Teradata SQL如何将“按日期”转移到“日期范围”？

Question

我有 6 亿行，如下表 1 所示。在 Teradata SQL 中，如何将“按日期”转换为按“日期范围”？

+-----------+-------+------------+----------+
| ProductID | Store | Trans_Date | Cost_Amt |
+-----------+-------+------------+----------+
|     20202 |  2320 | 2018-01-02 |  $9.23   |
|     20202 |  2320 | 2018-01-03 |  $9.23   |
|     20202 |  2320 | 2018-01-04 |  $9.23   |
|     20202 |  2320 | 2018-01-05 |  $9.38   |
|     20202 |  2320 | 2018-01-06 |  $9.38   |
|     20202 |  2320 | 2018-01-07 |  $9.38   |
|     20202 |  2320 | 2018-01-08 |  $9.23   |
|     20202 |  2320 | 2018-01-09 |  $9.23   |
|     20202 |  2320 | 2018-01-10 |  $9.23   |
+-----------+-------+------------+----------+

想要的输出：

+-----------+-------+------------+------------+----------+
| ProductID | Store | Start Date |  End Date  | Cost_Amt |
+-----------+-------+------------+------------+----------+
|     20202 |  2320 | 2018-01-02 | 2018-01-04 |  $9.23   |
|     20202 |  2320 | 2018-01-05 | 2018-01-07 |  $9.38   |
|     20202 |  2320 | 2018-01-08 | 2018-01-10 |  $9.23   |
+-----------+-------+------------+------------+----------+

Answer 1

如果每个 ProductID,Store,Trans_Date 有一行（没有丢失日期），您可以应用专有的 Teradata 语法来规范重叠的日期范围：

SELECT 
   ProductID
  ,Store
  -- split the period back into seperate dates
  ,Begin(pd) AS StartDate
  ,Last(pd) AS EndDate -- must be LAST not END to match your logic
  ,Cost_Amt
FROM
 (
   SELECT NORMALIZE 
      ProductID
     ,Store
     -- NORMALIZE only works with periods, so create it on the fly
     ,PERIOD(Trans_Date,Trans_Date+1) AS pd
     ,Cost_Amt
   FROM bigtable
 ) AS dt

对于create table，还有一个normalize 选项。

Answer 2

我将示例扩展为：

CREATE TABLE bigtable(
   ProductID  INTEGER 
  ,Store      INTEGER 
  ,Trans_Date DATE 
  ,Cost_Amt   VARCHAR(10)
);
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-02','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-03','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-04','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-05','$9.38');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-06','$9.38');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-07','$9.38');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-08','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-09','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-10','$9.23');
INSERT INTO bigtable(ProductID,Store,Trans_Date,Cost_Amt) VALUES (20202,2320,'2018-01-11','$9.38');

这个查询用于显示派生表：

select
    *
    , row_number() over(partition by ProductID,Store           order by Trans_Date) rn1
    , row_number() over(partition by ProductID,Store,Cost_Amt  order by Trans_Date) rn2
    , row_number() over(partition by ProductID,Store           order by Trans_Date)
    - row_number() over(partition by ProductID,Store,Cost_Amt  order by Trans_Date) grp
from bigtable
order by ProductID,Store,Trans_Date
;

计算我们稍后需要的“grp”值：

|    | ProductID | Store |     Trans_Date      | Cost_Amt | rn1 | rn2 | grp |
|----|-----------|-------|---------------------|----------|-----|-----|-----|
|  1 |     20202 |  2320 | 02.01.2018 00:00:00 | $9.23    |   1 |   1 |   0 |
|  2 |     20202 |  2320 | 03.01.2018 00:00:00 | $9.23    |   2 |   2 |   0 |
|  3 |     20202 |  2320 | 04.01.2018 00:00:00 | $9.23    |   3 |   3 |   0 |
|  4 |     20202 |  2320 | 05.01.2018 00:00:00 | $9.38    |   4 |   1 |   3 |
|  5 |     20202 |  2320 | 06.01.2018 00:00:00 | $9.38    |   5 |   2 |   3 |
|  6 |     20202 |  2320 | 07.01.2018 00:00:00 | $9.38    |   6 |   3 |   3 |
|  7 |     20202 |  2320 | 08.01.2018 00:00:00 | $9.23    |   7 |   4 |   3 |
|  8 |     20202 |  2320 | 09.01.2018 00:00:00 | $9.23    |   8 |   5 |   3 |
|  9 |     20202 |  2320 | 10.01.2018 00:00:00 | $9.23    |   9 |   6 |   3 |
| 10 |     20202 |  2320 | 11.01.2018 00:00:00 | $9.38    |  10 |   4 |   6 |

现在计算日期范围：

select
      ProductID
    , Store
    , Cost_Amt
    , grp
    , min(Trans_Date) start_date
    , max(Trans_Date) end_date
from (
    select
        *
        , row_number() over(partition by ProductID,Store           order by Trans_Date)
        - row_number() over(partition by ProductID,Store,Cost_Amt  order by Trans_Date) grp
    from bigtable
    ) d
group by
      ProductID
    , Store
    , Cost_Amt
    , grp
;

结果如下：

|    | ProductID | Store | Cost_Amt | grp |  (No column name)   |  (No column name)   |
|----|-----------|-------|----------|-----|---------------------|---------------------|
|  1 |     20202 |  2320 | $9.23    |   0 | 02.01.2018 00:00:00 | 04.01.2018 00:00:00 |
|  2 |     20202 |  2320 | $9.23    |   3 | 08.01.2018 00:00:00 | 10.01.2018 00:00:00 |
|  3 |     20202 |  2320 | $9.38    |   3 | 05.01.2018 00:00:00 | 07.01.2018 00:00:00 |
|  4 |     20202 |  2320 | $9.38    |   6 | 11.01.2018 00:00:00 | 11.01.2018 00:00:00 |

另见：http://rextester.com/PJRU91378

Answer 3

这不是很优雅，但似乎是正确的：

select A.ProductID , A.Store , A.Trans_Date as Start_Date ,  
(select max(Trans_Date) from Table1 as C
           where A.ProductID = C.ProductID
           and   A.Store = C.Store
           and   A.Cost_Amt = C.Cost_Amt
           and   C.Trans_Date < (
             select coalesce(min(Trans_Date),'9999-01-01') 
             from Table1 as D
             where A.ProductID = D.ProductID
             and   A.Store = D.Store
             and   A.Cost_Amt <> D.Cost_Amt
             and   A.Trans_Date < D.Trans_Date
             )
           ) as End_Date  , A.Cost_Amt
from Table1 as A 
where A.Trans_Date = (select min(Trans_Date) 
             from Table1 as B
             where A.ProductID = B.ProductID
             and   A.Store = B.Store
             and   A.Cost_Amt = B.Cost_Amt
             and   B.Trans_Date > (
             select coalesce(max(Trans_Date),'0001-01-01') 
             from Table1 as E
             where A.ProductID = E.ProductID
             and   A.Store = E.Store
             and   A.Cost_Amt <> E.Cost_Amt
             and   A.Trans_Date > E.Trans_Date
             )
             )

现场示例在这里：http://sqlfiddle.com/#!15/7f8c4/12