在SQL中以破坏序列升序/降序对结果最小值/最大值进行分组

Posted 2023-03-31

【中文标题】在SQL中以破坏序列升序/降序对结果最小值/最大值进行分组

【英文标题】：Grouping Result Min/Max in a breaking series ascending/descending order in SQL

【发布时间】：2016-03-19 05:43:54

【问题描述】：

如果序列破坏了数据顺序，我想在升序/降序序列中选择最小值和最大值

假设我有按 DateTime 顺序排列的数据：

LogDate      StartValue EndValue    Multiplier  DiffValue
2016-02-08   7661.25    7677.62     6.94        16.37
2016-02-09   7677.62    7693.02     6.94        15.4
2016-02-10   7693.02    7709.82     6.94        16.8
2016-02-11   7709.82    7727.08     6.94        17.26
2016-02-12   7727.08    7740.93     6.94        13.85
2016-02-13   3.02       12.22       6.94        9.2
2016-02-14   12.22      20.73       6.94        8.51
2016-02-15   20.73      37.04       6.94        16.31
2016-02-16   37.04      52.56       7           15.52
2016-02-17   52.56      67.82       7           15.26
2016-02-18   67.82      83.66       7           15.84
2016-02-19   83.66      98.77       7           15.11
2016-02-20   98.77      108.37      7           9.61

我想要这样的结果：

LogDateMin  LogDateMax  StartValue  EndValue    Multiplier  SumOfDiffValue
2016-02-08  2016-02-12  7661.25     7740.93     6.94        79.68
2016-02-13  2016-02-15  3.02        37.04       6.94        34.02
2016-02-16  2016-02-20  37.04       108.37      7           71.34

这里我也按 Multiplier 对结果进行分组并得到 defValue 的总和

我们怎样才能做到这一点

请帮忙

【问题讨论】：

上例中如果序列中断数据顺序是什么意思

查看秒表的示例，其数据按升序流动，但如果用户重置手表，它将从零或最小值重新启动 CurrValue

【参考方案1】：

如果我理解正确，“中断”是指随着时间的推移，值之间的最小值阈值。

为了获得结果，我使用了LEAD 和LAG 函数来查找中断，因为它们在当前记录之前和之后提供值，而不使用self JOIN。

然后，我创建了仅包含“中断”附近的第一条和最后一条记录的组。 resultset 包含日期和值作为行，因此需要 UNPIVOT。

最终查询应如下所示：

declare @Threshold NUMERIC(18, 2) = 1000

;with DeltaCte as (
    SELECT DateTime, CurrValue, 
        LAG(CurrValue, 1, CurrValue - @Threshold - 1) OVER (ORDER BY DateTime) AS PrevVal, 
        LEAD(CurrValue, 1, CurrValue - @Threshold - 1) OVER (ORDER BY DateTime) AS NextVal
    FROM RawData
)
,GroupsCTE AS (
    select DateTime, CurrValue, CurrValue - PrevVal AS Delta1, CurrValue - NextVal AS Delta2, 
        (ROW_NUMBER() OVER (ORDER BY DateTime) + 1) / 2 AS GroupNo
    FROM DeltaCte
    WHERE ABS(CurrValue - PrevVal) > @Threshold OR ABS(CurrValue - NextVal) > @Threshold
)
SELECT GroupNo, MIN(d) AS DateTimeMin, MAX(d) DateTimeMax, 
    MIN(v) AS CurrValueMin, MAX(v) CurrValueMax
from GroupsCTE
UNPIVOT (v FOR nValue IN ([CurrValue])) AS P1
UNPIVOT (d FOR nDate IN ([DateTime])) AS P2
GROUP BY GroupNo

[编辑]

如果“break”表示升序中断，上面的查询会稍微简单一些：

;with DeltaCte as (
    SELECT DateTime, CurrValue, 
        LAG(CurrValue, 1, CurrValue + 1) OVER (ORDER BY DateTime) AS PrevVal, 
        LEAD(CurrValue, 1, CurrValue - 1) OVER (ORDER BY DateTime) AS NextVal
    FROM RawData
)
,GroupsCTE AS (
    select DateTime, CurrValue, CurrValue - PrevVal AS Delta1, CurrValue - NextVal AS Delta2, (ROW_NUMBER() OVER (ORDER BY DateTime) + 1) / 2 AS GroupNo
    FROM DeltaCte
    WHERE (CurrValue - PrevVal < 0) OR (NextVal - CurrValue <  0)
)
SELECT GroupNo, MIN(d) AS DateTimeMin, MAX(d) DateTimeMax, 
    MIN(v) AS CurrValueMin, MAX(v) CurrValueMax
from GroupsCTE
UNPIVOT (v FOR nValue IN ([CurrValue])) AS P1
UNPIVOT (d FOR nDate IN ([DateTime])) AS P2
GROUP BY GroupNo

基本上，增量到阈值的比较被替换为增量到 0 的比较。

【讨论】：

非常感谢@alexei