使用 SQL 计算 group/partition 的累积百分位数

Question

我想计算 SQL 中给定 partition/group 的累积百分位数。例如输入数据看起来像 -

CustID     Product ID     quantity_purchased    
1          111                2
2          111                3
3          111                2 
4          111                5
1          222                2
2          222                6
4          222                7
6          222                2

我想获得每个产品 ID 组的累积个百分位数。输出应该是 -

Product ID    min      25%      50%      75%     max
    111        2        2       2.5      3.5      5
    222        2        2       2.5      5.25     7

所以，基本上对于产品 ID 111，我只需要为产品 ID 111 取 quantity_purchased 的百分位数，但是当我在该列中继续进行时，百分位数应该是产品 ID 222 的累积意义，将考虑产品 ID 111 和产品 ID 222 的 quantity_purchased 值计算百分位数 (2,3,2,5,2,6 ,7,2).同样，如果数据中有产品 ID 333，那么对于产品 ID 333，我将根据与产品 111、产品 222 和产品 333 关联的所有 quantity_purchased 值计算百分位数，并将结果存储在产品中333行。如何使用 SQL?

实现此目的

Answer 1

这使用 PERCENTILE_CONT instead of PERCENTILE_DISC 返回 val 的主要区别是基于使用线性插值的连续分布，其中值不完美排列 - 根据您的用例，这可能会提供更准确的数据点。 :-)

select
    ProductID,
    min(Quantity_Purchased::float) min,
    PERCENTILE_CONT(.25) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "25%",
    PERCENTILE_CONT(.50) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "50%",
    PERCENTILE_CONT(.75) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "75%" ,
    max(Quantity_Purchased) max
from
    cte
group by
    1

在 Snowflake

中复制|粘贴|运行

with cte as (
select
    1 CustID,
    111 ProductID,
    2 Quantity_Purchased
union
select
    2 CustID,
    111 ProductID,
    3 Quantity_Purchased
union
select
    3 CustID,
    111 ProductID,
    2 Quantity_Purchased
union
select
    4 CustID,
    111 ProductID,
    5 Quantity_Purchased
union
select
    1 CustID,
    222 ProductID,
    2 Quantity_Purchased
union
select
    2 CustID,
    222 ProductID,
    6 Quantity_Purchased
union
select
    4 CustID,
    222 ProductID,
    7 Quantity_Purchased
union
select
    6 CustID,
    222 ProductID,
    2 Quantity_Purchased
)  
select
    ProductID,
    min(Quantity_Purchased::float) min,
    PERCENTILE_CONT(.25) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "25%",
    PERCENTILE_CONT(.50) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "50%",
    PERCENTILE_CONT(.75) WITHIN GROUP (ORDER BY Quantity_Purchased ) as "75%" ,
    max(Quantity_Purchased) max
from
    cte
group by
    1

Answer 2

这非常好奇，但我认为您需要扩展每个产品 ID 的数据：

select t.product_id, min(t2.quantity_purchased), max(t2.quantity_purchased),
       percentile_cont(0.25) within group (order by t2.quantity_purchased),
       percentile_cont(0.50) within group (order by t2.quantity_purchased),
       percentile_cont(0.75) within group (order by t2.quantity_purchased)
from t join
     t t2
     on t2.product_id <= t.product_id
group by t1.product_id;

使用 SQL 计算 group/partition 的累积百分位数

Calculate cumulative percentiles using SQL for a group/partition

sql

snowflake-schema