How to add Conditions on SQL Percentile Window function?

I want to do a special query to evaluate team members. Each member will have a score, and those with score that is greater than 80th percentile of that team’s score will get a bonus. However, I want to add a condition so that the 80th percentile score is calculated based only on those scores > 0.

For example, if team A has

[0, 0, 0.6, 0.6, 0.8, 0.8]

then the percentile will be calculated using

[0.6, 0.6, 0.8, 0.8]

which the result will be 0.8. Else, if the team only has scores = 0, then team_80th_score will be 0.

The table looks like this, with the team_80th_score* is the desired result.

team| member | score | team_80th_score*
----+-----------+------+--------
A   | Alex   |  0    |     0.8 
A   | Abby   |  0    |     0.8   
A   | Hunt   |  0.6  |     0.8  
A   | Tyler  |  0.6  |     0.8 
A   | Jack   |  0.8  |     0.8 
A   | Mile   |  0.8  |     0.8 
B   | John   |  0    |     0 
B   | Amy    |  0    |     0 
B   | Alice  |  0    |     0

I use Hive SQL, and understand that building on this fundamental window function will be the way to go

select team, member, score, 
percentile_approx(score, 0.8) over (partition by team) as team_80th_score
from table;

But I don’t know how to include the conditions to just consider score > 0 (for a case like team A), and if sum(score) group by team is 0 then 0 end as team_80th_score (for a case like team B).

What would you recommend me to do in this case?

Answer

Hmmmm . . . One method is to include logic in the partition by. Note that this returns meaningless values when score = 0:

select team, member, score, 
       percentile_approx(score, 0.8) over (partition by team, (case when score > 0 then 1 else 0 end) as team_80th_score
from table;

To fix that, use an outer case expression:

select team, member, score, 
       (case when score > 0
             then percentile_approx(score, 0.8) over (partition by team, (case when score > 0 then 1 else 0 end))
        end) as team_80th_score
from table;

Advertisement

Answer