aggregation according to different conditions on same column - sql

Question

I have a table #tbl like below, i need to write a query like if there are more than 3 records availble
for particular cid then avg(val of particular cid ) for particular cid should be dispalyed against each id and if there are less than
3 records availble for particular cid then avg(val of all records availble).
Please suggest.
declare #tbl table(id int, cid int, val float )
insert into #tbl
values(1,100,20),(2,100,30),(3,100,25),(4,100,31),(5,100,50),
(6,200,30),(7,200,30),(8,300,90)

Answer 1

Your description is not clear, but I believe you need windowed functions:
WITH cte AS (
SELECT *, COUNT(*) OVER(PARTITION BY cid) AS cnt
FROM #tbl
)
SELECT id, (SELECT AVG(val) FROM cte) AS Av
FROM cte
WHERE cnt <=3
UNION ALL
SELECT id, AVG(val) OVER(PARTITION BY cid) AS Av
FROM cte
WHERE cnt > 3
ORDER BY id;
DBFiddle Demo
EDIT:
SELECT id,
CASE WHEN COUNT(*) OVER(PARTITION BY cid) <= 3 THEN AVG(val) OVER()
ELSE AVG(val) OVER(PARTITION BY cid)
END
FROM #tbl
ORDER BY id;
DBFiddle Demo2

Answer 2

You can try with the following. First calculate the average for each Cid depending in it's number of occurences, then join each Cid with the Id to display all table.
;WITH CidAverages AS
(
SELECT
T.cid,
Average = CASE
WHEN COUNT(1) >= 3 THEN AVG(T.val)
ELSE (SELECT AVG(Val) FROM #tbl) END
FROM
#tbl AS T
GROUP BY
T.cid
)
SELECT
T.*,
C.Average
FROM
#tbl AS T
INNER JOIN CidAverages AS C ON T.cid = C.cid

Answer 3

Given the clarifications in comments, I am thinking this is the intention
declare #tbl table(id int, cid int, val float )
insert into #tbl
values(1,100,20),(2,100,30),(3,100,25),(4,100,31),(5,100,50),
(6,200,30),(7,200,30),(8,300,90);
select distinct
cid
, case
when count(*) over (partition by cid) > 3 then avg(val) over (partition by cid)
else avg (val) over (partition by 1)
end as avg
from #tbl;
http://dbfiddle.uk/?rdbms=sqlserver_2017&fiddle=fdf4c4457220ec64132de7452a034976
cid avg
100 31.2
200 38.25
300 38.25
There are a number of aspects of a query like this that when run at scale though are going to be pretty bad on the query plan, I'd want to test this at a larger scale and tune before using.
The description was not clear on what happened if it was exactly 3, it mentions 'more than 3' and 'less than 3' - within this code the 'more than' was used to determine which category it was in, and less than interpreted to mean 'less than or equal to 3'