Weekly Streaks - Snowflake - snowflake-cloud-data-platform

Question

I have various questions about consecutive customer visits, it spans from a 90 days, to a monthly, quarterly, and currently the question is weekly. So I really am looking for easy to duplicate and edit code that can cover my bases.
I found some helpful articles to get me started, but am having some trouble deciphering the output, and is causing me some trouble with validating.
the code used is:
with output as
(select distinct
unqclient
,dategroup
,datediff('week',min(weekofclass),max(weekofclass)) as streak
,min(weekofclass) as startdate
,max(weekofclass) as enddate
from
(select
*
,dateadd('week',-rank,weekofclass) as dategroup
from
(select distinct
concat(v.location, v.clientid) as unqclient
,cast(v.classdate as date) as classdate2
,date_trunc('week',classdate2) as weekofclass
,dense_rank() over (partition by unqclient order by weekofclass) as rank
from visit_data as v
where (missed=false and cancelled=false)
)
)
group by 1,2)
select
output.*
,c.location
,c.emailname
,s.studioname
from output
left join clients as c
on concat(c.location,c.clientid)=output.unqclient
left join studios as s
on c.location=s.location
where streak>1
and date_trunc('week',enddate)>=date_trunc('week',current_date()-7)
order by streak desc
The days for the start and end dates are right (showing the monday of the week their first and last visit in the streak occurred). But I don't understand what 'dategroup' is outputting, they do not correspond to first visits of that client or take place during the streak.
Example of Output from code above:
Client
Date Group
Streak
Start Date
End Date
A
2020-03-02
116
2020-06-15
2022-09-05
B
2017-04-24
122
2020-05-04
2022-09-05

Answer 1

So if we start with some fake data with weekly streaks:
with visit_data(location, clientid, classdate, missed, cancelled ) as (
select * from values
-- (1, 10, '2022-06-15 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-06-22 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-06-29 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-06 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-20 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-27 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-03 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-10 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-17 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-24 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-31 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-7 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-14 10:12:13'::timestamp, false, false)
), clients(location, clientid, emailname) as (
select * from values
(1, 10, 'email_10#example.com')
), studios(location, studioname) as (
select * from values
(1, 'location_1')
)
we can move the two inner sub-select into one block and inspect it:
select distinct
v.location
,v.clientid
,date_trunc('week', v.classdate) as weekofclass
,dense_rank() over (partition by v.location, v.clientid order by weekofclass) as rank
,dateadd('week', -rank, weekofclass) as dategroup
from visit_data as v
where missed = false and cancelled = false
gives:
LOCATION
CLIENTID
WEEKOFCLASS
RANK
DATEGROUP
1
10
2022-08-15 00:00:00.000
1
2022-08-08 00:00:00.000
1
10
2022-08-22 00:00:00.000
2
2022-08-08 00:00:00.000
1
10
2022-08-29 00:00:00.000
3
2022-08-08 00:00:00.000
1
10
2022-09-05 00:00:00.000
4
2022-08-08 00:00:00.000
1
10
2022-09-12 00:00:00.000
5
2022-08-08 00:00:00.000
we can drop the timestamp -> date cast (if that is what is happening) as date trunc will reduce it down more, if it was a string inside visit_data it shouldn't be, can be inline case with:
,date_trunc('week', v.classdate::date) as weekofclass
next the date group can be calculated at the same time.
Datagroup is just a clustering group, it meaning is not really important, how it works is the dense_rank will give increasing numbers over the distinct truncated weeks, and minusing that rank from the week, if they are part of the same batch will end up with the same result
If we put some "gaps in the data"
with visit_data(location, clientid, classdate, missed, cancelled ) as (
select * from values
(1, 10, '2022-06-15 10:12:13'::timestamp, false, false),
(1, 10, '2022-06-22 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-06-29 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-06 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-20 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-27 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-03 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-10 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-17 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-24 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-31 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-7 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-14 10:12:13'::timestamp, false, false)
)
we get:
LOCATION
CLIENTID
WEEKOFCLASS
RANK
DATEGROUP
1
10
2022-06-13 00:00:00.000
1
2022-06-06 00:00:00.000
1
10
2022-06-20 00:00:00.000
2
2022-06-06 00:00:00.000
1
10
2022-07-04 00:00:00.000
3
2022-06-13 00:00:00.000
1
10
2022-07-11 00:00:00.000
4
2022-06-13 00:00:00.000
1
10
2022-07-18 00:00:00.000
5
2022-06-13 00:00:00.000
1
10
2022-08-15 00:00:00.000
6
2022-07-04 00:00:00.000
1
10
2022-08-22 00:00:00.000
7
2022-07-04 00:00:00.000
1
10
2022-08-29 00:00:00.000
8
2022-07-04 00:00:00.000
1
10
2022-09-05 00:00:00.000
9
2022-07-04 00:00:00.000
1
10
2022-09-12 00:00:00.000
10
2022-07-04 00:00:00.000
we see the rank steps in ones, but some week have a gap, and thus the minus gives a different number. This is called Gaps And Islands
After that, the GROUP BY and DISTINCT in the output is redunant, you should only need one or the other, not both
in the outer select, the date_trunc('week',enddate) is not needed as the date is already truncated to week, and functions should be avoid on WHERE clauses, and it can get pushed as a HAVING into the output CTE.
thus with the data CTE's included:
with visit_data(location, clientid, classdate, missed, cancelled ) as (
select * from values
(1, 10, '2022-06-15 10:12:13'::timestamp, false, false),
(1, 10, '2022-06-22 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-06-29 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-06 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-13 10:12:13'::timestamp, false, false),
(1, 10, '2022-07-20 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-07-27 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-03 10:12:13'::timestamp, false, false),
-- (1, 10, '2022-08-10 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-17 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-24 10:12:13'::timestamp, false, false),
(1, 10, '2022-08-31 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-7 10:12:13'::timestamp, false, false),
(1, 10, '2022-09-14 10:12:13'::timestamp, false, false)
), clients(location, clientid, emailname) as (
select * from values
(1, 10, 'email_10#example.com')
), studios(location, studioname) as (
select * from values
(1, 'location_1')
), inner_a as(
select distinct
v.location
,v.clientid
,date_trunc('week', v.classdate) as weekofclass
,dense_rank() over (partition by v.location, v.clientid order by weekofclass) as rank
,dateadd('week', -rank, weekofclass) as dategroup
from visit_data as v
where missed = false and cancelled = false
order by rank
)
--with
,output as (
select --distinct
location
,clientid
,dategroup
,min(weekofclass) as startdate
,max(weekofclass) as enddate
,datediff('week', startdate, enddate) as streak
from inner_a
group by 1,2,3
having enddate >= date_trunc('week',current_date()-7)
)
select
o.*
,c.location
,c.emailname
,s.studioname
from output as o
left join clients as c
on c.location = o.location
and c.clientid = o.clientid
left join studios as s
on c.location = s.location
where streak>1
order by streak desc
gives:
LOCATION
CLIENTID
DATEGROUP
STARTDATE
ENDDATE
STREAK
LOCATION
EMAILNAME
STUDIONAME
1
10
2022-07-04 00:00:00.000
2022-08-15 00:00:00.000
2022-09-12 00:00:00.000
4
1
email_10#example.com
location_1