按月分组获取数据集的百分位数

问题描述

ew。这是一个真正的脑筋急转弯。首先，我用于测试的表架构为：

Create Table scores 
( 
    Id int not null identity(1,1) primary key clustered
    , [Date] datetime not null
    , score int not null
)

现在，首先，我在sql 2008中使用CTE计算值以检查答案，然后构建了一个可以在sql 2000中使用的解决方案。因此，在sql 2008中，我们执行以下操作：

;With 
    SummaryStatistics As
    (
        Select Year([Date]) As YearNum
            , Month([Date]) As MonthNum
            , Min(score) As Minscore
            , Max(score) As Maxscore
            , Avg(score) As Avgscore
        From scores
        Group By Month([Date]), Year([Date])
    )
    , percentiles As
    (
        Select Year([Date]) As YearNum
            , Month([Date]) As MonthNum
            , score
            , NTile( 100 ) Over ( Partition By Month([Date]), Year([Date]) Order By score ) As percentile
        From scores
    )
    , Reportedpercentiles As
    (
        Select YearNum, MonthNum
            , Min(Case When percentile = 45 Then score End) As percentile45
            , Min(Case When percentile = 55 Then score End) As percentile55
        From percentiles
        Where percentile In(45,55)
        Group By YearNum, MonthNum
    )
Select SS.YearNum, SS.MonthNum
    , SS.Minscore, SS.Maxscore, SS.Avgscore
    , RP.percentile45, RP.percentile55
From SummaryStatistics As SS
    Join Reportedpercentiles As RP
        On  RP.YearNum = SS.YearNum
            And RP.MonthNum = SS.MonthNum
Order By SS.YearNum, SS.MonthNum

现在是一个 sql 2000解决方案。本质上，诀窍是使用几个临时表来计算分数的出现。

If object_id('tempdb..#Working') is not null
    DROP TABLE #Working
GO
Create Table #Working 
    (
    YearNum int not null
    , MonthNum int not null
    , score int not null
    , Occurances int not null
    , Constraint PK_#Working Primary Key Clustered ( MonthNum, YearNum, score )
    )
GO
Insert #Working(MonthNum, YearNum, score, Occurances)
Select Month([Date]), Year([Date]), score, Count(*)
From scores
Group By Month([Date]), Year([Date]), score
GO
If object_id('tempdb..#SummaryStatistics') is not null
    DROP TABLE #SummaryStatistics
GO
Create Table #SummaryStatistics
    (
    MonthNum int not null
    , YearNum int not null
    , score int not null
    , Occurances int not null
    , Cumulativetotal int not null
    , percentile float null
    , Constraint PK_#SummaryStatistics Primary Key Clustered ( MonthNum, YearNum, score )
    )
GO
Insert #SummaryStatistics(YearNum, MonthNum, score, Occurances, Cumulativetotal)
Select W2.YearNum, W2.MonthNum, W2.score, W2.Occurances, Sum(W1.Occurances)-W2.Occurances
From #Working As W1
    Join #Working As W2 
        On W2.YearNum = W1.YearNum
            And W2.MonthNum = W1.MonthNum
Where W1.score <= W2.score
Group By W2.YearNum, W2.MonthNum, W2.score, W2.Occurances

Update #SummaryStatistics
Set percentile = SS.Cumulativetotal * 100.0 / MonthTotal.Total
From #SummaryStatistics As SS
    Join    (
            Select SS1.YearNum, SS1.MonthNum, Max(SS1.Cumulativetotal) As Total
            From #SummaryStatistics As SS1
            Group By SS1.YearNum, SS1.MonthNum
            ) As MonthTotal
        On MonthTotal.YearNum = SS.YearNum
            And MonthTotal.MonthNum = SS.MonthNum

Select GeneralStats.*, percentiles.percentile45, percentiles.percentile55
From    (
        Select  Year(S1.[Date]) As YearNum
            , Month(S1.[Date]) As MonthNum
            , Min(S1.score) As Minscore
            , Max(S1.score) As Maxscore
            , Avg(S1.score) As Avgscore
        From scores As S1
        Group By Month(S1.[Date]), Year(S1.[Date])
        ) As GeneralStats
    Join    (
            Select SS1.YearNum, SS1.MonthNum
                , Min(Case When SS1.percentile >= 45 Then score End) As percentile45
                , Min(Case When SS1.percentile >= 55 Then score End) As percentile55
            From #SummaryStatistics As SS1
            Group By SS1.YearNum, SS1.MonthNum 
            ) As percentiles
        On percentiles.YearNum = GeneralStats.YearNum
            And percentiles.MonthNum = GeneralStats.MonthNum

解决方法

我有一个带有全部记录的SQL表，如下所示：

| Date       | Score |
+ -----------+-------+
| 01/01/2010 |     4 |
| 02/01/2010 |     6 |
| 03/01/2010 |    10 |
  ...
| 16/03/2010 |     2 |

我将其绘制在图表上，因此我在图表上得到一条漂亮的线，表示随时间推移的得分。迷人的。

现在，我需要做的是在图表上包括平均得分，以便我们可以看到平均得分随时间的变化，因此我可以简单地将其添加到组合中：

SELECT 
    YEAR(SCOREDATE) 'Year',MONTH(SCOREDATE) 'Month',MIN(SCORE) MinScore,AVG(SCORE) AverageScore,MAX(SCORE) MaxScore
FROM SCORES
GROUP BY YEAR(SCOREDATE),MONTH(SCOREDATE) 
ORDER BY YEAR(SCOREDATE),MONTH(SCOREDATE)

到目前为止没有问题。

问题是，如何轻松计算每个时间段的百分位数？我不确定这是正确的短语。我总共需要的是：

图表上的分数线（简单）
图表上的平均值线（简单）
图表上的一条线显示了95％的分数占据的频段（已绊脚步）

这是我不明白的第三个。我需要计算5％的百分位数，我可以单独这样做：

SELECT MAX(SubQ.SCORE) FROM
    (SELECT TOP 45 PERCENT SCORE 
    FROM SCORES
    WHERE YEAR(SCOREDATE) = 2010 AND MONTH(SCOREDATE) = 1
    ORDER BY SCORE ASC) AS SubQ

SELECT MIN(SubQ.SCORE) FROM
    (SELECT TOP 45 PERCENT SCORE 
    FROM SCORES
    WHERE YEAR(SCOREDATE) = 2010 AND MONTH(SCOREDATE) = 1
    ORDER BY SCORE DESC) AS SubQ

但是我无法弄清楚如何获得所有月份的表格。

| Date       | Average | 45% | 55% |
+ -----------+---------+-----+-----+
| 01/01/2010 |      13 |  11 |  15 |
| 02/01/2010 |      10 |   8 |  12 |
| 03/01/2010 |       5 |   4 |  10 |
  ...
| 16/03/2010 |       7 |   7 |   9 |

此刻，我将不得不把这些加载到我的应用程序中，然后自己计算这些数字。或运行大量的单个查询并整理结果。

sql sql