Aggregate only Sequential values - sql

Question

I have a table with 3 columns
Create table test
(
Created Datetime
, Flag Bit
, Amount Money
)
that looks like this
Created Flag Amount
2019-12-01 00:00:00.000 1 50,40
2019-11-21 00:00:00.000 1 50,40
2019-11-06 00:00:00.000 0 50,40
2019-10-04 00:00:00.000 1 50,40
2019-09-08 00:00:00.000 1 50,40
2019-09-01 00:00:00.000 0 50,40
2019-08-04 00:00:00.000 1 50,40
2019-07-24 00:00:00.000 1 50,40
2019-07-23 00:00:00.000 1 50,40
2019-06-01 00:00:00.000 0 50,40
2019-05-05 00:00:00.000 0 50,40
2019-04-25 00:00:00.000 1 50,40
2019-03-11 00:00:00.000 0 50,40
2019-02-03 00:00:00.000 0 50,40
2019-02-02 00:00:00.000 0 50,40
2019-02-01 00:00:00.000 0 50,40
2019-01-31 00:00:00.000 1 50,40
2019-01-26 00:00:00.000 0 50,40
2019-01-26 00:00:00.000 0 50,40
2019-01-01 00:00:00.000 1 50,40
As you can see it is ordered by Created in descending order.
Imagine that all these rows are Transactions. When the flag is one we have a checkpoint. So for example from line 20 to 17 is one period(Always counting from older to newer). From line 17 to 12 is another period and so on.
Please notice that in lines 9,8 and 7 we have 3 consecutive flags with a value of 1. When this happens, having consecutive 1s without 0s, i want to treat all the consecutive 1s as a group. I want them to appear as one row with Summed amount and keeping the MIN(Created) of them.
For example for rows 9-7 I want to group it in one row with where amount has a value of 151.2,flag a value of 1 and Created a value of 2019-07-23 00:00:00.000 (the min(date) of the three rows.
An example output of this table would be the following.
Created Flag Amount
2019-11-21 00:00:00.000 1 100,80
2019-11-06 00:00:00.000 0 50,40
2019-09-08 00:00:00.000 1 100,80
2019-09-01 00:00:00.000 0 50,40
2019-07-23 00:00:00.000 1 151,20
2019-06-01 00:00:00.000 0 50,40
2019-05-05 00:00:00.000 0 50,40
2019-04-25 00:00:00.000 1 50,40
2019-03-11 00:00:00.000 0 50,40
2019-02-03 00:00:00.000 0 50,40
2019-02-02 00:00:00.000 0 50,40
2019-02-01 00:00:00.000 0 50,40
2019-01-31 00:00:00.000 1 50,40
2019-01-26 00:00:00.000 0 50,40
2019-01-26 00:00:00.000 0 50,40
2019-01-01 00:00:00.000 1 50,40

Answer 1

If you just want to collapse adjacent "1"s, then one approach is to assign a grouping based on the count of preceding 0s and aggregate. So for aggregating the "1"s:
select min(created), 1 as flag, sum(amount)
from (select t.*,
sum(1 - flag) over (order by created) as grouping
from t
) t
where flag = 1
group by grouping;
This does not quite work when we include 0s, because the 0s would get combined with the 1s. So I think the simplest method is union all:
select min(created), 1 as flag, sum(amount)
from (select t.*,
sum(1 - flag) over (order by created) as grouping
from t
) t
where flag = 1
group by grouping
union all
select created, flag, amount
from t
where flag = 0;
I originally misinterpreted the question as wanting a summary for all periods, not just the adjacent "1"s. You can do this with a cumulative sum to identify the groups:
select t.*,
sum(flag) over (order by created) as grouping
from t;
And then use a subquery to aggregate this:
select min(created), max(created), count(*) as num_transactions,
sum(amount) as total_amount
from (select t.*,
sum(flag) over (order by created) as grouping
from t
) t
group by grouping;

Answer 2

You want to aggregate all consecutive rows flagged 1. You can achieve this with a running count of rows flagged 0. You can see in the table below that flag + running count of zeros identifies the groups.
Created | Amount | Flag | COUNT_0
-----------+--------+------+--------
2019-12-01 | 50,40 | 1 | 0 \ both rows flag=1, count_0=0 => one group
2019-11-21 | 50,40 | 1 | 0 /
2019-11-06 | 50,40 | 0 | 1 > the only row with flag=0, count_0=1 => one group
2019-10-04 | 50,40 | 1 | 1 \ both rows flag=1, count_0=1 => one group
2019-09-08 | 50,40 | 1 | 1 /
2019-09-01 | 50,40 | 0 | 2 > the only row with flag=0, count_0=2 => one group
2019-08-04 | 50,40 | 1 | 2 \
2019-07-24 | 50,40 | 1 | 2 | all three rows flag=1, count_0=2 => one group
2019-07-23 | 50,40 | 1 | 2 /
2019-06-01 | 50,40 | 0 | 3 > the only row with flag=0, count_0=3 => one group
2019-05-05 | 50,40 | 0 | 4 > the only row with flag=0, count_0=4 => one group
2019-04-25 | 50,40 | 1 | 4 > the only row with flag=1, count_0=4 => one group
2019-03-11 | 50,40 | 0 | 5 > the only row with flag=0, count_0=5 => one group
2019-02-03 | 50,40 | 0 | 6 > the only row with flag=0, count_0=6 => one group
2019-02-02 | 50,40 | 0 | 7 > the only row with flag=0, count_0=7 => one group
2019-02-01 | 50,40 | 0 | 8 > the only row with flag=0, count_0=8 => one group
2019-01-31 | 50,40 | 1 | 8 > the only row with flag=1, count_0=8 => one group
2019-01-26 | 50,40 | 0 | 9 > the only row with flag=0, count_0=9 => one group
2019-01-26 | 50,40 | 0 | 10 > the only row with flag=0, count_0=10 => one group
2019-01-01 | 50,40 | 1 | 10 > the only row with flag=1, count_0=10 => one group
The related query:
select min(created), min(flag), sum(amount)
from
(
select
m.*,
count(case when flag = 0 then 1 end) over (order by created) as count_0
from mytable m
)
group by flag, count_0
order by min(created);