Snowflake MATCH_RECOGNIZE 跳过不重要的事件

Question

我有以下按发生时间排序的事件：

e4 -> e2 -> e2 -> e3 -> e10 -> e4

如果 e2 事件发生然后 e4 发生（e2 在 e4 之前），我应该如何编写 MATCH_RECOGNIZE 的 PATTERN 部分来匹配记录，无论这两个事件之间是否有 0 个或更多其他事件？

e4 -> e2 -> e2 -> e3 -> e10 -> e4 - matched
e4 -> e2 -> e4 - matched
e4 -> e4 -> e2 -> e3 - not matched
e2 -> e10 -> e2 -> e5 -> e4 - matched

Answer 1

所以这四个序列，它们可以最小匹配：

WITH data AS (
    SELECT * FROM VALUES 
     (1,'e4',1),(1,'e2',2),(1,'e2',3),(1,'e3',4),(1,'e10',5),(1,'e4',6),
     (2,'e4',1),(2,'e2',2),(2,'e4',3),
     (3,'e4',1),(3,'e4',2),(3,'e2',3),(3,'e3',4),
     (4,'e2',1),(4,'10',2),(4,'e2',3),(4,'e5',4),(4,'e4',5)
)
SELECT * FROM data 
match_recognize(
    partition by column1
    order by column3
    measures
        match_number() as "MATCH_NUMBER",
        match_sequence_number() as msq,
        classifier() as cl
    all rows per match with unmatched rows
    PATTERN (d1 d2* d3)
    DEFINE d1 as column2 = 'e2',
        d2 as column2 NOT IN ('e2','e4'),
        d3 as column2 = 'e4'
)
ORDER BY 1,3;

给予：

COLUMN1	COLUMN2	COLUMN3	MATCH_NUMBER	MSQ	CL
1	e4	1
1	e2	2
1	e2	3	1	1	D1
1	e3	4	1	2	D2
1	e10	5	1	3	D2
1	e4	6	1	4	D3
2	e4	1
2	e2	2	1	1	D1
2	e4	3	1	2	D3
3	e4	1
3	e4	2
3	e2	3
3	e3	4
4	e2	1
4	10	2
4	e2	3	1	1	D1
4	e5	4	1	2	D2
4	e4	5	1	3	D3

但是如果你说你想要“匹配”，那么也许你只想要范围细节，因此：

SELECT * FROM data 
match_recognize(
    partition by column1
    order by column3
    measures
        first_value(column1) as batch,
        first_value(column3) as seq_start,
        last_value(column3) as seq_end,
        match_number() as "MATCH_NUMBER",
        match_sequence_number() as msq,
        classifier() as cl
    one row per match
    PATTERN (d1 d2* d3)
    DEFINE d1 as column2 = 'e2',
        d2 as column2 NOT IN ('e2','e4'),
        d3 as column2 = 'e4'
)
ORDER BY 1,3;

可能是你想要的：

COLUMN1	BATCH	SEQ_START	SEQ_END	MATCH_NUMBER	MSQ	CL
1	1	3	6	1	4	D3
2	2	2	3	1	2	D3
4	4	3	5	1	3	D3