Microsoft SQL 和 R、存储过程和 k-means

Question

我是新来的，希望得到帮助和帮助。

但是，我正在开发新的 Microsoft Sql Server Management Studio (2016)，使用它的新功能，这意味着与 R 的集成。首先，我的目标是创建一个存储过程，该过程使用 x 和 y 列执行 K-Means 聚类。

问题是我被卡在了中间，因为我无法拒绝对我的案例的在线文档。

这里是脚本

    CREATE TABLE [dbo].[ModelTable]
    (
    column_name1 varchar(8000)
    )
    ;

    CREATE TABLE [dbo].[ResultTable]
    (
    column_name1 varchar(8000),
    column_name2 varchar(8000),
    column_name3 varchar(8000),
    column_name4 varchar(8000)
    )
    ;

    CREATE PROCEDURE [dbo].[kmean]  

    AS  
    BEGIN  
    DECLARE @inquery nvarchar(max) = N'  
              select  name,x,y FROM [dbtable] 

   '  
   -- then I decide to insert the model in a table: this is similar to  the documentation, but I am not sure it fits well.  

   INSERT INTO [dbo].[ModelTable] 
   EXEC sp_execute_external_script @language = N'R',  
                                   @script = N'  

   ## Here I create model: this is one of the biggest problem, because I tried to create a data frame with the data, but I do not know if here,
   ## in the R code, the data are read in this way. Generally in "pure"  R, I write data.frame(sourcedata$x,sourcedata$y), but here, where   is source of data?
   ## In the documentation it is used ImputDataSet, so maybe I could do: 

    trained_model <- kmeans(data.frame(ImputDataSet$x,ImputDataSet$y),8)  

    -- If everything is ok (doubtfully) I should have the model. And here, the part that I really cannot handle.
    -- I'd like to have a table [ResultTable] with name, variable x, variable y,  and trainedmodel$cluster.

    ',  
                              @input_data_1 = @inquery,  
                              @output_data_1_name = N'trained_model'  
     ;  


    END  
    GO  

    EXEC kmean

嗯，还有很多问题等等，由于这是MSSMS中比较新的功能，所以网上的帮助等也不是很多。提前致谢

Answer 1

我们可以尝试以下方法：

CREATE TABLE #tempData (x float not null, y float not null); 
INSERT INTO #tempData   VALUES (0, 0), (0.1, 0.1), (1, 1), (1.1, 1.1);

CREATE TABLE #output (x float, y float, Cluster int); 

INSERT INTO #output
EXECUTE  sp_execute_external_script
                @language = N'R'
              , @script = N'                
                        trained_model <- kmeans(df[, c("x", "y")], 2)
                        df$cluster <- trained_model$cluster
                        '
              , @input_data_1 = N'SELECT * from #tempData'
              , @output_data_1_name = N'df'
              , @input_data_1_name = N'df';

SELECT *
FROM #output

输出：

x   y   Cluster
0   0   1
0.1 0.1 1
1   1   2
1.1 1.1 2

请注意，我将输入和输出数据指定为 df。默认值是 InputDataSet 和 OutputDataSet.

如果您有更长的 R 脚本：我建议您在 R 环境中编写和测试它们，然后将它们保存在一个包中，然后只需加载并调用它们。

Microsoft SQL 和 R、存储过程和 k-means

Microsoft SQL and R, stored procedure and k-means

sql-server

r

k-means