Extracting a (sampled) time series from an SQL DB - sql-server

Question

I have an MS SQL data base which contains values stored with their time stamps. So my result table looks like this:
date value
03.01.2016 11
19.01.2016 22
29.01.2016 33
17.02.2016 44
01.03.2016 55
06.03.2016 66
The time stamps don't really follow much of a pattern. Now, I need to extract weekly data from this: (sampled on Friday, for example)
date value
01.01.2016 11 // friday
08.01.2016 11 // next friday
15.01.2016 11
22.01.2016 22
29.01.2016 33
05.02.2016 33
12.02.2016 33
19.02.2016 44
26.02.2016 44
04.03.2016 55
11.03.2016 66
Is there a reasonable way to do this directly in T-SQL?
I could reformat the result table using a C# or Matlab program, but it seems a bit weird, because I seem to again query the result table...

Answer 1

You Could possibly use a CROSS JOIN or INNER JOIN. I would personally go with the INNER JOIN as its much more efficient.
SAMPLE DATA:
CREATE TABLE #Temp(SomeDate DATE
, SomeValue VARCHAR(10));
INSERT INTO #Temp(SomeDate
, SomeValue)
VALUES
('20160103'
, 11),
('20160119'
, 22),
('20160129'
, 33),
('20160217'
, 44),
('20160301'
, 55),
('20160306'
, 66)
QUERY USING CROSS JOIN:
;WITH T
AS (SELECT *
FROM #Temp),
D
AS (
SELECT SomeDate
, SomeValue
FROM #Temp AS A
UNION
SELECT DATEADD(day, 7, SomeDate)
, SomeValue
FROM #Temp AS B
UNION
SELECT DATEADD(day, 14, SomeDate)
, SomeValue
FROM #Temp AS C)
SELECT D.*
FROM T
CROSS JOIN D
WHERE T.SomeValue = D.SomeValue
ORDER BY SomeValue
, SomeDate;
RESULT:
QUERY USING INNER JOIN:
;WITH T
AS (SELECT *
FROM #Temp),
D
AS (
SELECT SomeDate
, SomeValue
FROM #Temp AS A
UNION
SELECT DATEADD(day, 7, SomeDate)
, SomeValue
FROM #Temp AS B
UNION
SELECT DATEADD(day, 14, SomeDate)
, SomeValue
FROM #Temp AS C)
SELECT D.*
FROM T
INNER JOIN D
ON T.SomeValue = D.SomeValue
ORDER BY SomeValue
, SomeDate;
RESULT:

Answer 2

This solution supports a maximum time window of 252 weeks from the first value time.
First row of your desired output is missing, because that friday is before the first value.
If needed, you can add it by mean of a UNION with a min of the table.
DECLARE #tbl TABLE ( [date] date, [value] int )
INSERT INTO #tbl
VALUES
('2016-01-03','11'),
('2016-01-19','22'),
('2016-01-29','33'),
('2016-02-17','44'),
('2016-03-01','55'),
('2016-03-06','66')
;WITH DATA
AS (
SELECT (S+P+Q) WeekNum, DATEADD( week, S + P + Q, MinDate ) Fridays, SubFri, [value]
FROM ( SELECT 1 S UNION SELECT 2 UNION SELECT 3 UNION SELECT 4 UNION SELECT 5 UNION SELECT 6 UNION SELECT 7 ) A
CROSS JOIN ( SELECT 0 P UNION SELECT 7 UNION SELECT 14 UNION SELECT 21 UNION SELECT 28 UNION SELECT 35 ) B
CROSS JOIN ( SELECT 0 Q UNION SELECT 42 UNION SELECT 84 UNION SELECT 126 UNION SELECT 168 UNION SELECT 210 ) C
CROSS JOIN (
SELECT
min ( DATEADD( day, -8 - DATEPART(weekday,[date]), [date] ) ) MinDate,
max ( DATEADD( day, 13 - DATEPART(weekday,[date]), [date] ) ) MaxDate
FROM #tbl
) MD
LEFT JOIN ( SELECT DATEADD( day, 6 - DATEPART(weekday,[date]), [date] ) SubFri, [value] FROM #tbl ) Val
ON SubFri<=DATEADD( week, S + P + Q, MinDate )
WHERE DATEADD( week, S + P + Q, MinDate )<=MaxDate
)
SELECT DATA.Fridays, DATA.value
FROM DATA
INNER JOIN
(
SELECT Fridays, max(SubFri) MaxSubFri
FROM DATA
GROUP BY Fridays
) idx
ON DATA.Fridays=idx.Fridays
AND SubFri=MaxSubFri
ORDER BY Fridays

Answer 3

You should be able to use DATENAME to get all the records of a certain day:
SELECT *
FROM table
WHERE DATENAME(WEEKDAY, date) = 'Friday'
This causes a scan in the query plan though so it would be advisable to have another column with the day of the week and you could just select WHERE dayOfWeekCol = 'Friday'

Answer 4

I found my own solution, which I find more readable. I'm first using a WHILE loop to generate the dates I'm looking for. Then I 'join' these dates to the actual data table using an OUTER APPLY, which looks up 'last value before a specific date'. Here's the code:
-- prepare in-memory table
declare #tbl table ( [date] date, [value] int )
insert into #tbl
values
('2016-01-03','11'),
('2016-01-19','22'),
('2016-01-29','33'),
('2016-02-17','44'),
('2016-03-01','55'),
('2016-03-06','66')
-- query
declare #startDate date='2016-01-01';
declare #endDate date='2016-03-31';
with Fridays as (
select #startDate as fridayDate
union all
select dateadd(day,7,fridayDate) from Fridays where dateadd(day,7,fridayDate)<=#endDate
)
select *
from
Fridays f
outer apply (
select top(1) * from #tbl t
where f.fridayDate >= t.[date]
order by t.[value] desc
) as result
option (maxrecursion 10000)
Gives me:
fridayDate date value
---------- ---------- -----------
2016-01-01 NULL NULL
2016-01-08 2016-01-03 11
2016-01-15 2016-01-03 11
2016-01-22 2016-01-19 22
2016-01-29 2016-01-29 33
2016-02-05 2016-01-29 33
2016-02-12 2016-01-29 33
2016-02-19 2016-02-17 44
2016-02-26 2016-02-17 44
2016-03-04 2016-03-01 55
2016-03-11 2016-03-06 66
2016-03-18 2016-03-06 66
2016-03-25 2016-03-06 66
Thanks for everybody's ideas and support though!