Question

我有下表。我需要找到每个“Seq”列的最小/最大日期，其中有1或连续1。每行只能有一个“1”

Animal  Calendar_Date   SeqA    SeqB    SeqC    SeqD    SeqE
Cat     2/5/2017        0       0       0       1       0
Cat     2/6/2017        1       0       0       0       0
Cat     2/7/2017        1       0       0       0       0
Cat     2/8/2017        1       0       0       0       0
Cat     2/9/2017        1       0       0       0       0
Cat     2/10/2017       0       0       0       0       1
Cat     2/11/2017       0       0       0       0       1
Cat     2/12/2017       0       0       0       0       1
Cat     2/13/2017       0       0       0       0       1
Dog     2/5/2017        1       0       0       0       0
Dog     2/6/2017        1       0       0       0       0
Dog     2/7/2017        0       1       0       0       0
Dog     2/8/2017        0       1       0       0       0
Dog     2/9/2017        1       0       0       0       0
Dog     2/10/2017       1       0       0       0       0
Dog     2/11/2017       0       0       0       1       0
Dog     2/12/2017       0       0       0       1       0
Dog     2/13/2017       0       0       0       1       0

期望的结果是这样的。我尝试过使用row_number，rank和dense_rank但是还没能解决这个问题。

Animal  Sequence    min         Max
Cat     D           2/5/2017    2/5/2017
Cat     A           2/6/2017    2/9/2017
Cat     E           2/10/2017   2/13/2017
Dog     A           2/5/2017    2/6/2017
Dog     B           2/7/2017    2/8/2017
Dog     A           2/9/2017    2/10/2017
Dog     D           2/11/2017   2/13/2017

以下答案来自dnoeth的工作。但是，在我的数据集中有一些实例，其中Animal和Calendar_Date有多个Seq填充了1。这意味着每Seq每Animal Calendar_Date不是唯一的Seq。我不得不修改代码并运行它五次（每个union一个）和SELECT Animal, cast(Min(Trim(Both ',' FROM Seq)) as varchar(8)) as POS_Type, Min(Calendar_Date) as Min_Date, Max(Calendar_Date) as Max_Date FROM ( SELECT Animal, Calendar_Date, Seq, -- calculate groups of consecutive values Sum(flag) Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS Unbounded Preceding) AS grp FROM ( SELECT Animal, Calendar_Date, -- combine all columns into one CASE WHEN SeqA = 1 THEN 'SeqA' ELSE null END -- CASE WHEN SeqB = 1 THEN 'SeqB' ELSE null END || -- CASE WHEN SeqC = 1 THEN 'SeqC' ELSE null END || -- CASE WHEN SeqD = 1 THEN 'SeqD' ELSE null END || -- CASE WHEN SeqE = 1 THEN 'SeqE' ELSE null END AS Seq, -- check if current and previous row are different CASE WHEN Min(Seq) -- or LAG in TD16.10 Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq THEN 0 ELSE 1 END AS flag FROM Table_A --and SeqA = 1 ) AS dt where seq is not null ) AS dt GROUP BY Animal, grp Union SELECT Animal, Min(Trim(Both ',' FROM Seq)) as POS_Type, Min(Calendar_Date) as Min_Date, Max(Calendar_Date) as Max_Date FROM ( SELECT Animal, Calendar_Date, Seq, -- calculate groups of consecutive values Sum(flag) Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS Unbounded Preceding) AS grp FROM ( SELECT Animal, Calendar_Date, -- combine all columns into one -- CASE WHEN SeqA = 1 THEN 'SeqA' ELSE null END CASE WHEN SeqB = 1 THEN 'SeqB' ELSE null END -- CASE WHEN SeqC = 1 THEN 'SeqC' ELSE null END || -- CASE WHEN SeqD = 1 THEN 'SeqD' ELSE null END || -- CASE WHEN SeqE = 1 THEN 'SeqE' ELSE null END AS Seq, -- check if current and previous row are different CASE WHEN Min(Seq) -- or LAG in TD16.10 Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq THEN 0 ELSE 1 END AS flag FROM Table_A --and SeqA = 1 ) AS dt where seq is not null ) AS dt GROUP BY Animal, grp Union SELECT Animal, Min(Trim(Both ',' FROM Seq)) as POS_Type, Min(Calendar_Date) as Min_Date, Max(Calendar_Date) as Max_Date FROM ( SELECT Animal, Calendar_Date, Seq, -- calculate groups of consecutive values Sum(flag) Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS Unbounded Preceding) AS grp FROM ( SELECT Animal, Calendar_Date, -- combine all columns into one -- CASE WHEN SeqA = 1 THEN 'SeqA' ELSE null END -- CASE WHEN SeqB = 1 THEN 'SeqB' ELSE null END CASE WHEN SeqC = 1 THEN 'SeqC' ELSE null END -- CASE WHEN SeqD = 1 THEN 'SeqD' ELSE null END -- CASE WHEN SeqE = 1 THEN 'SeqE' ELSE null END AS Seq, -- check if current and previous row are different CASE WHEN Min(Seq) -- or LAG in TD16.10 Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq THEN 0 ELSE 1 END AS flag FROM Table_A --and SeqA = 1 ) AS dt where seq is not null ) AS dt GROUP BY Animal, grp Union SELECT Animal, Min(Trim(Both ',' FROM Seq)) as POS_Type, Min(Calendar_Date) as Min_Date, Max(Calendar_Date) as Max_Date FROM ( SELECT Animal, Calendar_Date, Seq, -- calculate groups of consecutive values Sum(flag) Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS Unbounded Preceding) AS grp FROM ( SELECT Animal, Calendar_Date, -- combine all columns into one -- CASE WHEN SeqA = 1 THEN 'SeqA' ELSE null END -- CASE WHEN SeqB = 1 THEN 'SeqB' ELSE null END -- CASE WHEN SeqC = 1 THEN 'SeqC' ELSE null END CASE WHEN SeqD = 1 THEN 'SeqD' ELSE null END -- CASE WHEN SeqE = 1 THEN 'SeqE' ELSE null END AS Seq, -- check if current and previous row are different CASE WHEN Min(Seq) -- or LAG in TD16.10 Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq THEN 0 ELSE 1 END AS flag FROM Table_A --and SeqA = 1 ) AS dt where seq is not null ) AS dt GROUP BY Animal, grp Union SELECT Animal, Min(Trim(Both ',' FROM Seq)) as POS_Type, Min(Calendar_Date) as Min_Date, Max(Calendar_Date) as Max_Date FROM ( SELECT Animal, Calendar_Date, Seq, -- calculate groups of consecutive values Sum(flag) Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS Unbounded Preceding) AS grp FROM ( SELECT Animal, Calendar_Date, -- combine all columns into one -- CASE WHEN SeqA = 1 THEN 'SeqA' ELSE null END -- CASE WHEN SeqB = 1 THEN 'SeqB' ELSE null END -- CASE WHEN SeqC = 1 THEN 'SeqC' ELSE null END -- CASE WHEN SeqD = 1 THEN 'SeqD' ELSE null END CASE WHEN SeqE = 1 THEN 'SeqE' ELSE null END AS Seq, -- check if current and previous row are different CASE WHEN Min(Seq) -- or LAG in TD16.10 Over (PARTITION BY Animal ORDER BY Calendar_Date ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq THEN 0 ELSE 1 END AS flag FROM Table_A --and SeqA = 1 ) AS dt where seq is not null ) AS dt GROUP BY Animal, grp它们在一起运行，如下所示。

s = set()
for x in [ {1,2}, {3,4}, {5,1} ]:
    s |= x

Answer 1

由于您只有五列，一种方法是手动取消它们，并做一个简单的GROUP BY来获得结果：

SELECT Animal, Sequence, MIN(Calendar_Date), MAX(Calendar_Date)
FROM (
    SELECT Animal, 'A' AS Sequence, Calendar_Date FROM MyTable WHERE SeqA=1
    UNION ALL
    SELECT Animal, 'B' AS Sequence, Calendar_Date FROM MyTable WHERE SeqB=1
    UNION ALL
    SELECT Animal, 'C' AS Sequence, Calendar_Date FROM MyTable WHERE SeqC=1
    UNION ALL
    SELECT Animal, 'D' AS Sequence, Calendar_Date FROM MyTable WHERE SeqD=1
    UNION ALL
    SELECT Animal, 'E' AS Sequence, Calendar_Date FROM MyTable WHERE SeqE=1
) dt
GROUP BY Animal, Sequence

Answer 2

您需要嵌套的OLAP函数：

SELECT Animal,
   Min(Trim(Both ',' FROM Seq)), 
   Min(Calendar_Date), 
   Max(Calendar_Date) 
FROM
 (
   SELECT Animal, Calendar_Date, Seq,
      -- calculate groups of consecutive values
      Sum(flag)
      Over (PARTITION BY Animal
            ORDER BY Calendar_Date
            ROWS Unbounded Preceding) AS grp
   FROM
    (
      SELECT Animal, Calendar_Date,
         -- combine all columns into one
         CASE WHEN SeqA = 1 THEN 'A' ELSE ',' END ||
         CASE WHEN SeqB = 1 THEN 'B' ELSE ',' END ||
         CASE WHEN SeqC = 1 THEN 'C' ELSE ',' END ||
         CASE WHEN SeqD = 1 THEN 'D' ELSE ',' END ||
         CASE WHEN SeqE = 1 THEN 'E' ELSE ',' END AS Seq,
         -- check if current and previous row are different
         CASE WHEN Min(Seq) -- or LAG in TD16.10
              Over (PARTITION BY Animal
                    ORDER BY Calendar_Date
                    ROWS BETWEEN 1 Preceding AND 1 Preceding) = Seq
              THEN 0
              ELSE 1
         END AS flag
      FROM tab
    ) AS dt
 ) AS dt
GROUP BY Animal, grp

Answer 3

根据您的数据，这可能是最有效的：

select animal,
       (case when seqA = 1 then 'A'
             when seqB = 1 then 'B'
             when seqC = 1 then 'C'
             when seqD = 1 then 'D'
             when seqE = 1 then 'E'
        end) as sequence,
       min(calendar_date), max(calendar_date)     
from t
group by seqA, seqB, seqC, seqD, seqE;

这是有效的，因为“seq”列中只有一列在每行中都有一个值。

老实说，我建议在表中添加一个计算列并使用它：

alter table animal
    add sequence as (case when seqA = 1 then 'A'
                          when seqB = 1 then 'B'
                          when seqC = 1 then 'C'
                          when seqD = 1 then 'D'
                          when seqE = 1 then 'E'
                     end);

然后你可以这样做：

select animal, sequence, min(calendar_date), max(calendar_date)     
from t
group by animal, sequence;

不幸的是，Teradata不支持计算列。我最初回答后添加了Teradata标签。但是，您可以使用视图获得相同的效果。

根据虚拟列和连续日期查找最小/最大日期范围

3 个答案: