Average sum ignoring max or min - sql

Question

I want to use AVG to get the average of some values, but ignoring the max and min values only if they are 1.5 bellow or above the second max and min values. I will put some examples:
Example 1:
SELECT *
FROM (
SELECT 100.5 v FROM DUAL UNION
SELECT 101.5 v FROM DUAL UNION
SELECT 103.1 v FROM DUAL ) D
I need this result, ignoring the 103.1 value:
100.5
101.5
Example 2:
SELECT *
FROM (
SELECT 100.5 v FROM DUAL UNION
SELECT 101.5 v FROM DUAL UNION
SELECT 103.1 v FROM DUAL UNION
SELECT 106.2 v FROM DUAL) D
I need this result, ignoring only the 106.2 value:
100.5
101.1
103.1
Example 3:
SELECT *
FROM (
SELECT 100.0 v FROM DUAL UNION
SELECT 102.0 v FROM DUAL UNION
SELECT 103.0 v FROM DUAL UNION
SELECT 105.0 v FROM DUAL UNION
SELECT 107.0 v FROM DUAL) D
I need this result, ignoring 100.0 and 107.0 values:
102.0
103.0
105.0
When there is only two values it doesnt matter.
With the right result, I can AVG(value) correctly.

Answer 1

You need a combination of analytic functions (lead/lag) and conditional aggregation. Here's what I came up with. Note that I allow for multiple groups, the "adjusted" average must be computed for each group separately (a common task in statistics when you must throw out the outliers in each group, when they exist):
with
inputs ( id, val ) as (
select 101, 33 from dual union all
select 102, 23 from dual union all
select 102, 22.8 from dual union all
select 103, 30 from dual union all
select 103, 40 from dual union all
select 104, 90 from dual union all
select 104, 92 from dual union all
select 104, 92 from dual union all
select 104, 91.5 from dual union all
select 104, 91.7 from dual
)
-- End of simulated inputs (for testing only, not part of the solution).
-- SQL query begins BELOW THIS LINE. Use your actual table and column names.
select id,
avg ( case when cnt >= 3
and ( lag_val is null and lead_val - val >= 1.5
or
lead_val is null and val - lag_val >= 1.5
)
then null
else val
end
) as adjusted_avg_val
from (
select id, val, count(val) over (partition by id) as cnt,
lag ( val ) over ( partition by id order by val ) as lag_val,
lead ( val ) over ( partition by id order by val ) as lead_val
from inputs
)
group by id
;
Output:
ID ADJUSTED_AVG_VAL
--- ----------------
101 33
102 22.9
103 35
104 91.8

Answer 2

Try using the following combination of row_number lead and lag
with cte as (
SELECT 100.5 v FROM DUAL UNION ALL
SELECT 101.5 v FROM DUAL UNION ALL
SELECT 103.1 v FROM DUAL UNION ALL
SELECT 106.2 v FROM DUAL)
-- end of sample data
select avg(v)
from
(
select row_number() over (order by v desc) arn,
row_number() over (order by v) drn,
lag(v) over (order by v) av,
lead(v) over (order by v) dv,
v
from cte
) t
where (arn != 1 and drn != 1) or -- if they are no maximum nor minumum
(drn = 1 and v + 1.5 > dv) or -- if they are minimum
(arn = 1 and v - 1.5 < av) or -- if they are maximum
(av is null and arn < 3) or -- if there are just two ore one value
(dv is null and drn < 3) -- if there are just two ore one value

Answer 3

In SQL you simply just need to express the result, so ...
WITH D as(
SELECT 100.0 v FROM DUAL UNION
SELECT 102.0 FROM DUAL UNION
SELECT 103.0 FROM DUAL UNION
SELECT 105.0 FROM DUAL UNION
SELECT 107.0 FROM DUAL)
SELECT avg(v)
FROM D
where (v < (select max(v) from D )
and ((select max(v) from D )
-(select max(v) from D where v !=
(select max(v) from D ) ) > 1.5))
or
(v > (select min(v) from D )
and ((select min(v) from D )
+(select min(v) from D where v !=
(select min(v) from D ) ) > 1.5))
... should do the trick!
But thinking ahead... the below version may also be useful ;)
WITH D as(
SELECT 1 PK, 100.0 v FROM DUAL UNION
SELECT 1,102.0 FROM DUAL UNION
SELECT 1,103.0 FROM DUAL UNION
SELECT 1,105.0 FROM DUAL UNION
SELECT 1,107.0 FROM DUAL)
SELECT PK,avg(v)
FROM D
where (v < (select max(v) from D group by PK)
and ((select max(v) from D group by PK)
-(select max(v) from D where v !=
(select max(v) from D group by PK) group by PK) > 1.5))
or
(v > (select min(v) from D group by PK)
and ((select min(v) from D group by PK)
+(select min(v) from D where v !=
(select min(v) from D group by PK) group by PK) > 1.5))
GROUP BY PK
In real life though you would consider the execution plans of the above in a large dataset too (homework).
For any further clarifications, I am at your disposal via comments.
Sincerely,
Ted