BigQuery：检索两列中唯一的行，否则检索第三列最大的行

Question

我有一个BigQuery表my_table ，如下所示：

+---------+---------+-------+------------------+----------+--------+-----+--------+
| poll_id | user_id | count |    timestamp     | timezone | answer | age | gender |
+---------+---------+-------+------------------+----------+--------+-----+--------+
|       1 |       1 |     5 | 2019-08-06 11:00 |        1 | no     |  25 | male   |
|       1 |       1 |    10 | 2019-08-06 10:00 |        1 | no     |  25 | male   |
|       1 |       1 |    10 | 2019-08-06 10:30 |        1 | yes    |  25 | male   |
|       1 |       2 |    10 | 2019-08-06 11:00 |        1 | no     |  35 | male   |
|       1 |       2 |    20 | 2019-08-06 11:00 |        1 | no     |  35 | male   |
|       1 |       2 |    35 | 2019-08-06 11:00 |        1 | NULL   |  35 | male   |
|       2 |       1 |    10 | 2019-08-06 10:35 |        1 | no     |  25 | male   |
|       3 |       1 |    10 | 2019-08-06 10:35 |        1 | NULL   |  25 | male   |
+---------+---------+-------+------------------+----------+--------+-----+--------+

我想检索满足以下要求的行：

• 如果该行具有 poll_id和user_id的唯一组合，则在answer具有非NULL值的情况下包括该行
• 如果该行没有 poll_id和user_id的唯一组合：
  • 在answer列中包含count最大但非NULL的行
    • 如果有两行具有相同的count （且answer为非NULL），则包括timestamp最大的行

我还希望能够将搜索限制为特定的日期和时区，例如日期为2019-08-06且时区为1，并且我不想在user_id检索值为NULL的行。

到目前为止，我已经尝试了以下标准SQL语句：

  SELECT
    t1.poll_id,
    t1.user_id,
    t1.count,
    t1.timestamp,
    t1.timezone,
    t1.answer,
    t1.age,
    t1.gender,
  FROM
    `my_table` t1
  LEFT JOIN
    `my_table` t2
  ON
    t1.poll_id = t2.poll_id
    AND t1.user_id = t2.user_id
    AND t1.count < t2.count
    AND t2.answer IS NOT NULL
    AND DATE(t2.timestamp, "+1:00") = "2019-08-06"
  WHERE
    t1.user_id IS NOT NULL
    AND t1.answer IS NOT NULL
    AND DATE(t1.timestamp, "+1:00") = "2019-08-06"
    AND t1.timezone = 1   
    AND t2.count IS NULL

所示表的预期结果是：

+---------+---------+-------+------------------+----------+--------+-----+--------+
| poll_id | user_id | count |    timestamp     | timezone | answer | age | gender |
+---------+---------+-------+------------------+----------+--------+-----+--------+
|       1 |       1 |    10 | 2019-08-06 10:30 |        1 | yes    |  25 | male   | // count = 10 and largest timestamp
|       1 |       2 |    20 | 2019-08-06 11:00 |        1 | no     |  35 | male   | // count = 20 (the 35 row had NULL in 'answer')
|       2 |       1 |    10 | 2019-08-06 10:35 |        1 | no     |  25 | male   | // unique 'poll_id', 'user_id' combination
+---------+---------+-------+------------------+----------+--------+-----+--------+

但是，似乎有两个问题：

1. 如果有多个具有相同（最大） count数值的行，则将检索所有这些行。 这意味着在此示例中将同时检索第2行和第3行。
2. 如果poll_id和user_id组合恰好有两行，即使它们具有不同的count数值，也不会检索到这两行。

至少看起来是这样。 我很难跟踪问题，当然，还要找出正确的查询。

任何帮助，将不胜感激。

Answer 1

对于这种类型的查询，通常使用row_number() 。 我认为这符合您的描述：

select t.*
from (select t.*,
             row_number() over (partition by poll_id, user_id order by count desc, timestamp desc) as seqnum
      from my_table t
      where answer is not nll
     ) t
where seqnum = 1;

Answer 2

以下是BigQuery标准SQL

#standardSQL
SELECT * EXCEPT(pos) 
FROM (
  SELECT *, 
    ROW_NUMBER() OVER(PARTITION BY poll_id, user_id ORDER BY count DESC, timestamp DESC) AS pos
  FROM `project.dataset.table`
  WHERE NOT answer IS NULL
  AND NOT user_id IS NULL
  AND timezone = 1
  AND SUBSTR(timestamp, 1, 10) = '2019-08-06'
)
WHERE pos = 1   

如果适用于您问题中的样本数据-结果为

Row poll_id user_id count   timestamp           timezone    answer  age gender   
1   1       1       10      2019-08-06 10:30    1           yes     25  male     
2   1       2       20      2019-08-06 11:00    1           no      35  male     
3   2       1       10      2019-08-06 10:35    1           no      25  male