在 Trino/Presto SQL 中：创建一个新列来说明（枚举）每个组的顺序变化

Question

我想计算一个新列，该列根据每组和该组内的序列变化分配 ID。

考虑下表：

# | user_id | my_date    | color  |
# |---------|------------|--------|
# | a       | 2023-02-01 | red    | ----
# | a       | 2023-03-22 | red    |    |
# | a       | 2023-03-30 | red    |    | this is period *1* for user_id = a
# | a       | 2023-06-10 | red    |    |
# | a       | 2023-06-11 | red    | ----
# | a       | 2023-07-03 | green  |
# | a       | 2023-07-09 | green  |
# | a       | 2024-01-11 | green  |
# | a       | 2024-02-11 | yellow |
# | a       | 2024-02-12 | yellow |
# | a       | 2024-02-13 | yellow |
# | a       | 2024-02-14 | yellow |
# | b       | 2022-10-20 | blue   |
# | b       | 2022-10-21 | blue   |
# | b       | 2022-10-22 | blue   |
# | b       | 2022-10-23 | brown  | ----
# | b       | 2022-10-24 | brown  |    | this is period *2* for user_id = b
# | b       | 2022-10-25 | brown  | ----
# | b       | 2022-10-26 | blue   |
# | b       | 2022-10-27 | blue   |

对于每个

user_id

，如果我们按顺序（升序）沿着

my_date

，我们可以看到存在具有相同

color

特征的“周期”或序列。

我想创建一个新列来说明周期/序列。

预期产量

# | user_id | my_date    | color  | period_number |
# |---------|------------|--------|---------------|
# | a       | 2023-02-01 | red    | 1             |
# | a       | 2023-03-22 | red    | 1             |
# | a       | 2023-03-30 | red    | 1             |
# | a       | 2023-06-10 | red    | 1             |
# | a       | 2023-06-11 | red    | 1             |
# | a       | 2023-07-03 | green  | 2             |
# | a       | 2023-07-09 | green  | 2             |
# | a       | 2024-01-11 | green  | 2             |
# | a       | 2024-02-11 | yellow | 3             |
# | a       | 2024-02-12 | yellow | 3             |
# | a       | 2024-02-13 | yellow | 3             |
# | a       | 2024-02-14 | yellow | 3             |
# | b       | 2022-10-20 | blue   | 1             |
# | b       | 2022-10-21 | blue   | 1             |
# | b       | 2022-10-22 | blue   | 1             |
# | b       | 2022-10-23 | brown  | 2             |
# | b       | 2022-10-24 | brown  | 2             |
# | b       | 2022-10-25 | brown  | 2             |
# | b       | 2022-10-26 | blue   | 3             |
# | b       | 2022-10-27 | blue   | 3             |

SQL 方言

我使用基于 Trino SQL 的 AWS Athena。

可重复的数据

WITH my_table AS (
    SELECT *
    FROM (VALUES
        ('a', DATE '2023-02-01', 'red'),
        ('a', DATE '2023-03-22', 'red'),
        ('a', DATE '2023-03-30', 'red'),
        ('a', DATE '2023-06-10', 'red'),
        ('a', DATE '2023-06-11', 'red'),
        ('a', DATE '2023-07-03', 'green'),
        ('a', DATE '2023-07-09', 'green'),
        ('a', DATE '2024-01-11', 'green'),
        ('a', DATE '2024-02-11', 'yellow'),
        ('a', DATE '2024-02-12', 'yellow'),
        ('a', DATE '2024-02-13', 'yellow'),
        ('a', DATE '2024-02-14', 'yellow'),
        ('b', DATE '2022-10-20', 'blue'),
        ('b', DATE '2022-10-21', 'blue'),
        ('b', DATE '2022-10-22', 'blue'),
        ('b', DATE '2022-10-23', 'brown'),
        ('b', DATE '2022-10-24', 'brown'),
        ('b', DATE '2022-10-25', 'brown'),
        ('b', DATE '2022-10-26', 'blue'),
        ('b', DATE '2022-10-27', 'blue')
    ) AS t(user_id, my_date, color)
)

SELECT *
FROM my_table; 

Answer 1

参见示例

with my_table AS (
    SELECT *
    FROM (VALUES
        ('a',  '2023-02-01', 'red'),
        ('a',  '2023-03-30', 'red'),
        ('a',  '2023-06-10', 'red'),
        ('a',  '2023-06-11', 'red'),
        ('a',  '2023-07-03', 'green'),
        ('a',  '2023-07-09', 'green'),
        ('a',  '2024-01-11', 'green'),
        ('a',  '2024-02-11', 'yellow'),
        ('a',  '2024-02-12', 'yellow'),
        ('a',  '2024-02-13', 'yellow'),
        ('a',  '2024-02-14', 'yellow'),
        ('b',  '2022-10-20', 'blue'),
        ('b',  '2022-10-21', 'blue'),
        ('b',  '2022-10-22', 'blue'),
        ('b',  '2022-10-23', 'brown'),
        ('b',  '2022-10-24', 'brown'),
        ('b',  '2022-10-25', 'brown'),
        ('b',  '2022-10-26', 'blue'),
        ('b',  '2022-10-27', 'blue')
    ) AS t(user_id, my_date, color)
)

查询

,periods as(select user_id,color,my_date
  ,(select min(my_date) from my_table t2 
    where t2.user_id=t.user_id and t2.my_date>t.my_date 
     and t2.color<>t.color) next_date
from my_table t
)
,gr as(
select user_id,color,next_date
  ,min(my_date)mindt 
from periods
group by user_id,color,next_date
)
,res as(
select t.* 
from my_table t
left join gr g on g.user_id=t.user_id
  and g.mindt<=t.my_date
)
select user_id,color,my_date,count(*) period_number
from res
group by user_id,color,my_date
order by user_id,my_date;