Teradata 分区顺序事件

Question

Teradata - 下面是两节到达车站的轨道车。 773 清空 (RI/RE) 然后加载 (RI/RL) 然后离开 TD。 819 只是清空然后离开。我想在 table 下面创建结果。我试过在汽车上使用 case 语句分组，但同一辆车多次经过车站，所以 max 和 min 给了我 unpredictable 结果。我已经阅读了有关分区的内容，但我很难想象这一点。任何帮助表示赞赏。

*注意：最好将初始搜索限制在 EVT_CD='TA' AND STN=DEST 以消除任何仅通过车站的汽车。但我不能将整个记录集限制为那个，因为 TD 有不同的目的地。

CAR_NUMB    EVT_DT  EVT_TM  EVT_CD  EVST_CD WB_ID   STN     DEST
773 03/08/2016  19.05.00    TA          582016  BOSTON      BOSTON  
773 03/12/2016  04.04.00    AP  PU      582016  BOSTON      BOSTON  
773 03/12/2016  14.35.00    RI  RE      412016  BOSTON      BOSTON  
773 03/12/2016  14.37.00    AP  PL      412016  BOSTON      BOSTON  
773 03/12/2016  14.45.00    RI  RL      812016  BOSTON      HOUSTON 
773 03/14/2016  12.22.00    TD          812016  BOSTON      HOUSTON 
819 03/04/2016  17.50.00    TA          362016  STLOUIS     STLOUIS  
819 03/06/2016  13.50.00    AP  PU      362016  STLOUIS     STLOUIS  
819 03/06/2016  17.27.55    RI  RE      042016  STLOUIS     STLOUIS  
819 03/07/2016  00.37.00    RI  PR      042016  STLOUIS     PORTLAND
819 03/11/2016  01.47.00    TD          042016  STLOUIS     PORTLAND     

Desired output: 
CAR_NUMB    TA              AP          RIRE        RIRL        TD
773 03/08/2016 19.05.00 03/12/20..  03/12/20..  03/12/20..  03/14/2016  12.22.00
819 03/04/2016 17.50.00 03/06/20..  03/06/20..      null    03/11/2016 01.47.00

上面的[..]。我为了格式化把时间戳截掉了。

Answer 1

您可以使用 CASE 使用窗口聚合函数在接下来的 5 行中搜索给定 EVT_CD 的最小值 date/time。

我将日期和时间合并到时间戳中，因为它更易于使用：

SELECT tab.*
  ,CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND) AS TA
  ,MIN(CASE WHEN EVT_CD = 'AP'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) 
   OVER (PARTITION BY CAR_NUMB
         ORDER BY EVT_DT, EVT_TM
         ROWS BETWEEN CURRENT ROW AND 5 FOLLOWING) AS AP
  ,MIN(CASE WHEN EVT_CD = 'RI' AND EVST_CD = 'RE'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) 
   OVER (PARTITION BY CAR_NUMB 
         ORDER BY EVT_DT, EVT_TM
         ROWS BETWEEN CURRENT ROW AND 5 FOLLOWING) AS RIRE
  ,MIN(CASE WHEN EVT_CD = 'RI' AND EVST_CD = 'RL'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) 
   OVER (PARTITION BY CAR_NUMB 
         ORDER BY EVT_DT, EVT_TM
         ROWS BETWEEN CURRENT ROW AND 5 FOLLOWING) AS RIRL
  ,MIN(CASE WHEN EVT_CD = 'TD'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) 
   OVER (PARTITION BY CAR_NUMB 
         ORDER BY EVT_DT, EVT_TM
         ROWS BETWEEN CURRENT ROW AND 5 FOLLOWING) AS TD
FROM tab
QUALIFY -- finally return only the starting row
   EVT_CD = 'TA'

如果 TA 和 TD 之间的行数可以大于 5，则必须相应地调整 ROWS。如果数字变化很大并且缺少 EVT_CD 此方法可能会失败并报告下一次旅行的数据。这个问题可以通过添加另一个步骤来解决：

SELECT 
   CAR_NUMB
  ,TA
  ,CASE WHEN   AP < TD THEN AP END AS AP
  ,CASE WHEN RIRE < TD THEN RIRE END AS RIRE
  ,CASE WHEN RIRL < TD THEN RIRL END AS RIRL
  ,TD
FROM
 (
   previous query
 ) AS dt

如果 TD 可能丢失，您可以采用不同的方法：找到之前的 TA 时间戳并按其分组：

SELECT 
   CAR_NUMB
  ,TA
  ,MIN(CASE WHEN EVT_CD = 'AP'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) AS AP
  ,MIN(CASE WHEN EVT_CD = 'RI' AND EVST_CD = 'RE'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) AS RIRE
  ,MIN(CASE WHEN EVT_CD = 'RI' AND EVST_CD = 'RL'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) AS RIRL
  ,MIN(CASE WHEN EVT_CD = 'TD'
            THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
       END) AS TD
FROM
 (
   SELECT tab.*
     ,MAX(CASE WHEN EVT_CD = 'TA'
               THEN CAST(EVT_DT AS TIMESTAMP(0)) + (EVT_TM - TIME '00:00:00' HOUR TO SECOND)
          END) 
      OVER (PARTITION BY CAR_NUMB
            ORDER BY EVT_DT, EVT_TM
            ROWS UNBOUNDED PRECEDING) AS TA
   FROM tab
   -- maybe QUALIFY TA IS NOT NULL?
 ) AS dt
GROUP BY
   CAR_NUMB
  ,TA

Teradata 分区顺序事件

Teradata partitioning sequential events

teradata

partition