1. 技术文章
  2. % of participants that are retained during six months

% of participants that are retained during six months

由 夙愿已清 提交于 2019-12-05 04:16:42
Roman Pekar

Well I should say that is not an easy one. The main problem was to solve 'at least twice a week for a sixth months' part - it's easy to calculate twice a week, but it should be 6 continous months!

While I've tried to solve it, I've found absolutely brilliant answer by Niels van der Rest - Finding continuous ranges in a set of numbers. So I'll give you general query for the Part 1, you can change parameters and get result for Part 2:

declare @Weeks int, @PerWeek int, @StartDate date, @EndDate date, @count

select
    @StartDate = '20120701',
    @EndDate = '20130630',
    @Weeks = 26, -- 6 month or 26 weeks
    @PerWeek = 2 -- twice per week

select @count = count(distinct A.ID)
from Attendance as A
where
    A.Attendence_date between @StartDate and @EndDate and
    A.ID not in (select T.ID from Deenrolled as T) and
    A.ID not in (select T.ID from Inactive as T)

;with CTE as (
    -- Week numbers, filter by dates
    select
        A.ID,
        datediff(dd, @StartDate, A.Attendence_date) / 7 as Wk
    from Attendance as A
    where
        A.Attendence_date between @StartDate and @EndDate and
        A.ID not in (select T.ID from Deenrolled as T) and
        A.ID not in (select T.ID from Inactive as T)
  ), CTE2 as (
    -- Group by week, filter less then @PerWeek per week, calculate row number
    select
        Wk, ID,
        row_number() over (partition by ID order by Wk) as Row_Num
    from CTE
    group by Wk, ID
    having count(*) >= @PerWeek
)
-- Final query - group by difference between week and row_number
select 100 * cast(count(distinct ID) as float) / @count
from CTE2
group by ID, Wk - Row_Num
having count(*) >= @Weeks

I've created SQL FIDDLE EXAMPLE, you can test the query.

Part 3 is easy

declare @PerWeek int, @StartDate date

select
    @StartDate = '20130101',
    @PerWeek = 2 -- twice per week

select @count = count(distinct A.ID)
from Attendance as A
where
    A.Attendence_date >= @StartDate and
    A.ID not in (select T.ID from Deenrolled as T) and
    A.ID not in (select T.ID from Inactive as T)

;with CTE as (
    -- Week numbers, filter by dates
    select
        A.ID,
        datediff(dd, @StartDate, A.Attendence_date) / 7 as Wk
    from Attendance as A
    where
        A.Attendence_date >= @StartDate and
        A.ID not in (select T.ID from Deenrolled as T) and
        A.ID not in (select T.ID from Inactive as T)
  ), CTE2 as (
    -- Group by week, filter less then @PerWeek per week
    select distinct ID
    from CTE
    group by Wk, ID
    having count(*) >= @PerWeek
)
select 100 * cast(count(*) as float) / @count from CTE2

Part 4 seems a bit unclear for me, could you clarify?

Give these a shot (changed because I missed a huge part of the question)

    SELECT  B.ID FROM
(SELECT number
      FROM master.dbo.spt_values
      WHERE TYPE = 'P' AND number < datediff(week, '07/01/2012', '06/30/2013') ) AS W
    JOIN
(SELECT A.ID, weeknum
  FROM
    (SELECT ID,  datediff(week, '07/01/2012',Attendence_date) AS weeknum
      FROM Attendance
      WHERE Attendence_date  BETWEEN '07/01/2012' AND '06/30/2013'
        AND ID NOT IN (SELECT ID FROM Deenrolled)
        AND ID NOT IN (SELECT ID FROM Inactive)) AS A
  GROUP BY A.ID, A.weeknum
  HAVING COUNT(A.ID) > 2) AS B ON W.number = B.weeknum
GROUP BY B.ID
HAVING COUNT(W.number) = datediff(week, '07/01/2012', '06/30/2013');

SELECT  B.ID FROM
(SELECT number
      FROM master.dbo.spt_values
      WHERE TYPE = 'P' AND number < datediff(week, '01/01/2013', '06/30/2013') ) AS W
    JOIN
(SELECT A.ID, weeknum
  FROM
    (SELECT ID,  datediff(week, '01/01/2013',Attendence_date) AS weeknum
      FROM Attendance
      WHERE Attendence_date  BETWEEN '01/01/2013' AND '06/30/2013'
        AND ID NOT IN (SELECT ID FROM Deenrolled)
        AND ID NOT IN (SELECT ID FROM Inactive)) AS A
  GROUP BY A.ID, A.weeknum
  HAVING COUNT(A.ID) > 2) AS B ON W.number = B.weeknum
GROUP BY B.ID
HAVING COUNT(W.number) = datediff(week, '07/01/2012', '06/30/2013');

SELECT  B.ID FROM
(SELECT number
      FROM master.dbo.spt_values
      WHERE TYPE = 'P' AND number < datediff(week, '01/01/2013', '06/30/2013') ) AS W
    JOIN
(SELECT A.ID, weeknum
  FROM
    (SELECT ID,  datediff(week, '01/01/2013',GetDate()) AS weeknum
      FROM Attendance
      WHERE Attendence_date  BETWEEN '01/01/2013' AND GetDate()
        AND ID NOT IN (SELECT ID FROM Deenrolled)
        AND ID NOT IN (SELECT ID FROM Inactive)) AS A
  GROUP BY A.ID, A.weeknum
  HAVING COUNT(A.ID) > 2) AS B ON W.number = B.weeknum
GROUP BY B.ID
HAVING COUNT(W.number) = datediff(week, '07/01/2012', GetDate());

SELECT DISTINCT(Attendance.ID)
  FROM Attendance
  WHERE Attendance.ID NOT IN (SELECT ID FROM Deenrolled)
      AND ID NOT IN (SELECT ID FROM Inactive);

and an sqlfiddle to help you out: http://sqlfiddle.com/#!6/97230/3 Good luck!

标签
sql
sql-server
date
date-range
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!