Adding labels to row based on condition of prior/next row - sql

Question

I have sample data like the following in Snowflake. I'd like to assign groupings (without aggregation based on the grp_start -> grp_end (basically when one grp_start = 1 I want to assign it a label, and assign each sequential row the same ID until grp_end is equal to 1. That would constitute a single grp. Then the next grp should have a different label and follow the same logic.
Note: If a single row is a grp_start = 1 and grp_end = 1 I want it to have a single grp label for that row as well, and thus following the pattern.
The data needs to be partitioned by id and ordered by start_time as well. Please see the below sample data and the mockup of what the desired result is to be. Ideally, I need this to scale to large amounts of data.
Current data:
create or replace temporary table grp_test (id char(4), start_time date, grp_start int, grp_end int)
as select * from values
('0001','2021-01-10',1,0),
('0001','2021-01-11',0,0),
('0001','2021-01-14',0,1),
('0001','2021-07-01',1,1),
('0001','2021-09-25',1,0),
('0001','2021-09-29',0,1),
('0002','2022-11-04',1,0),
('0002','2022-11-25',0,1);
select * from grp_test;
Desired result mockup:
create or replace temporary table desired_result (id char(4), start_time date, grp_start int, grp_end int, label int)
as select * from values
('0001','2021-01-10',1,0,0),
('0001','2021-01-11',0,0,0),
('0001','2021-01-14',0,1,0),
('0001','2021-07-01',1,1,1),
('0001','2021-09-25',1,0,2),
('0001','2021-09-29',0,1,2),
('0002','2022-11-04',1,0,0),
('0002','2022-11-25',0,1,0);
select * from desired_result;

Answer 1

so changing the setup data to:
create or replace temporary table grp_test as
select * from values
('0001','2021-01-10'::date,1,0),
('0001','2021-01-11'::date,0,0),
('0001','2021-01-14'::date,0,1),
('0001','2021-01-15'::date,0,0),
('0001','2021-07-01'::date,1,1),
('0001','2021-09-25'::date,1,0),
('0001','2021-09-29'::date,0,1),
('0002','2022-11-04'::date,1,0),
('0002','2022-11-25'::date,0,1)
t(id, start_time, grp_start, grp_end);
We can use two CONDITIONAL_TRUE_EVENT's, this allows us to know we we are outside the end, but before a start, and thus can alter the label to null.
select d.*
,CONDITIONAL_TRUE_EVENT(grp_start=1) over (partition by id order by start_time) as s_e
,CONDITIONAL_TRUE_EVENT(grp_end=1) over (partition by id order by start_time) as e_e
,iff(s_e != e_e OR grp_end = 1, s_e, null) as label
from grp_test as d
order by 1,2;
ID
START_TIME
GRP_START
GRP_END
S_E
E_E
LABEL
0001
2021-01-10
1
0
1
0
1
0001
2021-01-11
0
0
1
0
1
0001
2021-01-14
0
1
1
1
1
0001
2021-01-15
0
0
1
1
0001
2021-07-01
1
1
2
2
2
0001
2021-09-25
1
0
3
2
3
0001
2021-09-29
0
1
3
3
3
0002
2022-11-04
1
0
1
0
1
0002
2022-11-25
0
1
1
1
1
If you don't actually care about label rows after an end, but before the next start, you can just use a single CONDITIONAL_TRUE_EVENT
select d.*
,CONDITIONAL_TRUE_EVENT(grp_start=1) over (partition by id order by start_time) as label
from grp_test as d
order by 1,2;
ID
START_TIME
GRP_START
GRP_END
LABEL
0001
2021-01-10
1
0
1
0001
2021-01-11
0
0
1
0001
2021-01-14
0
1
1
0001
2021-01-15
0
0
1
0001
2021-07-01
1
1
2
0001
2021-09-25
1
0
3
0001
2021-09-29
0
1
3
0002
2022-11-04
1
0
1
0002
2022-11-25
0
1
1

Answer 2

Here's a solution that uses two nested window functions, max and dense_rank. Snowflake (as well as most other DBMSs) doesn't allow you to nest two window functions, so we'll process the first one in a subquery and the second one in the query itself.
The key to this method is to assign a common date-value to all members of the group, in this case the start date of the group, then dense_rank will give a 1 to all the records tied for first place, a 2 to the next group, etc. So we want the max(Start_Time) of the records with grp_start=1 at or before this time for every row in grp_test.
max(Case When grp_start=1 Then Start_Time End)
Over (Partition By ID Order By Start_Time
Rows Between Unbounded Preceding And Current Row) as grp_start_time
So put it all together with
Select ID, Start_Time, Grp_Start, Grp_End,
dense_rank(grp_start_time) Over (Partition By ID) as label
From (
Select ID, Start_Time, Grp_Start, Grp_End,
max(Case When grp_start=1 Then Start_Time End)
Over (Partition By ID Order By Start_Time
Rows Between Unbounded Preceding And Current Row) as grp_start_time
From grp_test
)
Order by ID,Start_Time
METHOD 2
You can simplify this considerably if you are certain grp_start must only contain zeros and ones. This one simply creates a running sum of grp_start:
Select ID, Start_Time, Grp_Start, Grp_End,
sum(Grp_Start) Over (Partition By ID Order By Start_Time
Rows Between Unbounded Preceding And Current Row) as label
Order by ID,Start_Time