T-SQL Statement OVER Clause Ranking Functions - sql

Question

I have a table which looks like this:
Field1 Field2 ValidFrom
200 a 01.01.1999
200 b 01.01.2015
210 c 01.01.2015
210 c 01.01.2010
Now I try to generate a select-statement with an additional column, that increments when Field1 stays the same, but Field2 changes. The order of the value in the extraColumn should depend on the ValidFrom, which means e.g. 1 for 01.01.1999 and 2 for 01.01.2015 (and not the other way around!).
When Field1 changes, the value in the extraColumn should start with 1 again. It should keep the same value, when the combination of Field1 and Field2 doesn't change.
So the result I'd like to have would look like this:
extraColumn Field1 Field2 ValidFrom
1 200 a 01.01.1999
2 200 b 01.01.2015
1 210 c 01.01.2015
1 210 c 01.01.2010
I tried to get this result by using this query with the RANK() function:
select RANK() OVER (
PARTITION BY [Field1], [Field2]
ORDER BY [ValidFrom] DESC
) as 'extraColumn'
,Field1 ,Field2 ,ValidFrom
FROM table1
Unfortunately this did not work as I expected it to and it did kind of the opposite of which I wanted, so my result looked like:
extraColumn Field1 Field2 ValidFrom
1 200 a 01.01.1999
1 200 b 01.01.2015
1 210 c 01.01.2015
2 210 c 01.01.2010
Any ideas what I did wrong?

Answer 1

Here I have answered a similar problem. Below is the adaptation to your conditions:
-- Preparation
declare #t table (
Field1 int,
Field2 char,
ValidFrom date
);
insert into #t (Field1, Field2, ValidFrom)
values
(200, 'a', '19990101'),
(200, 'b', '20150101'),
(210, 'c', '20150101'),
(210, 'c', '20100101');
-- The query
with cte as (
select t.*,
lag(t.Field2) over(partition by t.Field1 order by t.ValidFrom) as [Prev2]
from #t t
)
select c.Field1, c.Field2, c.ValidFrom,
sum(case when c.Prev2 = c.Field2 then 0 else 1 end)
over(partition by c.Field1 order by c.ValidFrom) as [ExtraColumn]
from cte c;
I only hope you aren't going to run this against millions or records, because 2 partitionings won't make it easy on the CPU and memory. Oh yes, and you need SQL Server 2012 or later for this to work.

Answer 2

Use DENSE_RANK, you need to use Field2 in order by to get the required result not ValidFrom
DENSE_RANK() OVER (PARTITION BY [Field1] ORDER BY [Field2])

Answer 3

Try it like this...
WITH
cte_MaxDate AS (
SELECT
*,
MaxDate = MAX(td.ValidFrom) OVER (PARTITION BY td.Feild1, td.Field2)
FROM
#TestData td
)
SELECT
DENSE_RANK() OVER (PARTITION BY md.Feild1 ORDER BY md.MaxDate, md.Field2),
md.Feild1, md.Field2, md.ValidFrom--, md.MaxDate
FROM
cte_MaxDate md;

Answer 4

I think you will have to use recursion here to calculate row-by-row whatever field2 changed or not.
WITH CTE_RN AS
(
SELECT *
, ROW_NUMBER() OVER (PARTITION BY Field1 ORDER BY ValidFrom) RN
FROM Table1
)
, RCTE AS
(
SELECT *, 1 AS ExtraColumn
FROM CTE_RN WHERE RN = 1
UNION ALL
SELECT c.*
, CASE WHEN r.Field2 = c.Field2 THEN r.ExtraColumn ELSE r.ExtraColumn + 1 END
FROM RCTE r
INNER JOIN CTE_RN c ON r.Field1 = c.Field1 AND r.RN + 1 = c.RN
)
SELECT *
FROM RCTE
ORDER BY Field1, ValidFrom
OPTION (MAXRECURSION 0)
SQLFiddle DEMO
(I added few more rows for more complex sample)