Hive 如果标记<&燃气轮机；如果配置单元中的标记=0，则重置为0？

如果tag=0，如何获取运行总和并将运行总和重置为零？就像上面的样本一样。TIA

您需要做的是为1和0的每个部分创建“组”。您可以通过创建一个布尔标志，然后对该列进行累加求和来获得组。在此基础上，您可以按在子查询中创建的每个组对原始

标记

列进行累积求和

查询：

customer    txn_date    tag running_sum
A           1-Jan-17    1   1
A           2-Jan-17    1   2
A           3-Jan-17    1   3
A           4-Jan-17    1   4
A           5-Jan-17    1   5
A           6-Jan-17    1   6
A           7-Jan-17    0   0
A           8-Jan-17    1   1
A           9-Jan-17    1   2
A           10-Jan-17   1   3
A           11-Jan-17   0   0
A           12-Jan-17   0   0
A           13-Jan-17   1   1
A           14-Jan-17   1   2
A           15-Jan-17   0   0

SELECT customer
  , txn_date
  , tag
  , SUM(tag) OVER (PARTITION BY customer, flg_sum ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS running_sum
FROM (
  SELECT *
    , SUM(tag_flg) OVER (PARTITION BY customer ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS flg_sum
  FROM (
    SELECT *
      , CASE WHEN  tag = 1 THEN 0 ELSE 1 END AS tag_flg
    FROM database.table ) x ) y

输出：

customer    txn_date    tag running_sum
A           1-Jan-17    1   1
A           2-Jan-17    1   2
A           3-Jan-17    1   3
A           4-Jan-17    1   4
A           5-Jan-17    1   5
A           6-Jan-17    1   6
A           7-Jan-17    0   0
A           8-Jan-17    1   1
A           9-Jan-17    1   2
A           10-Jan-17   1   3
A           11-Jan-17   0   0
A           12-Jan-17   0   0
A           13-Jan-17   1   1
A           14-Jan-17   1   2
A           15-Jan-17   0   0

SELECT customer
  , txn_date
  , tag
  , SUM(tag) OVER (PARTITION BY customer, flg_sum ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS running_sum
FROM (
  SELECT *
    , SUM(tag_flg) OVER (PARTITION BY customer ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS flg_sum
  FROM (
    SELECT *
      , CASE WHEN  tag = 1 THEN 0 ELSE 1 END AS tag_flg
    FROM database.table ) x ) y

---

您正在尝试编写SQL查询吗？如果是，您已经尝试过什么？看《谢谢你》@gobrewers14！成功了！实际上，我正在探索在这个问题上使用滞后和超前函数，但没有用。