优先消除重复行

Question

我正在研究一个将历史与预测相结合的存储过程。我有一个位列（PHP），它指定投影是否优先于历史。我还有一列指定数据是来自历史表还是投影表。我的存储过程的输出如下所示：

CaseId     Year     Projection    PHP    Gas   Oil
  1        2004         0          1    
  1        2005         0          1    
  1        2005         1          1    
  1        2006         1          1    
  1        2007         1          1    
  1        2008         1          1    
  1        2009         1          1    
  2        2003         0          0    
  2        2004         0          0    
  2        2005         0          0    
  2        2005         1          0    
  2        2006         1          0    
  2        2007         1          0    
  2        2008         1          0    
  2        2006         1          0    

在此示例中，我需要删除第二行，因为对于 CaseId 1 投影具有优先权，因此应删除重叠的历史日期。此外，应该删除 CaseId 2 的第四行，因为历史优先。

CaseId     Year     Projection    PHP    Gas   Oil  
  1        2004         0          1    
  1        2005         1          1    
  1        2006         1          1    
  1        2007         1          1    
  1        2008         1          1    
  1        2009         1          1    
  2        2003         0          0    
  2        2004         0          0    
  2        2005         0          0    
  2        2006         1          0    
  2        2007         1          0    
  2        2008         1          0    
  2        2006         1          0

我需要在 CaseId 中标记重复年份，然后比较 Projection 和 PHP 列并删除它们不匹配的行。

这是我正在处理的查询：

SELECT      rcl.ReportRunCaseId AS CaseId, 
            year(rce.EcoDate) as Year,
            1 as Projection,
            cpq.ProjectionHasPrecedence as PHP,
            rce.GrossOil as Oil,                
            rce.GrossGas as Gas    
  from  phdreports.PhdRpt.ReportCaseList_28 rcl 
         inner join phdreports.PhdRpt.RptCaseEco_28 rce on
            rce.ReportRunCaseId = rcl.ReportRunCaseId
         inner join dbo.caseQualifier cq on 
            cq.CorpScenarioId = 1 and 
            cq.CaseCaseId = rcl.ReportRunCaseId and 
            cq.CorpQualifierTypeId = 1
         inner join dbo.caseProjectionQualifier cpq on 
            cpq.CaseCaseId = rcl.ReportRunCaseId and 
            cpq.CorpQualifierId = cq.QualifierHasData 
where rcl.ReportRunCaseId <=2
group by year(rce.EcoDate), rcl.ReportRunCaseId, cpq.ProjectionHasPrecedence, rce.GrossGas, rce.GrossOil

union all

select      rmp.ReportRunCaseId AS CaseId, 
            year(rmp.EcoDate) as Year,
            0 as Projection,
            cpq.ProjectionHasPrecedence as PHP,
            rmp.GrossOil as Oil,
            rmp.GrossGas as Gas              
from PhdReports.PhdRpt.RptMonthlyProduction_50 rmp
        inner join dbo.caseQualifier cq on 
          cq.CorpScenarioId = 1 and 
          cq.CaseCaseId = rmp.ReportRunCaseId and 
          cq.CorpQualifierTypeId = 1
        inner join dbo.caseProjectionQualifier cpq on 
          cpq.CaseCaseId = rmp.ReportRunCaseId and 
          cpq.CorpQualifierId = cq.QualifierHasData 
where rmp.ReportRunCaseId <= 2
group by year(rmp.EcoDate), rmp.ReportRunCaseId, cpq.ProjectionHasPrecedence, rmp.GrossGas, rmp.GrossOil 

如何消除 Projection 和 PHP 不匹配的重复年份？

Answer 1

ROW_NUMBER() 函数在这里应该可以帮助您：

WITH Data AS
(   SELECT      rcl.ReportRunCaseId AS CaseId, 
                year(rce.EcoDate) as Year,
                1 as Projection,
                cpq.ProjectionHasPrecedence as PHP,
                rce.GrossOil as Oil,                
                rce.GrossGas as Gas    
      from  phdreports.PhdRpt.ReportCaseList_28 rcl 
             inner join phdreports.PhdRpt.RptCaseEco_28 rce on
                rce.ReportRunCaseId = rcl.ReportRunCaseId
             inner join dbo.caseQualifier cq on 
                cq.CorpScenarioId = 1 and 
                cq.CaseCaseId = rcl.ReportRunCaseId and 
                cq.CorpQualifierTypeId = 1
             inner join dbo.caseProjectionQualifier cpq on 
                cpq.CaseCaseId = rcl.ReportRunCaseId and 
                cpq.CorpQualifierId = cq.QualifierHasData 
    where rcl.ReportRunCaseId <=2
    group by year(rce.EcoDate), rcl.ReportRunCaseId, cpq.ProjectionHasPrecedence, rce.GrossGas, rce.GrossOil

    union all

    select      rmp.ReportRunCaseId AS CaseId, 
                year(rmp.EcoDate) as Year,
                0 as Projection,
                cpq.ProjectionHasPrecedence as PHP,
                rmp.GrossOil as Oil,
                rmp.GrossGas as Gas              
    from PhdReports.PhdRpt.RptMonthlyProduction_50 rmp
            inner join dbo.caseQualifier cq on 
              cq.CorpScenarioId = 1 and 
              cq.CaseCaseId = rmp.ReportRunCaseId and 
              cq.CorpQualifierTypeId = 1
            inner join dbo.caseProjectionQualifier cpq on 
              cpq.CaseCaseId = rmp.ReportRunCaseId and 
              cpq.CorpQualifierId = cq.QualifierHasData 
    where rmp.ReportRunCaseId <= 2
    group by year(rmp.EcoDate), rmp.ReportRunCaseId, cpq.ProjectionHasPrecedence, rmp.GrossGas, rmp.GrossOil
), Data2 AS
(   SELECT  *, 
            RowNum = ROW_NUMBER() OVER(PARTITION BY CaseId, Year 
                                        ORDER BY CASE WHEN PHP = Projection THEN 0 ELSE 1 END DESC, PHP DESC, Projection DESC)
    FROM    Data
)
SELECT  CaseId, Year, Projection, PHP, Oil, Gas
FROM    Data2
WHERE   RowNum - 1;

只考虑最后一点，因为第一个只是您在公用表表达式中的查询：

RowNum = ROW_NUMBER() OVER(PARTITION BY CaseId, Year 
                            ORDER BY CASE WHEN PHP = Projection THEN 0 ELSE 1 END DESC, PHP DESC, Projection DESC)

在这里，我们给每个 caseId、year 元组一个等级，按照 PHP 是否等于投影进行排序。然后最后一部分将结果限制为每个元组的第一行，因此如果存在相等的行，则将采用该行，如果没有相等的行，则将使用不相等的行。

您可能需要在 order by 中添加更多标准以确保结果是确定性的，即如果您有两行相同的 caseId/Year，其中 PHP 和 projection 都为 1，请确保选择同一行每次。

Answer 2

我不知道您的查询与该问题有什么关系。所以，让我假设您有一个查询：

select CaseId, Year, Projection, PHP, Gas, Oil 
from t

有了这个，你可以用row_number()做你想做的事：

select CaseId, Year, Projection, PHP, Gas, Oil
from (select CaseId, Year, Projection, PHP, Gas, Oil,
             row_number() over (partition by CaseId, Year
                                order by Projection + PHP desc
                               ) as seqnum
      from t
     ) t
where seqnum = 1;

这将根据设置的标志数对行进行优先级排序。在CaseId = 2 的示例中，两行包含相同的值。这将返回这些行之一。如果要在它们之间进行选择，则需要另一列，因此请指定优先级。