如何使用 PostgreSQL 9.2 计算游戏中每个级别的百分位数

Question

I have a table of game logs. Like this:

Level Shuffle_Count
  1        3
  2        1
  2        2
  2        1
  3        0
  3        4

That means whenever a user plays a level, a row is added to table. These rows have the level data showing which level was played by user and the shuffle_count data showing how many times shuffle happened during that level.

I want to know how many times shuffle occurs in every level by calculating the median of shuffle_count for every level. In the below code, I can find the median of level 2 separately. Firstly, I create a temporary table which orders shuffle_counts and divide them to 4 even groups with ntile. Then I select the min shuffle_count which has value of 3 within the new column named quartile.

with ranked_test as (
    SELECT shuffle_count, ntile(4) OVER (ORDER BY shuffle_count) AS quartile FROM ch.public.game_log WHERE level = 2
)
SELECT min(shuffle_count) FROM ranked_test
WHERE quartile = 3
GROUP BY quartile;

This is the table created before selecting min shuffle_count where quartile = 3 (which is median approximately):

Shuffle_Count quartile
     0           1
     0           1
     2           2
     3           2
     4           3
     8           3
     12          4
     19          4

So far so good. But the problem is that I have over 1000 levels and I can't do that manually for each level. I need the median value of shuffle_count for every level from 1 to 1000. I know this could be done with one row in PostgreSQL 9.4 but I unfortunately don't have that option right now.

I couldn't make this happen with a simple Group By. I guess I need more complex query including FOR or something.

Do you have any idea, guys? Thanks in advance.

Answer 1

I think that this should do it for your use case:

with ranked_test as (
    select 
        level,
        shuffle_count, 
        ntile(4) over(partition by level order by shuffle_count) quartile 
    from ch.public.game_log
)
select level, quartile , min(shuffle_count) 
from ranked_test
where quartile = 3
group by level, quartile;

This is basically an extended version of your working query:

• in the CTE, we remove the filter on level in the subquery, and add it to the partition by of the window function instead

• the outer query, we add the level to the select and group by clause