/articles/Advanced+Querying/61716/
Heh... that's where I was going with this... I just wanted to know if he wanted his original table to be updated or if I was going to have to use a temp table.
--Jeff Moden
Change is inevitable... Change for the better is not.
October 24, 2008 at 8:22 am
Graber,
You are very close to what I need. But, row 16 should start with 1 and not row 15 becuase:
- the conter needs to start only when the code is = a in row 15 the first criteria is correct but code start with b ant not with a
criteria:
- If the value is between 10 and 15 and the code is = a then start count once I find the first value that meets the criteria up to 10 and start all over once I find another value that meets the criteria again.
Thank you so much!
October 24, 2008 at 8:28 am
You just need to add one criteria to the CASE statement. Like this:
[font="Courier New"]
UPDATE #Counter2
SET @Counter = Counter = CASE WHEN (@Counter >= 1 AND @Counter < 10) OR ([Value] >= 10 AND [Value] <=15 AND Code = 'a')
THEN @Counter + 1
ELSE 0 -- restart the counter
END
,@RowNumber = RowNumber
FROM #Counter2
[/font]
October 24, 2008 at 8:29 am
Jeff Moden (10/24/2008)
ggraber (10/24/2008)
P.S. You may want to read Jeff's cool article on Running Totals which explains how this approach works http://www.sqlservercentral.com/articles/Advanced+Querying/61716/Heh... that's where I was going with this... I just wanted to know if he wanted his original table to be updated or if I was going to have to use a temp table.
I figured. Funny how a lot of the questions on the forum revolve around the same few concepts.
October 24, 2008 at 8:39 am
JohnDBA (10/23/2008)
Please help me!I have this values that I need to count based on condition.
- If the value is between 10 and 15 and the code is = a then start count once I find the first value that meets the criteria up to 10 and start all over once I find another value that meets the criteria again.
John,
GGraber is exactly on the right track but there's a couple of things that are missed.
Didn't start the count in regards to Code "a" .
Used "0" instead of NULL.
Forgot to include a Clustered Index to GUARANTEE the order of the update (although it worked without in this example, ya gotta have it to guarantee the order... and this only works with Updates, not Selects).
So, using your fine test data and GGraber's good code, here's a very high performance solution using the "running total" method that GGraber cited... as with GGraber's code, the details are in the comments...
--===== Create a temporary working table with a dummy unique identifier (RowNumber)
-- that will also be the object of the Clustered Index as a PK. Notice also,
-- that we've added a "Counter" column to hold the result of your request.
CREATE TABLE #Counter2
(
RowNumber INT IDENTITY(1,1),
StartDate DATETIME,
Value FLOAT,
Code CHAR(1),
Counter INT
)
--===== Copy all existing data from original table into our results table
-- in the correct order.
INSERT INTO #Counter2(StartDate, Value, Code)
SELECT StartDate, Value, Code
FROM #Counter
ORDER BY StartDate
--===== MUST have a clustered primary key to GUARANTEE this will work
ALTER TABLE #Counter2
ADD PRIMARY KEY CLUSTERED (RowNumber)
--===== Create the required local variables
DECLARE @Counter INT, -- Start value is NULL
@RowNumber -- Dummy variable to "anchor" the "running" update
--===== Calculate and store the value of the counter in the table using a "pseudo-cursor"
-- the order of which is driven by a forced scan on the Clustered Index.
-- Rules for the counter value are...
--
-- 1. Any value not in the following criteria will result in NULL for the Counter
-- 2. Data must be processed in order by StartDate.
-- 3. Counter will start at "1" at the first occurance where the Value is between 10 and 15 and
-- and the Code is "a".
-- 4. Count will increase by 1
-- 5. When the Counter reaches the value of 10, reset the Counter to NULL
-- 6. Do until all rows in RowNumber order have been updated.
UPDATE #Counter2
SET @Counter = Counter = CASE WHEN ([Value] BETWEEN 10 AND 15) AND Code = 'a' AND @Counter IS NULL THEN 1 -- Rules 1 & 3
WHEN @Counter = 10 THEN NULL -- Rule 5
ELSE @Counter + 1 -- Rule 4
END,
@RowNumber = RowNumber
FROM #Counter2 WITH(INDEX(0)) --LOOK! Must refer to the clustered index or could bomb! -- Rules 2 and 6
--===== Display the results in RowNumber order
SELECT * FROM #Counter2 ORDER BY RowNumber
--===== Housekeeping so we can run this example again, if we want.
-- This is NOT necessary in final production stored procs because
-- it will automatically drop at the end of the proc (how poetic ;-)
DROP TABLE #Counter2
Someone will likely figure out and post a way to do this with ROW_NUMBER() or RANK(), but this solution works on SQL Server 2000, as well.
--Jeff Moden
Change is inevitable... Change for the better is not.
Viewing 15 posts - 1 through 15 (of 28 total)
You must be logged in to reply to this topic. Login to reply