Question

I have a table in a SQL Server database which stores historical data on daily basis. The structure is shown below:

UploadDate    TypeID    Value1   Value2
-------------------------------------------
2012-01-08    1         NEG      1998-02-05
2012-01-08    2         NEG      1999-02-09
2012-01-08    3         STABLE   1997-02-06
2012-02-08    1         NEG      1998-02-05
2012-02-08    2         NEG      1999-02-09
2012-03-08    1         POS      2012-03-08
2012-03-08    2         STABLE   2012-01-08

As you can see above for the TypeID 1 & 2, Value1 and Value2 has changed on 2012-03-08

My requirement is such that I have to show only those rows which have changed from previous values.

In this case since TypeID 1 & 2 have changed than it should show the current and most nearest previous value. And for TypeID 3 since it has not changed, it will will only show the most current values. The result set would look something like below:

UploadDate    TypeID    Value1   Value2
-------------------------------------------
2012-01-08    3         STABLE   1997-02-06
2012-02-08    1         NEG      1998-02-05
2012-02-08    2         NEG      1999-02-09
2012-03-08    1         POS      2012-03-08
2012-03-08    2         STABLE   2012-01-08

Any idea how I can tackle this using SQL?

Answer 1

Uninspired version uses self-join on ordered set to check the value of chronologically previous row of the same typeid. If there is no previous row or values are different the row is output.

; with numbered as (
  select *,
         row_number() over (order by typeid, uploaddate) rn
    from table1
)
select n1.*
  from numbered n1
  left join numbered n2
    on n1.TypeID = n2.TypeID
   and n1.rn + 1 = n2.rn
 where (n2.rn is null 
    or n1.value1 <> n2.value1
    or n1.value2 <> n2.value2)
 order by typeid, uploaddate

Here is Sql Fiddle with example.

UPDATE: another variant which does not require self-join but does require group by. Each timeline of same typeid, value1 and value2 are given unique group_number which is used later to extract max(uploaddate) for the group.

; with numbered as (
  select *,
         row_number() over (order by typeid, uploaddate)
       - row_number() over (partition by typeid, value1, value2 
                            order by uploaddate) group_number
    from table1
)
select max(uploaddate) uploaddate, typeid, value1, value2
  from numbered
group by typeid, value1, value2, group_number
order by typeid, uploaddate

Another Sql Fiddle.