How to produce such results using SQL - sql

Question

I edited my question as it seems like people misunderstood what I wanted.
I have a table which has the following columns:
Company
Transaction ID
Transaction Date
The result I want is:
| COMPANY | Transaction ID |Transaction Date | GROUP
|---------------------|------------------|------------------|----------
| Company A | t_0001 | 01-01-2014 | 1
| Company A | t_0002 | 02-01-2014 | 1
| Company A | t_0003 | 04-01-2014 | 1
| Company A | t_0003 | 10-01-2014 | 2
| Company B | t_0004 | 02-01-2014 | 1
| Company B | t_0005 | 02-01-2014 | 1
| Company C | t_0006 | 03-01-2014 | 1
| Company C | t_0007 | 05-01-2014 | 2
where the transactions and dates are firstly group into companies. The transactions within the company are sorted from the earliest to the latest. The transactions are checked, row by row, if the previous transaction was performed less than 3 days ago in a moving window period.
For example, t_0002 and t_0001 are less than 3 days apart so they fall under group 1. t_0003 and t_0002 are less than 3 days apart so they fall under group 1 even though t_0003 and t_0003 are >= 3 days apart.
I figured the way to go about doing this is to group the data by companies first, following by sorting the transactions by the dates, but I got stuck after this. Like what methods are there I could use to produce this results? Any help on this?
P.S. I am using SQL Server 2014.

Answer 1

I have determined days difference between each company following by transaction id. so if days difference is less than 3 goes to group 1 other are 2. Based on your requirement alter the lag clause and use it.
select *,isnull(
case when datediff(day,
lag([Transaction Date]) over(partition by company order by [transaction id]),[Transaction Date])>=2
then
2
end ,1)group1
from #Table1

Answer 2

If you don't care about the numbering in groups, use
select *,
dense_rank() over(partition by company order by transaction_date) -
(select count(distinct transaction_date) from t
where t1.company=company
and datediff(dd,transaction_date,t1.transaction_date) between 1 and 2) grp
from t t1
order by 1,3
Sample Demo
If continuous numbers are needed for groups, use
select company,transaction_id,transaction_date,
dense_rank() over(partition by company order by grp) grp
from (select *, dense_rank() over(partition by company order by transaction_date) -
(select count(distinct transaction_date) from t
where t1.company=company
and datediff(dd,transaction_date,t1.transaction_date) between 1 and 2) grp
from t t1
) x
order by 1,3

Answer 3

create table xn (
[Company] char(1),
[Transaction ID] char(6),
[Transaction Date] date,
primary key ([Company], [Transaction ID], [Transaction Date])
);
insert into xn values
('A', 't_0001', '2014-01-01'),
('A', 't_0002', '2014-01-02'),
('A', 't_0003', '2014-01-04'),
('A', 't_0003', '2014-01-10'),
('B', 't_0004', '2014-01-02'),
('B', 't_0005', '2014-01-02'),
('C', 't_0006', '2014-01-03'),
('C', 't_0007', '2014-01-05');
Each query builds on the one before. There are more concise ways to write queries like this, but I think this way helps when you're learning window functions like lag(...) over (...).
The first one here brings the previous transaction date into the "current" row.
select
[Company],
[Transaction ID],
[Transaction Date],
lag ([Transaction Date]) over (partition by [Company] order by [Transaction Date]) as [Prev Transaction Date]
from xn
This query determines the number of days between the "current" transaction date and the previous transaction date.
select
[Company],
[Transaction ID],
[Transaction Date],
[Prev Transaction Date],
DateDiff(d, [Prev Transaction Date], [Transaction Date]) as [Days Between]
from (select
[Company],
[Transaction ID],
[Transaction Date],
lag ([Transaction Date]) over (partition by [Company] order by [Transaction Date]) as [Prev Transaction Date]
from xn) x
This does the grouping based on the number of days.
select
[Company],
[Transaction ID],
[Transaction Date],
case when [Days Between] between 0 and 3 then 1
when [Days Between] is null then 1
when [Days Between] > 3 then 2
else 'Ummm'
end as [Group Num]
from (
select
[Company],
[Transaction ID],
[Transaction Date],
[Prev Transaction Date],
DateDiff(d, [Prev Transaction Date], [Transaction Date]) as [Days Between]
from (select
[Company],
[Transaction ID],
[Transaction Date],
lag ([Transaction Date]) over (partition by [Company] order by [Transaction Date]) as [Prev Transaction Date]
from xn) x
) y;