Select Top n Values Across Columns and Average - sql-server

Question

I have a table that has 13 columns, one with type varchar(25) and the rest with type `int (holding values for each month of the year).
For each row, I would like to pick the top 6 int values from the 12 columns and calculate the average of those values.
I know how to select the top n from a given column, but how do you do it across multiple columns?

Answer 1

select ID,
(
select avg(C)
from (
select top(6) C
from (values(C1),(C2),(C3),(C4),(C5),(C6),(C7),
(C8),(C9),(C10),(C11),(C12)) as T(C)
order by C desc
) as T
) as C
from YourTable
SQL Fiddle
For SQL Server 2005 it would look like this since you can't use the Table Value Constructor
select ID,
(
select avg(C)
from (
select top(6) C
from (select C1 union all
select C2 union all
select C3 union all
select C4 union all
select C5 union all
select C6 union all
select C7 union all
select C8 union all
select C9 union all
select C10 union all
select C11 union all
select C12) as T(C)
order by C desc
) as T
) as C
from YourTable
SQL Fiddle
And for SQL Server 2000 this could work for you.
select T1.ID,
avg(C) as C
from (
select ID, C1 as C from YourTable union all
select ID, C2 from YourTable union all
select ID, C3 from YourTable union all
select ID, C4 from YourTable union all
select ID, C5 from YourTable union all
select ID, C6 from YourTable union all
select ID, C7 from YourTable union all
select ID, C8 from YourTable union all
select ID, C9 from YourTable union all
select ID, C10 from YourTable union all
select ID, C11 from YourTable union all
select ID, C12 from YourTable
) as T1
where (
select count(*)
from (
select ID, C1 as C from YourTable union all
select ID, C2 from YourTable union all
select ID, C3 from YourTable union all
select ID, C4 from YourTable union all
select ID, C5 from YourTable union all
select ID, C6 from YourTable union all
select ID, C7 from YourTable union all
select ID, C8 from YourTable union all
select ID, C9 from YourTable union all
select ID, C10 from YourTable union all
select ID, C11 from YourTable union all
select ID, C12 from YourTable
) as T2
where T1.ID = T2.ID and
T1.C <= T2.C
) <= 6
group by T1.ID
SQL Fiddle
I would not expect this to be particularly fast. Perhaps a better option is to store an intermediate result in a temp table.
create table #T
(
ID varchar(25),
C int
)
insert into #T
select ID, C1 as C from YourTable union all
select ID, C2 from YourTable union all
select ID, C3 from YourTable union all
select ID, C4 from YourTable union all
select ID, C5 from YourTable union all
select ID, C6 from YourTable union all
select ID, C7 from YourTable union all
select ID, C8 from YourTable union all
select ID, C9 from YourTable union all
select ID, C10 from YourTable union all
select ID, C11 from YourTable union all
select ID, C12 from YourTable
select T1.ID,
avg(C) as C
from #T as T1
where (
select count(*)
from #T as T2
where T1.ID = T2.ID and
T1.C <= T2.C
) <= 6
group by T1.ID
drop table #T

Answer 2

Firstly, it is important to understand that using TOP in conjunction with aggregates will not limit the aggregation but rather the result set. Look at this example:
SELECT TOP 2 SUM(col) FROM
(SELECT 1 AS col
UNION
SELECT 2
UNION
SELECT 3)sq
The result is still "6".
Secondly, aggregation does NOT work across columns, only rows. You would need to evaluate them manually. Probably the most efficient way would be to create a table from the row like this:
SELECT
(SELECT MAX(myval) FROM (values (col1), (col2), (col3), (col4)) as all_values(myval))
FROM (SELECT 1 as col1, 2 as col2, 3 as col3, 4 as col4)sq

Answer 3

If you are using SQL Server 2005 or later you can unpivot the table, then rank the values, and last calculate the averages the for the top 6 values per identifier.
something like this:
;WITH UnPivoted AS (
SELECT pk, MonthID, MonthNumber, MonthValue
FROM
(SELECT pk, Month1, Month2, Month3, Month4, Month5, Month6, Month7, Month8, Month9, Month10, Month11, Month12
FROM pvt) p
UNPIVOT
(pk FOR MonthNumber IN
(Month1, Month2, Month3, Month4, Month5, Month6, Month7, Month8, Month9, Month10, Month11, Month12)
)AS unpvt
),
UnPivotedRanked AS (
SELECT pk, MonthValue, RANK() OVER(PARTITION BY pk ORDER BY MonthValue DESC) AS pkRanked
FROM UnPivoted
GROUP BY pk
)
SELECT pk, AVG(MonthValue) AS Top6Average
FROM UnPivotedRanked
WHERE pkRanked < 6
GROUP BY pk