get number of occurrences taking 1 for every 3 days group - sql

Question

I've got a table for employees that usually get late to work. I need to send a report to Human Resources showing every user that got late, taking into account that I just can count a warning per user if that user got late in a period of 3 days at least 1 time within period.
The first data I need is the total number of warnings to be sent for HR manager to evaluate global "lateness".
Users that got late just one day will receive one warning, but if they got late twice or more the warnings they'll receive depend if they received a warning within a 3 days period counting from day one.
Let's see with an example:
Joe monday 9th
Mark monday 9th
Tim monday 9th
Joe tuesday 10th
Joe wednesday 11th
Joe Thursday 12th
Tim Friday 13th
Taking the data from table above as an example.
Joe will receive 2 warnings: first for monday, and second for Thursday. Tuesday and Wednesday will be discarded because they belonged to the same 3 day period.
Mark will receive just one warning for monday.
Tim will receive 2 warnings. First for monday and second for Friday.
Maybe this number is not possible to get using standard sql query and some cursors need to be done.
Thanks in advance

Answer 1

Ok... some information is missing here, as correctly stated in both comments (Quassnoi and Mr. Llama).
RDBMS in use can impact on the solution as this has to do with date functions and date algebra and not all RDBMS share the same extended function set. I have presumed MySQL 5.5, which is quite common and can be tested on SQLFiddle.
Your 3 day period is also a bit vague. Is it the same for all employees, or on what does it depend? Is it 3 days, or half a week (mon-wed, thu-sat)? What happens after that? Do we use sun-tue, then wed-fri and so on? I have presumed half-weeks (sun-wed, thu-sat) the same for all employees.`So a period is identified by year, week-of-year, half-week.
The last point you should clear up is the expected result set. Do you want a list of warning days or a count or what? I have presumed a list of warnings with the date of each (I've taken the first date for each combination of employee & period).
Creating your sample dataset with these statements:
CREATE TABLE LateEntrances (
Employee VARCHAR(20),
DateLate DATE
);
INSERT INTO LateEntrances VALUES('Joe' ,'2014.06.09');
INSERT INTO LateEntrances VALUES('Mark','2014.06.09');
INSERT INTO LateEntrances VALUES('Tim' ,'2014.06.09');
INSERT INTO LateEntrances VALUES('Joe' ,'2014.06.10');
INSERT INTO LateEntrances VALUES('Joe' ,'2014.06.11');
INSERT INTO LateEntrances VALUES('Joe' ,'2014.06.12');
INSERT INTO LateEntrances VALUES('Tim' ,'2014.06.13');
The following query solves your problem:
SELECT i.Employee, i.YearLate, i.WeekLate, i.PeriodLate, MIN(i.DateLate)
FROM (
SELECT Employee, DateLate,
YEAR(DateLate) AS YearLate,
WEEKOFYEAR(DateLate) AS WeekLate,
FLOOR(DAYOFWEEK(DateLate)/4) AS PeriodLate
FROM LateEntrances
) i
GROUP BY i.Employee, i.YearLate, i.WeekLate, i.PeriodLate;
(SQLFiddle here)
The three columns YearLate, WeekLate and PeriodLate identify the warning period. You could concatenate them in a single period identification column:
SELECT i.Employee, i.PeriodLate, MIN(i.DateLate)
FROM (
SELECT Employee, DateLate,
CONCAT_WS('*',
YEAR(DateLate) ,
WEEKOFYEAR(DateLate) ,
FLOOR(DAYOFWEEK(DateLate)/4)
) AS PeriodLate
FROM LateEntrances
) i
GROUP BY i.Employee, i.PeriodLate;
... or you could hide them alltogether (in the SELECT), even though you must still use them to GROUP BY:
SELECT i.Employee, MIN(i.DateLate)
FROM (
SELECT Employee, DateLate,
CONCAT_WS('*',
YEAR(DateLate) ,
WEEKOFYEAR(DateLate) ,
FLOOR(DAYOFWEEK(DateLate)/4)
) AS PeriodLate
FROM LateEntrances
) i
GROUP BY i.Employee, i.PeriodLate;
You can also easily change the period calculation logic to something else, like strict 3 days of year, or 3 days of month periods. There are many possibilities.
...as per the assumptions I made. Clear up the open points and I'll try to make the answer better. But in the meantime this should be enough to get you started.