0

如果我有一个包含这样数据的表：

x1      y1      x2      y2

A001     1      B001    2
A001     1      B002    2
A002     2      A001    1
C001     2      B003    3
C002     1      B003    3

什么看起来像 SQL 查询（Microsoft SQL 服务器），以实现这样的结果：

GroupId      x      y
1            A001   1
1            B001   2
1            B002   2
1            A002   2
2            C001   2
2            C002   1
2            B003   3

这是关于分组相等的对，例如：如果 a == b 和 b == c 然后 a == c

4

2 回答 2

2

如果不需要递归，可以使用dense_rank()函数：

;with cte1(x, y, xp, yp) as (
  select x1, y1, x2, y2 from Table1 union all
  select x2, y2, x1, y1 from Table1
), cte2 as (
  select
      x, y, xp, yp,
      case
          when x < xp then x + cast(y as nvarchar(max))
          else xp + cast(yp as nvarchar(max))
      end as grp
   from cte1  
)
select distinct
   x, y,
   dense_rank() over (order by grp) as grp
from cte2
order by grp, x, y

参见sql fiddle with example
希望对您有所帮助

更新实际上，使用表变量更容易做到这一点：

declare @Table2 table (x varchar(4), y int, grp int)
declare @Table3 table (x varchar(4), y int, xp varchar(4), yp int)
declare @i int = 1

insert into @Table2
select x, y, row_number() over (order by x, y) as grp
from
(
    select distinct x1, y1 from @Table1 union
    select distinct x2, y2 from @Table1
) as a(x, y)

insert into @Table3
select x1, y1, x2, y2 from @Table1 union
select x2, y2, x1, y1 from @Table1

while @i > 0
begin
    update T2 set
        grp = T4.grp
    from @Table2 as T2
        inner join @Table3 as T3 on T3.x = T2.x and T3.y = T2.y
        inner join @Table2 as T4 on T4.x = T3.xp and T4.y = T3.yp
    where T4.grp < T2.grp

    select @i = @@rowcount
end

select x, y, dense_rank() over (order by grp)
from @Table2

sql小提琴

于 2013-08-08T06:43:54.693 回答
1

好吧，经过一番尝试，我发现了以下内容：

DECLARE @mod int; SET @mod=1;
DECLARE @newgrp int; SET @newgrp=1;
CREATE TABLE tbl([x1] varchar(4), [y1] int, [x2] varchar(4), [y2] int);
INSERT INTO tbl
    ([x1], [y1], [x2], [y2])
          SELECT 'A001', 1, 'B001', 2  -- modified input to create chained equalities:
UNION ALL SELECT 'B001', 2, 'B002', 2  -- --> replaced A001 1 by B001 2
UNION ALL SELECT 'A002', 2, 'B002', 1  
UNION ALL SELECT 'C001', 2, 'B003', 3
UNION ALL SELECT 'C002', 1, 'B003', 3;

SELECT  x,y,0 grp INTO tmp FROM (
 SELECT x1 x,y1 y FROM tbl union SELECT x2 x, y2 y FROM tbl ) t;

-- set first seed: grp=1 on first ID only ...
UPDATE TOP(1) tmp SET grp=1;

-- now iteratively populate the tmp table 
WHILE @newgrp>0                         -- for each group
BEGIN 
 WHILE @mod>0 
  BEGIN                          -- in case of chained equalities
   UPDATE t2 SET grp=tmp.grp FROM tmp 
   INNER JOIN ( SELECT x1,x2 FROM tbl 
          UNION SELECT x2,x1 FROM tbl ) -- do group assignments in both directions!
              tt     ON tt.x1 = tmp.x AND tmp.grp>0
   INNER JOIN tmp t2 ON  t2.x = x2 AND t2.grp=0
   SET @mod=@@ROWCOUNT;
  END 
 -- OK, move on to the next group and then repeat the game ...
 UPDATE TOP(1) tmp SET grp=(SELECT MAX(grp) FROM tmp)+1 WHERE grp=0
 SELECT @newgrp=@@ROWCOUNT, @mod=1;
END

-- show the result
SELECT * FROM tmp

结果：

x    y   grp
---- --- ---
A001 1   1
A002 2   1
B001 2   1
B002 2   1
B003 3   2
C001 2   2
C002 1   2

建议的示例脚本假定 y 列与比较无关（示例数据对于每个 x 值都有一个 y 值）。如有必要，当然可以将 y 列包含在比较过程中。

编辑：

Et violá （比较现在也包括 y 列）： 而且......这是与之配套的SQLfiddle（我一开始输入了太多分号 - 愚蠢的我）！

CREATE TABLE tbl([x1] varchar(4), [y1] int, [x2] varchar(4), [y2] int);
INSERT INTO tbl
    (x1, y1, x2, y2)
          SELECT 'A001', 1, 'B001', 2
UNION ALL SELECT 'A001', 1, 'B002', 2
UNION ALL SELECT 'A002', 2, 'A001', 1
UNION ALL SELECT 'D001', 3, 'B003', 3
UNION ALL SELECT 'D003', 1, 'D001', 3
UNION ALL SELECT 'D001', 1, 'A001', 1
UNION ALL SELECT 'C001', 2, 'B003', 3
UNION ALL SELECT 'C002', 1, 'B003', 3

-- start of processing ...

SELECT  x,y,0 grp INTO tmp FROM (
 SELECT x1 x,y1 y FROM tbl union SELECT x2 x, y2 y FROM tbl ) t;

DECLARE @mod int
SET @mod=1
DECLARE @newgrp int 
SET @newgrp=1

UPDATE TOP(1) tmp SET grp=1  -- set first grp-label (seed)

-- now iteratively populate the tmp table 
WHILE @newgrp>0                         -- for each group
BEGIN 
 WHILE @mod>0                           -- in case of chained equalities
 BEGIN
   UPDATE t2 SET grp=tmp.grp FROM tmp 
   INNER JOIN ( SELECT x1,y1,x2,y2 FROM tbl 
          UNION SELECT x2,y2,x1,y1 FROM tbl ) -- do group assignments in both directions!
              tt     ON tt.x1 = tmp.x AND tt.y1 = tmp.y AND tmp.grp>0
   INNER JOIN tmp t2 ON  t2.x = tt.x2 AND  t2.y = tt.y2 AND t2.grp=0
   SET @mod=@@ROWCOUNT
 -- OK, move on to the next group and then repeat the game ...
 END
 UPDATE TOP(1) tmp SET grp=(SELECT MAX(grp) FROM tmp)+1 WHERE grp=0
 SELECT @newgrp=@@ROWCOUNT, @mod=1
END

-- show the result
SELECT * FROM tmp
-- and drop tmp again
DROP TABLE tmp

我还添加了一些示例数据以显示链式相等（'D001' 3='B003' 3和 'D003' 1= 'D001' 3）以及具有不同 y 值（'D001',1和'D001' 3）的情况。这些while循环让我有些头疼，因为我一开始没有足够注意@@ROWCOUNT 的内容……现在它应该又可以工作了！

链式等式案例（递归）是此查询中的主要问题。如果没有，一切都可以在一个语句中完成，请参阅@Roman Pekar）

我的（扩展）示例的结果：

x       y    grp
------  ---  ----
A001    1    1
A002    2    1
B001    2    1
B002    2    1
B003    3    2
C001    2    2
C002    1    2
D001    1    1
D001    3    2
D003    1    2
于 2013-08-07T16:46:23.857 回答