How to fill in rows based on available data - sql

Question

Using Snowflake SQL.
So my table has 2 columns: hour and customerID. Every customer will have 2 rows, one corresponding to hour that he/she came into the store, and one corresponding to hour that he/she left the store. With this data, I want to create a table that has every hour that a customer has been in the store. For example, a customer X entered the store at 1PM and left at 5PM, so there would be 5 rows (1 for each hour) like the screenshot below.
Here's my attempt that's now:
select
hour
,first_value(customer_id) over (partition by customer_id order by hour rows between unbounded preceding and current row) as customer_id
FROM table

Answer 1

In Snowflake, you would typically use a table of numbers to solve this. You can use the table (generator ...) syntax to generate such derived table, and then join it with an aggregate query that computes the hour boundaries of each client with an inequality condition:
select t.customer_id, dateadd(hour, n.rn, t.min_hour) final_hour
from (
select t.customer_id, min(t.hour) min_hour, max(t.hour) max_hour
from mytable t
group by t.customer_id
) t
inner join (
select row_number() over(order by null) - 1 rn
from table (generator(rowcount => 24))
) n on dateadd(hour, n.rn, t.min_hour) <= t.max_hour
order by customer_id, final_hour
This would handle up to 24 hours of visit per customer. If you need more, then you can increase the parameter to the table generator.

Answer 2

so for the example case as shown in the test data, where there is only one days worth of data GMB's solution works fine.
once you get into many days (that can/cannot have overlapping store visits, lets just pretend you cannot overnight in the store)
which can be fixed via:
select t.hour::date, t.customer_id, min(t.hour) min_hour, max(t.hour) max_hour
from mytable t
group by 1,2
but multiple entries, ether requires tag data like:
with mytable as (
select * from values
('2019-04-01 09:00:00','x','in')
,('2019-04-01 15:00:00','x','out')
,('2019-04-02 12:00:00','x','in')
,('2019-04-02 14:00:00','x','out')
v(hour, customer_id, state)
)
or for it to be inferred:
with mytable as (
select * from values ('2019-04-01 09:00:00','x','in'),('2019-04-01 15:00:00','x','out')
,('2019-04-02 12:00:00','x','in'),('2019-04-02 14:00:00','x','out')
v(hour, customer_id, state)
)
select hour::date as day
,hour
,customer_id
,state
,BITAND(row_number() over(partition by day, customer_id order by hour), 1) = 1 AS in_dir
from mytable
order by 3,1,2;
giving:
DAY HOUR CUSTOMER_ID STATE IN_DIR
2019-04-01 2019-04-01 09:00:00 x in TRUE
2019-04-01 2019-04-01 15:00:00 x out FALSE
2019-04-02 2019-04-02 12:00:00 x in TRUE
2019-04-02 2019-04-02 14:00:00 x out FALSE
now this can be used with a LAG and QUALIFY to get true ranges the can handle multi-entries:
select customer_id
,day
,hour
,lead(hour) over (partition by customer_id, day order by hour) as exit_time
from infer_direction
qualify in_dir = true
which works by getting then next time for all rows of each day/customer, and after that (via the qualify) only keeping the rows 'in' rows.
then we can then join to the time of a day:
select dateadd('hour', row_number() over(order by null) - 1, '00:00:00'::time) as hour
from table (generator(rowcount => 24))
thus for it all woven together
with mytable as (
select hour::timestamp as hour, customer_id, state
from values
('2019-04-01 09:00:00','x','in')
,('2019-04-01 12:00:00','x','out')
,('2019-04-02 13:00:00','x','in')
,('2019-04-02 14:00:00','x','out')
,('2019-04-02 9:00:00','x','in')
,('2019-04-02 10:00:00','x','out')
v(hour, customer_id, state)
), infer_direction AS (
select hour::date as day
,hour::time as hour
,customer_id
,state
,BITAND(row_number() over(partition by day, customer_id order by hour), 1) = 1 AS in_dir
from mytable
), visit_ranges as (
select customer_id
,day
,hour
,lead(hour) over (partition by customer_id, day order by hour) as exit_time
from infer_direction
qualify in_dir = true
), time_of_day AS (
select dateadd('hour', row_number() over(order by null) - 1, '00:00:00'::time) as hour
from table (generator(rowcount => 24))
)
select t.customer_id
,t.day
,h.hour
from visit_ranges as t
join time_of_day h on h.hour between t.hour and t.exit_time
order by 1,2,3;
we get:
CUSTOMER_ID DAY HOUR
x 2019-04-01 09:00:00
x 2019-04-01 10:00:00
x 2019-04-01 11:00:00
x 2019-04-01 12:00:00
x 2019-04-02 09:00:00
x 2019-04-02 10:00:00
x 2019-04-02 13:00:00
x 2019-04-02 14:00:00