在 SUM over PARTITION 之后使用 LAG

Question

+---------+----------+---------+------------+-------+---------+---------------+
| Product | Version  | Country |    Week    | sales | returns | Running_sales |
+---------+----------+---------+------------+-------+---------+---------------+
| Pdt1    | pdt1ver1 | Aus     | 2020M01W01 |    10 |       3 |             7 |
| pdt1    | pdt1ver1 | Fra     | 2020M01W01 |     8 |       2 |             6 |
| pdt1    | pdt1ver1 | Fra     | 2020M01W02 |    15 |       5 |            16 |
| pdt1    | pdt1ver2 | UK      | 2020M01W01 |    20 |       5 |            15 |
| pdt1    | pdt1ver2 | UK      | 2020M01W02 |    15 |       1 |            29 |
| pdt1    | pdt1ver2 | UK      | 2020M01w03 |     9 |       0 |            38 |
| pdt2    | pdt2ver1 | Fra     | 2020M01W01 |     5 |       1 |             4 |
| pdt2    | pdt2ver1 | Fra     | 2020M01W02 |     3 |       0 |             7 |
+---------+----------+---------+------------+-------+---------+---------------+

我有一个视图给我之前显示的数据和运行总数。这是一个简单的 select，在 select 语句中也有这个（已经从这里得到帮助）：

Sum(sales-returns) over(
  partition by Version, Country 
  Order by Week 
  ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
) Running_sales

我现在需要另一列来给我上周的 returns（但留在它应该在的子组中，这意味着我不能在第 1、2 行中包含任何内容, 4 和 7 因为没有前一行), 因此:

+---------+----------+---------+------------+-------+---------+---------------+------------------+
| Product | Version  | Country |    Week    | sales | returns | Running_sales | previous_returns |
+---------+----------+---------+------------+-------+---------+---------------+------------------+
| Pdt1    | pdt1ver1 | Aus     | 2020M01W01 |    10 |       3 |             7 |                  |
| pdt1    | pdt1ver1 | Fra     | 2020M01W01 |     8 |       2 |             6 |                  |
| pdt1    | pdt1ver1 | Fra     | 2020M01W02 |    15 |       5 |            16 |                2 |
| pdt1    | pdt1ver2 | UK      | 2020M01W01 |    20 |       5 |            15 |                  |
| pdt1    | pdt1ver2 | UK      | 2020M01W02 |    15 |       1 |            29 |                5 |
| pdt1    | pdt1ver2 | UK      | 2020M01w03 |     9 |       0 |            38 |                1 |
| pdt2    | pdt2ver1 | Fra     | 2020M01W01 |     5 |       1 |             4 |                  |
| pdt2    | pdt2ver1 | Fra     | 2020M01W02 |     3 |       0 |             7 |                1 |
+---------+----------+---------+------------+-------+---------+---------------+------------------+

我尝试将此添加到我的 select 声明中：

LAG(returns,1,0)
    OVER(partition by Version, Country 
      Order by Week 
      ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) previous_returns

但这显然不是那么容易。也许 window ？我需要再次使用分区吗？在这个问题上的任何提示将不胜感激。我正在使用 Teradata。

Answer 1

这应该很简单：

lag(returns) over(partition by version, country order by week) previous_returns

在 SUM over PARTITION 之后使用 LAG

Using LAG after a SUM over PARTITION

sql

teradata

window-functions