用最接近的上下非空值元素差值填充空值

Question

有没有办法在 Snowflake 中使用 sql 实现以下目标？

原文Table:
订单 |值 |
1 | 80 |
2 |空 |
3 |空 |
4 | 20 |

输出table:
订单 |值 |
1 | 80 |
2 | 60 |
3 | 40 |
4 | 20 |

我知道这需要相当复杂的 window 操作，并且可能超出标准 sql。

Answer 1

因此假设order只是步长的方向而不是步长的线性映射。否则 rn 可以替换为 ov 以匹配您原始订单中的步骤。

如果您同时运行多个数据集，它没有分区。

SELECT * 
    ,lead(v) ignore nulls over (order by ov) as nv
    ,lag(v) ignore nulls over (order by ov) as pv
    ,lead(nrn) ignore nulls over (order by ov) as nnrn
    ,lag(nrn) ignore nulls over (order by ov) as pnrn
    ,nvl(v, ((nv-pv)/(nnrn-pnrn)*(rn-pnrn))+pv) as value
FROM (
    SELECT *
        ,row_number() over (order by ov) as rn
        ,nvl2(v,rn,null) as nrn
    FROM values
        (1,80),
        (2,null),
        (3,null),
        (4,20)
         t(ov, v)
)
order by ov;

给出：

OV	V	RN	NRN	NV	PV	NNRN	PNRN	VALUE
1	80	1	1	20		4		80
2		2		20	80	4	1	60
3		3		20	80	4	1	40
4	20	4	4		80		1	20

或者如果你想让它看起来整洁：

WITH data as (
    SELECT *   
    FROM values
        (1,80),
        (2,null),
        (3,null),
        (4,20),
        (5,null),
        (6,null),
        (7,11)
         t(ov, v)
)
SELECT ov, v, value 
FROM (
    SELECT * 
        ,lead(v) ignore nulls over (order by ov) as nv
        ,lag(v) ignore nulls over (order by ov) as pv
        ,lead(nrn) ignore nulls over (order by ov) as nnrn
        ,lag(nrn) ignore nulls over (order by ov) as pnrn
        ,nvl(v, ((nv-pv)/(nnrn-pnrn)*(rn-pnrn))+pv) as value
    FROM (
        SELECT *
            ,row_number() over (order by ov) as rn
            ,nvl2(v,rn,null) as nrn
        FROM data
    )
)
order by ov;

OV	V	VALUE
1	80	80
2		60
3		40
4	20	20
5		17
6		14
7	11	11

用最接近的上下非空值元素差值填充空值

Fill null values with closest upper and lower non-null value element difference

sql

snowflake-cloud-data-platform