How to combine Date periods in SQL and create a composite timeline - sql

Question

In a oracle table for each family (unique id) can have several people (unique id) in different relationship for a date range. I would like to get a timeline created to obtain the FamilyType based on combinations of relationship for the time periods. An example is given below for a particular Family.
P1|-----Head---------------------------------------|
P2|--Partner--------------|
P3|---Child----------------------|
P4|---Child------------|
|=Single=|=Couple=|=Family=======|=SingleParent==|
Table has columns
FamilyId, PersonId, Relationship, StartDate, EndDate
Each | is a date (no time portion). The data guarantees that on a given date
* there will always be one person who is Head.
* There can be 0 or 1 Partner.
* There can be 0 or n child.
The rules are
* If there is only a Head the FamilyType is Single
* If there is a Head and a Partner the FamilyType is Couple
* If there is a Head , a Partner and 1 or more Children the FamilyType is Family
* If there is a Head and 1 or more Children the FamilyType is SingleParent
People can join or leave from a family on any date.
And people can change relationships. So following scenarios are possible
P1|----------Head--------------------|
P2|----partner------------|---Head--------|
P3|---Child----------------------|
P4|--Child-----------------|
|=Single=|=Couple=|=Family=======|=SingleParent==|
P1|----------Head--------------------|
P2|----partner------------|---Head--------|
P3|---Child----------------------|
P4|--Child-----------------|
p5|---Partner-----|
|=Single=|=Couple=|=Family=======================|
How can this be done using SQL in Oracle 11GR2 (working using SQL only and not using procedural code). I am trying to evaluate whether this is best done in SQL or C#. As a curiosity answer specific for SQL Server 2012 is also good to have.
The result should be rows with StartDate, EndDate and FamilyType.

Answer 1

you could do something like this:
with family_ranges(familyid, min_start, max_end, curr_date)
as (select familyid,
min(startdate),
max(enddate),
to_number(to_char(min(startdate), 'j'))
from family
group by familyid
union all
select familyid, min_start, max_end, curr_date+1
from family_ranges
where curr_date < to_number(to_char(max_end,'j')))
select familyid, min(curr_date) fromdate, max(curr_date) todate, state
from (select familyid, to_date(curr_date,'j') curr_date,
case when head = 'Y' and partner = 'Y' and child = 'Y' then 'Family'
when head = 'Y' and partner = 'Y' then 'Couple'
when head = 'Y' and child = 'Y' then 'SingleParent'
when head = 'Y' then 'Single'
end state
from (select f.familyid, d.curr_date, f.relationship
from family_ranges d
inner join family f
on f.familyid = d.familyid
and to_date(d.curr_date,'j') between f.startdate and f.enddate)
pivot (
max('Y')
for relationship in ('Head' as head, 'Partner' as partner, 'Child' as child)
))
group by familyid, state
order by familyid, fromdate;
forgive the nonsense with the date->julian. it's to work round a bug with 11.2.0.1-3 where date arithmetic fails with factored subqueries.
the fatored subquery part gets us a list of dates that the family spans. From that we join it back to family to work out who was in the family on that day.
select f.familyid, d.curr_date, f.relationship
from family_ranges d
inner join family f
on f.familyid = d.familyid
and to_date(d.curr_date,'j') between f.startdate and f.enddate;
now we pivot this to get a simple Y/N list
SQL> with family_ranges(familyid, min_start, max_end, curr_date)
2 as (select familyid,
3 min(startdate),
4 max(enddate),
5 to_number(to_char(min(startdate), 'j'))
6 from family
7 group by familyid
8 union all
9 select familyid, min_start, max_end, curr_date+1
10 from family_ranges
11 where curr_date < to_number(to_char(max_end,'j')))
12 select familyid, to_date(curr_date,'j') curr_date, head, partner, child
13 from (select f.familyid, d.curr_date, f.relationship
14 from family_ranges d
15 inner join family f
16 on f.familyid = d.familyid
17 and to_date(d.curr_date,'j') between f.startdate and f.enddate)
18 pivot (
19 max('Y')
20 for relationship in ('Head' as head, 'Partner' as partner, 'Child' as child)
21 );
FAMILYID CURR_DATE H P C
---------- --------- - - -
1 09-NOV-12 Y
1 11-NOV-12 Y
1 13-NOV-12 Y
1 23-NOV-12 Y
2 23-NOV-12 Y
2 28-NOV-12 Y Y
2 29-NOV-12 Y Y
1 30-NOV-12 Y Y
1 01-DEC-12 Y Y
1 03-DEC-12 Y Y
2 18-DEC-12 Y Y Y
2 20-DEC-12 Y Y Y
then its a simple case to get your required string from the rules and a group by to get the date ranges.
SQL> with family_ranges(familyid, min_start, max_end, curr_date)
2 as (select familyid,
3 min(startdate),
4 max(enddate),
5 to_number(to_char(min(startdate), 'j'))
6 from family
7 group by familyid
8 union all
9 select familyid, min_start, max_end, curr_date+1
10 from family_ranges
11 where curr_date < to_number(to_char(max_end,'j')))
12 select familyid, min(curr_date) fromdate, max(curr_date) todate, state
13 from (select familyid, to_date(curr_date,'j') curr_date,
14 case when head = 'Y' and partner = 'Y' and child = 'Y' then 'Family'
15 when head = 'Y' and partner = 'Y' then 'Couple'
16 when head = 'Y' and child = 'Y' then 'SingleParent'
17 when head = 'Y' then 'Single'
18 end state
19 from (select f.familyid, d.curr_date, f.relationship
20 from family_ranges d
21 inner join family f
22 on f.familyid = d.familyid
23 and to_date(d.curr_date,'j') between f.startdate and f.enddate)
24 pivot (
25 max('Y')
26 for relationship in ('Head' as head, 'Partner' as partner, 'Child' as child)
27 ))
28 group by familyid, state
29 order by familyid, fromdate;
FAMILYID FROMDATE TODATE STATE
---------- --------- --------- ------------
1 05-NOV-12 24-NOV-12 Single
1 25-NOV-12 14-DEC-12 Couple
1 15-DEC-12 24-JAN-13 Family
1 25-JAN-13 13-FEB-13 SingleParent
2 05-NOV-12 24-NOV-12 Single
2 25-NOV-12 14-DEC-12 Couple
2 15-DEC-12 13-FEB-13 Family
fiddle: http://sqlfiddle.com/#!4/484b5/1