T-SQL 中的 Theil-Sen 估计器函数

Question

有人有用 T-SQL 编写的Theil-Sen 回归函数吗？

我找到了一个用 Perl 编写的，但我无法将其重新编码为 SQL。

Answer 1

当我说我无法将其重新编码为 SQL 时，我在撒谎。我只是太懒了。这是带有使用示例的代码。

该代码基于TheiSen perl 库，使用QuickMedian。让我们定义一个新的表类型来轻松地将我们的数据传递给过程。

CREATE TYPE dbo.TheilSenInputDataTableType AS TABLE 
(
    ID INT IDENTITY(1,1),
    x REAL, 
    y REAL
)

请注意 ID 列，这里很重要，因为我们的解决方案使用 CROSS APPLY 语句来正确解释 TheilSen.pm 中的内部循环。

my ($x1,$x2,$y1,$y2);
foreach my $i(0 .. $n-2){
    $y1 = $y->[$i];
    $x1 = $x->[$i];
    foreach my $j($i+1 .. $n-1){
        $y2 = $y->[$j];
        $x2 = $x->[$j];

我们还需要一种新的数据类型来存储真实类型值的数组。

CREATE TYPE [dbo].[RealArray] AS TABLE(
    [val] [real] NULL
)

这是f_QuickMedian函数，返回给定数组的中位数。这要归功于 Itzik Ben-Gan。

CREATE FUNCTION [dbo].[f_QuickMedian](@RealArray RealArray READONLY)
RETURNS REAL
AS
BEGIN
    DECLARE @Median REAL;
    DECLARE @QMedian REAL;

    SELECT @Median = AVG(1.0 * val)
    FROM
    (
        SELECT o.val, rn = ROW_NUMBER() OVER (ORDER BY o.val), c.c
        FROM @RealArray AS o
        CROSS JOIN (SELECT c = COUNT(*) FROM @RealArray) AS c
    ) AS x
    WHERE rn IN ((c + 1)/2, (c + 2)/2);

    SELECT TOP 1 @QMedian = val FROM @RealArray
    ORDER BY ABS(val - @Median) ASC, val DESC

    RETURN @QMedian
END

和p_TheilSen估计器：

CREATE PROCEDURE [dbo].[p_TheilSen](
      @TheilSenInput TheilSenInputDataTableType READONLY
    , @m Real OUTPUT
    , @c Real OUTPUT
)
AS
BEGIN
    DECLARE 
        @m_arr RealArray
      , @c_arr RealArray;       

    INSERT INTO @m_arr
        SELECT m
        FROM 
        (
            SELECT  
                t1.x as x1
                , t1.y as y1
                , t2o.x as x2
                , t2o.y as y2
                , t2o.y-t1.y as [y2 - y1]
                , t2o.x-t1.x as [x2 - x1]
                , CASE WHEN (t2o.x <> t1.x) THEN  CAST((t2o.y-t1.y) AS Real)/(t2o.x-t1.x) ELSE NULL END AS [($y2-$y1)/($x2-$x1)]
                , CASE WHEN t1.y = t2o.y THEN 0
                  ELSE
                    CASE WHEN t1.x = t2o.x THEN NULL
                        ELSE 
                        -- push @M, ($y2-$y1)/($x2-$x1);
                        CAST((t2o.y-t1.y) AS Real)/(t2o.x-t1.x)
                    END
                  END as m
            FROM @TheilSenInput t1
            CROSS APPLY
                    (
                    SELECT  t2.x, t2.y
                    FROM    @TheilSenInput t2
                    WHERE   t2.ID > t1.ID
                     ) t2o
        ) t
        WHERE m IS NOT NULL 

    SELECT @m = dbo.f_QuickMedian(@m_arr)

    INSERT INTO @c_arr
        SELECT y - (@m * x)
            FROM @TheilSenInput

    SELECT @c = dbo.f_QuickMedian(@c_arr)

END

例子：

DECLARE 
      @in TheilSenInputDataTableType
    , @m Real
    , @c Real

INSERT INTO @in(x,y) VALUES (10.79,118.99)
INSERT INTO @in(x,y) VALUES (10.8,120.76)
INSERT INTO @in(x,y) VALUES (10.86,122.71)
INSERT INTO @in(x,y) VALUES (10.93,125.48)
INSERT INTO @in(x,y) VALUES (10.99,127.31)
INSERT INTO @in(x,y) VALUES (10.96,130.06)
INSERT INTO @in(x,y) VALUES (10.98,132.41)
INSERT INTO @in(x,y) VALUES (11.03,135.89)
INSERT INTO @in(x,y) VALUES (11.08,139.02)
INSERT INTO @in(x,y) VALUES (11.1,140.25)
INSERT INTO @in(x,y) VALUES (11.19,145.61)
INSERT INTO @in(x,y) VALUES (11.25,153.45)
INSERT INTO @in(x,y) VALUES (11.4,158.03)
INSERT INTO @in(x,y) VALUES (11.61,162.72)
INSERT INTO @in(x,y) VALUES (11.69,167.67)
INSERT INTO @in(x,y) VALUES (11.91,172.86)
INSERT INTO @in(x,y) VALUES (12.07,177.52)
INSERT INTO @in(x,y) VALUES (12.32,182.09)


EXEC p_TheilSen @in, @m = @m OUTPUT, @c = @c OUTPUT

SELECT @m
SELECT @c

返回：

m = 52.7079
c = -448.4853

只是为了比较，perl 版本为相同的数据集返回以下值：

m = 52.7078651685394
c = -448.484943820225

我使用 TheilSen 估算器来计算文件系统的 DaysToFill 指标。享受！