进行条件聚合

Question

由于我们的业务原因，我遇到了棘手的分组问题，我有一个 table 具有这样的值


----------------------------
| NAME  |  TYPE   |  VALUE  |
----------------------------
|   N1  |   T1    |   V1    |
|   N1  |   T2    |   V2    |
|   N1  |  NULL   |   V3    |
|   N2  |   T2    |   V4    |
|   N2  |  NULL   |   V5    |
|   N3  |  NULL   |   V6    |
-----------------------------

我需要将其分组，

第一级分组将按名称。
在第二层，
- 当可用类型为T1、T2和NULL时，将T1和NULL组合在一起，将T2单独分组。
- 当可用类型为T2和NULL时，将NULL与T2分组。
- 当 NULL 是唯一可用的类型时，保持原样。

上述 table 的预期 O/P 是，

----------------------------
|   N1  |   T1    | V1+V3   |
|   N1  |   T2    |   V2    |
|   N2  |   T2    | V4+V5   |
|   N3  |  NULL   |   V6    |
-----------------------------

如何在雪花中实现这一点 sql。或任何其他服务器，以便我可以在 Snowflake 中找到等效项。

Answer 1

以下查询应该有效：

SELECT t1.NAME, COALESCE(TYPE, MIN_TYPE), SUM(VALUE)
FROM mytable AS t1
JOIN (
   SELECT NAME, MIN(TYPE) AS MIN_TYPE
   FROM mytable
   GROUP BY NAME
) AS t2 ON t1.NAME = t2.NAME    
GROUP BY t1.NAME, COALESCE(TYPE, MIN_TYPE)

查询使用派生的 table 来提取每个 NAME 的 MIN(TYPE) 值。使用 COALESCE 我们可以将 NULL 转换为 T1 或 T2.

编辑：

您可以使用以下查询创建预期结果集的透视版本：

SELECT NAME, 
       CASE 
          WHEN T1SUM IS NULL THEN 0
          ELSE COALESCE(T1SUM, 0) + COALESCE(NULLSUM,0)
       END AS T1SUM,
       CASE 
          WHEN T1SUM IS NULL AND T2SUM IS NOT NULL 
             THEN COALESCE(T2SUM, 0) + COALESCE(NULLSUM,0)
          ELSE COALESCE(T2SUM, 0)
       END AS T2SUM,
       CASE 
          WHEN T1SUM IS NULL AND T2SUM IS NULL THEN COALESCE(NULLSUM,0)
          ELSE 0
       END AS NULLSUM  
FROM (
   SELECT NAME, 
          SUM(CASE WHEN TYPE = 'T1' THEN VALUE END) AS T1SUM,
          SUM(CASE WHEN TYPE = 'T2' THEN VALUE END) AS T2SUM,
          SUM(CASE WHEN TYPE IS NULL THEN VALUE END) AS NULLSUM
   FROM mytable
   GROUP BY NAME) AS t

Answer 2

所以在 Giorgos 的回答中，总计以旋转的形式给出，或者单行 be case 形式，每个案例行数不多，这可以写得更简单：

使用此数据：

WITH data_table(name, type, value) AS (
    SELECT * FROM VALUES
    (10, 1, 100 ),
    (10, 2, 200 ),
    (10, null, 400 ),
    (11, 2, 100 ),
    (11, null, 200 ),
    (12, null, 100 )
)

和这个SQL

SELECT name
    ,SUM(IFF(type=1, value, null)) as t1_val
    ,SUM(IFF(type=2, value, null)) as t2_val
    ,SUM(IFF(type is null, value, null)) as tnull_val
    ,IFF(t1_val is not null, t1_val + zeroifnull(tnull_val), null) as c1_sum
    ,IFF(t1_val is not null, t2_val, t2_val + zeroifnull(tnull_val)) as c2_sum
    ,IFF(t1_val is null AND t2_val is null, tnull_val, null) as c3_sum
FROM data_table
GROUP BY 1;

我们得到：

NAME	T1_VAL	T2_VAL	TNULL_VAL	C1_SUM	C2_SUM	C3_SUM
10	100	200	400	500	200	null
11	null	100	200	null	300	null
12	null	null	100	null	null	100

显示 10 行的空值和与 1 和绑定，对于 11 行，空值和与 2 和绑定，在 12 行中我们自己得到零和。

如果我们愿意，我们可以对这些值进行反透视，但是加入一个有 3 行的迷你 table，如下所示：

SELECT d.name,
    p.c2 as type,
    case p.c1
        WHEN 1 then d.c1_sum
        WHEN 2 then d.c2_sum
        ELSE d.c3_sum
    end as value
FROM (
    SELECT name
        ,SUM(IFF(type=1, value, null)) as t1_val
        ,SUM(IFF(type=2, value, null)) as t2_val
        ,SUM(IFF(type is null, value, null)) as tnull_val
        ,IFF(t1_val is not null, t1_val + zeroifnull(tnull_val), null) as c1_sum
        ,IFF(t1_val is not null, t2_val, t2_val + zeroifnull(tnull_val)) as c2_sum
        ,IFF(t1_val is null AND t2_val is null, tnull_val, null) as c3_sum
    FROM data_table
    GROUP BY 1
) AS d
JOIN (
    SELECT column1 as c1, column2 as c2 
    FROM VALUES (1,'T1'),(2,'T2'),(null,'null')
) AS p 
    ON ((d.c1_sum is not null AND p.c1 = 1) 
        OR (d.c2_sum is not null AND p.c1 = 2) 
        OR (d.c3_sum is not null AND p.c1 is null))
ORDER BY 1,2;

它给出了原始请求的输出：

NAME	TYPE	VALUE
10	T1	500
10	T2	200
11	T2	300
12	null	100

进行条件聚合

Making a conditional aggregate

sql

snowflake-cloud-data-platform