Find start date of similar contiguous records - sql

Question

I have a table called activities that stores the activities that employees are doing. It stores simple information such as if they are working or if they are on various types of leave, e.g. annual leave, sick leave, compassionate leave etc. The table stores the employee number, the type of activity and the date that the activity is on. Only 1 type of activity can occur on a single day and only days that are normally worked will have an activity attributed to them. For example if an employee is a Monday to Friday worker and is on annual leave for a week, the weekend dates are not included in the table as they are not days the employee normally works.
Below is a sample table:
╔══════════╦════════════╦══════════════╗
║ Employee ║ Date ║ Activity ║
╠══════════╬════════════╬══════════════╣
║ 12345 ║ 25/11/2016 ║ Work ║
║ 12345 ║ 24/11/2016 ║ Work ║
║ 12345 ║ 23/11/2016 ║ Work ║
║ 12345 ║ 22/11/2016 ║ Work ║
║ 12345 ║ 21/11/2016 ║ Work ║
║ 12345 ║ 18/11/2016 ║ Work ║
║ 12345 ║ 17/11/2016 ║ Work ║
║ 12345 ║ 16/11/2016 ║ Work ║
║ 12345 ║ 15/11/2016 ║ Sick Leave ║
║ 12345 ║ 14/11/2016 ║ Sick Leave ║
║ 12345 ║ 11/11/2016 ║ Sick Leave ║
║ 12345 ║ 10/11/2016 ║ Work ║
║ 12345 ║ 9/11/2016 ║ Work ║
║ 12345 ║ 8/11/2016 ║ Work ║
║ 12345 ║ 7/11/2016 ║ Work ║
║ 12345 ║ 4/11/2016 ║ Work ║
║ 12345 ║ 3/11/2016 ║ Sick Leave ║
║ 12345 ║ 2/11/2016 ║ Sick Leave ║
║ 12345 ║ 1/11/2016 ║ Work ║
║ 12345 ║ 31/10/2016 ║ Work ║
║ 67890 ║ 25/11/2016 ║ Annual Leave ║
║ 67890 ║ 24/11/2016 ║ Annual Leave ║
║ 67890 ║ 23/11/2016 ║ Annual Leave ║
║ 67890 ║ 22/11/2016 ║ Annual Leave ║
║ 67890 ║ 21/11/2016 ║ Annual Leave ║
║ 67890 ║ 18/11/2016 ║ Work ║
║ 67890 ║ 17/11/2016 ║ Work ║
║ 67890 ║ 16/11/2016 ║ Work ║
║ 67890 ║ 15/11/2016 ║ Sick Leave ║
║ 67890 ║ 14/11/2016 ║ Sick Leave ║
║ 67890 ║ 11/11/2016 ║ Sick Leave ║
║ 67890 ║ 10/11/2016 ║ Work ║
║ 67890 ║ 9/11/2016 ║ Work ║
║ 67890 ║ 8/11/2016 ║ Work ║
║ 67890 ║ 7/11/2016 ║ Work ║
║ 67890 ║ 4/11/2016 ║ Work ║
║ 67890 ║ 3/11/2016 ║ Annual Leave ║
║ 67890 ║ 2/11/2016 ║ Annual Leave ║
║ 67890 ║ 1/11/2016 ║ Work ║
║ 67890 ║ 31/10/2016 ║ Work ║
╚══════════╩════════════╩══════════════╝
For a given employee, date and activity, I need to work backwards from that date and find the start date of the most recent block of that given activity. A 'block' is any group of the same activity, so it could be 1 day or many days.
As an example, using the table above, let's say I need to find the start date of the most recent 'Sick Leave' for employee 12345 working backwards from a date of 20/11/2016. In this case I would be looking to get a value of '11/11/2016' as this was the start date for the most recent block of sick leave.
As another example, using the table above, let's say I need to find the start date of the most recent 'Annual Leave' for employee 67890 working backwards from a date of 20/11/2016. In this case I would be looking to get a value of '21/11/2016' as this was the start date for the most recent block of annual leave.

Answer 1

This is an example of a "gaps-and-islands" problem. You can get the periods of activity for an employee using the difference of row numbers approach:
select employee, activity, min(date), max(date)
from (select t.*,
row_number() over (partition by employee order by date) as seqnum_e,
row_number() over (partition by employee, activity order by date) as seqnum_ea
from t
) t
group by employee, activity, (seqnum_e - seqnum_ea);
You can then use this to answer your questions. For instance:
with ea as (
select employee, activity, min(date) as date_from, max(date) as date_to
from (select t.*,
row_number() over (partition by employee order by date) as seqnum_e,
row_number() over (partition by employee, activity order by date) as seqnum_ea
from t
) t
group by employee, activity, (seqnum_e - seqnum_ea)
)
select top 1 ea.*
from ea
where employee = 12345 and activity = 'Sick Leave'
order by date_from desc;
There are other solutions for particular questions, but this is likely to be the most general.