Moving grouped MEDIAN / Get the MEDIAN of specific months from the past IN T-SQL - tsql

Question

Let's say I have a table:
DATE
ID
VALUE
01.2010
1
100
02.2010
1
200
...
...
...
12.2010
1
300
01.2011
1
150
02.2011
1
250
...
...
...
12.2011
1
350
01.2012
1
200
02.2012
1
300
...
...
...
12.2012
1
400
I want to get a median of VALUE grouped by months i.e. get something like
DATE
ID
VALUE
MEDIAN
01.2010
1
100
100
02.2010
1
200
200
...
...
...
...
12.2010
1
300
300
01.2011
1
150
125 = (100+150)/2
02.2011
1
250
225 = (200+250)/2
...
...
...
...
12.2011
1
350
325 = (300+350)/2
01.2012
1
200
150
02.2012
1
300
250
...
...
...
...
12.2012
1
400
350
I have more ID in table so I would like to get this result for every ID.
I have tried doing
SELECT PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY VALUE) OVER (PARTITION BY Id, MONTH(Date) ORDER BY Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW)
but I get "The function 'PERCENTILE_CONT' may not have a window frame.
I've also tried the following (but also without any results):
SELECT PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY VALUE)
OVER (PARTITION BY Id, MONTH(Date))
FROM tab1 LEFT JOIN tab2
ON tab1.key = tab2.key
WHERE tab1.Date BETWEEN Min(Date) AND tab2.Date
EDIT
So far I have resolved it with
SELECT (CASE WHEN Date =2010 THEN PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY CASE WHEN Date = 2010 THEN VALUE ELSE NULL) OVER (PARTITION BY Id, MONTH(Date)) ELSE 0 END) +
(CASE WHEN Date =2011 THEN PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY CASE WHEN Date <= 2011 THEN VALUE ELSE NULL) OVER (PARTITION BY Id, MONTH(Date)) ELSE 0 END) +
(CASE WHEN Date =2012 THEN PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY CASE WHEN Date <= 2012 THEN VALUE ELSE NULL) OVER (PARTITION BY Id, MONTH(Date)) ELSE 0 END)
FROM tab1
But to be honest, I would like to have an resolution without assumption of a priori knowledge of dates. I've thought about WHILE LOOP and updating column while #MinYear <= #MaxYear where in every iteration #MinYear = #MinYear+1 but in this case I would have to create temporary tables which I'm trying to avoid.

Answer 1

My idea is to use (Value1+value2)/2 as median as your requirement is little complicated.
CREATE TABLE MedianData
(
[Date] VARCHAR(100)
,ID INT
,[Value] INT
)
INSERT INTO MedianData VALUES ('01.2010', 1, 100)
,('02.2010', 1, 200)
,('12.2010', 1, 300)
,('01.2011', 1, 150)
,('02.2011', 1, 250)
,('12.2011', 1, 350)
,('01.2012', 1, 200)
,('02.2012', 1, 300)
,('12.2012', 1, 400)
SELECT *
,ROW_NUMBER() OVER ( PARTITION BY Substring([Date],1,2 ) ORDER BY [Date] ) AS [row]
,Substring([Date],1,2 ) as [MONTH]
INTO #Temp_tbl2
FROM MedianData
SELECT
A.Date
,A.ID
,A.[Value]
--Logic is applied here. I used (Value1+value2)/2 as median
,CASE WHEN A.[row] = 3 THEN ( A.[Value] + ( SELECT T.[Value] FROM #Temp_tbl2
T where T.[MONTH] = Substring(A.[Date],1,2 ) AND T.[row] = 1 ) )/2
WHEN A.[row] != 1 THEN (A.total/2)
ELSE A.total END as [Median]
INTO #Temp_table
FROM
(
SELECT *
,ROW_NUMBER() OVER ( PARTITION BY Substring([Date],1,2 ) ORDER BY [Date] ) AS [row]
,SUM ([Value] ) OVER ( PARTITION BY Substring([Date],1,2 ) ORDER BY [Date] ) AS [total]
FROM MedianData
) AS A
--to make the table data order
SELECT MedianData.*, #Temp_table.Median
FROM MedianData
INNER JOIN #Temp_table
ON MedianData.[Date] = #Temp_table.[Date]
drop table #Temp_table
drop table #Temp_tbl2