Question

假设我们有一个带有'A'列的表，其值为0到N.我想为每列具有相同值的30％选择“A”列。

So if I have this:
A|  B
-------
0 hello
0 test
0 hi
1 blah1
1 blah2
1 blah3
1 blah4
1 blah5
1 blah6

Result:
A|  B
-------
0 hello
1 blah1
1 blah4

它可能是blah1或任何其他不是blah4的blah，而blah4可能是任何其他不是blah1的blah，基本上它可能是随机的或跳过。

顺便说一下，实际的桌子是巨大的，谈论太字节，所以想想性能。

Answer 1

尝试这样的事情：

DECLARE @YourTable table (A int, b varchar(10))
INSERT @YourTable VALUES (0, 'hello') --OP's data
INSERT @YourTable VALUES (0, 'test')
INSERT @YourTable VALUES (0, 'hi')
INSERT @YourTable VALUES (1, 'blah1')
INSERT @YourTable VALUES (1, 'blah2')
INSERT @YourTable VALUES (1, 'blah3')
INSERT @YourTable VALUES (1, 'blah4')
INSERT @YourTable VALUES (1, 'blah5')
INSERT @YourTable VALUES (1, 'blah6')

;WITH NumberedRows AS
(   SELECT 
        A,B,ROW_NUMBER() OVER (PARTITION BY A ORDER BY A,B) AS RowNumber
        FROM @YourTable
)
, GroupCounts AS
(   SELECT
        A,MAX(RowNumber) AS MaxA
        FROM NumberedRows
        GROUP BY A
)
SELECT
    n.a,n.b
    FROM NumberedRows           n
        INNER JOIN GroupCounts  c ON n.A=c.A
    WHERE n.RowNUmber<=(c.MaxA+1)*0.3

输出：

a           b
----------- ----------
0           hello
1           blah1
1           blah2

(3 row(s) affected)

编辑基于Andriy M的评论中的好主意

;WITH NumberedRows AS
(   SELECT 
        A,B,ROW_NUMBER() OVER (PARTITION BY A ORDER BY A,B) AS RowNumber
            ,COUNT(*) OVER (PARTITION BY A) AS TotalOf
        FROM @YourTable
)
SELECT
    n.a,n.b
    FROM NumberedRows            n
    WHERE n.RowNumber<=(n.TotalOf+1)*0.3
    ORDER BY A

输出：

a           b
----------- ----------
0           hello
1           blah1
1           blah2

(3 row(s) affected)

编辑这里是“随机”行，使用Andriy M的想法：

DECLARE @YourTable table (A int, b varchar(10))
INSERT @YourTable VALUES (0, 'hello') --OP's data
INSERT @YourTable VALUES (0, 'test')
INSERT @YourTable VALUES (0, 'hi')
INSERT @YourTable VALUES (1, 'blah1')
INSERT @YourTable VALUES (1, 'blah2')
INSERT @YourTable VALUES (1, 'blah3')
INSERT @YourTable VALUES (1, 'blah4')
INSERT @YourTable VALUES (1, 'blah5')
INSERT @YourTable VALUES (1, 'blah6')

;WITH NumberedRows AS
(   SELECT 
        A,B,ROW_NUMBER() OVER (PARTITION BY A ORDER BY newid()) AS RowNumber
        FROM @YourTable
)
, GroupCounts AS (SELECT A,COUNT(A) AS MaxA FROM NumberedRows GROUP BY A)
SELECT
    n.A,n.B
    FROM NumberedRows           n
        INNER JOIN GroupCounts  c ON n.A=c.A
    WHERE n.RowNUmber<=(c.MaxA+1)*0.3
    ORDER BY n.A

输出：

a           b
----------- ----------
0           hi
1           blah3
1           blah6

(3 row(s) affected)

Answer 2

这只使用一个子查询，因此只能通过你的集合。

SELECT a
    , b
FROM 
    (
        SELECT A
            , b
            , ROW_NUMBER()
                OVER(    PARTITION BY A
                        ORDER BY b
                    ) r
            , COUNT(b)
                OVER(    PARTITION BY A
                    ) ct
        FROM @YourTable
    ) n
WHERE n.r <= n.ct * 0.3

就像这样，虽然如果少于10个并且“额外”被发布到第一个箱子，它总是返回前3个。：

SELECT A
    , b
FROM 
    (
        SELECT A
            , b
            , NTILE(10)
                OVER(   PARTITION BY a
                        ORDER BY b
                    ) tens                          
        FROM @YourTable

    ) n
WHERE tens <= 3;

选择每列值的30％

2 个答案: