Pivot table with more then one record - sql

Question

I have data that is structured as follow:
LocationId, GroupId, DayOfWeek, Count, DatetimeValue15Min
2 9 4 5 2014-01-02 08:15:00.000
2 9 4 5 2014-01-02 09:15:00.000
I want to calculate the mode for each day, the data above already contains the count to know the mode. I have written a query with a pivot.
SELECT
pvt.LocationId, pvt.GroupId, [1], [2], [3], [4],[5]
FROM
#TempResult
PIVOT
(min ([DatetimeValue15Min])
FOR DayOfWeek IN ( [1], [2], [3], [4],[5])) AS pvt
In this case I have two modes but i want to show them both. My query returns in this case just the mode with the minimum value. I know that I can make a second query with the max value but what if i have more than two modes?
The output should be like:
LocationId GroupId 1 2 3 4 5
2 9 08:15, 09:15
I am using SQL Server 2005.

Answer 1

You are almost there. You just need to build the comma-separated list. A little xml type abuse works really well for that.
;WITH
t1 AS ( --Add a grouping id for quick reference
SELECT
[LocationId],[GroupId],[DayOfWeek],[DatetimeValue15Min],
DENSE_RANK() OVER(ORDER BY [LocationId],[GroupId],[DayOfWeek]) [i]
FROM #TempResult
),
t2 AS ( --Build a comma-separated list of all [DatetimeValue15Min] with same grouping id
SELECT [LocationId],[GroupId],[DayOfWeek],
CAST(REPLACE((SELECT CONVERT(time, [DatetimeValue15Min]) AS a FROM t1 WHERE [i] = t.[i] FOR xml PATH('')),'</a><a>',',') AS xml).value('a[1]','varchar(max)') [dtv_list]
FROM t1 t
)
SELECT pvt.LocationId, pvt.GroupId, [1], [2], [3], [4],[5]
FROM t2
PIVOT
(
min ([dtv_list])
FOR DayOfWeek IN ( [1], [2], [3], [4],[5])
) AS pvt
The xml trick works like this:
SELECT [DatetimeValue15Min] FOR XML -> <a>08:15</a><a>09:15</a>
replace '</a><a>' with ',' -> <a>08:15,09:15</a>
extract first node from xml -> '08:15,09:15'

Answer 2

SELECT LocationId, GroupId
, STUFF (
MAX (CASE WHEN DayOfWeek = 1 AND num = 1 THEN Value ELSE '' END)
+ MAX (CASE WHEN DayOfWeek = 1 AND num = 2 THEN Value ELSE '' END), 1, 2, '') AS [1]
, STUFF (
MAX (CASE WHEN DayOfWeek = 2 AND num = 1 THEN Value ELSE '' END)
+ MAX (CASE WHEN DayOfWeek = 2 AND num = 2 THEN Value ELSE '' END), 1, 2, '') AS [2]
, STUFF (
MAX (CASE WHEN DayOfWeek = 3 AND num = 1 THEN Value ELSE '' END)
+ MAX (CASE WHEN DayOfWeek = 3 AND num = 2 THEN Value ELSE '' END), 1, 2, '') AS [3]
, STUFF (
MAX (CASE WHEN DayOfWeek = 4 AND num = 1 THEN Value ELSE '' END)
+ MAX (CASE WHEN DayOfWeek = 4 AND num = 2 THEN Value ELSE '' END), 1, 2, '') AS [4]
, STUFF (
MAX (CASE WHEN DayOfWeek = 5 AND num = 1 THEN Value ELSE '' END)
+ MAX (CASE WHEN DayOfWeek = 5 AND num = 2 THEN Value ELSE '' END), 1, 2, '') AS [5]
FROM (
SELECT LocationId, GroupId, DayOfWeek, ', ' + CONVERT (VARCHAR(5), DatetimeValue15Min, 8) AS Value
, ROW_NUMBER() OVER(PARTITION BY LocationId, GroupId, DayOfWeek ORDER BY DatetimeValue15Min) AS num
FROM #TempResult
) T
GROUP BY LocationId, GroupId
UPD. Solution for unlimited number of values per day.
WITH cte AS (
SELECT LocationId, GroupId, DayOfWeek, CONVERT (VARCHAR(5), DatetimeValue15Min, 8) AS Value
, ROW_NUMBER() OVER(PARTITION BY LocationId, GroupId, DayOfWeek ORDER BY DatetimeValue15Min) AS num
FROM #TempResult
)
SELECT LocationId, GroupId
, MIN (CASE DayOfWeek WHEN 1 THEN Value END) AS [1]
, MIN (CASE DayOfWeek WHEN 2 THEN Value END) AS [2]
, MIN (CASE DayOfWeek WHEN 3 THEN Value END) AS [3]
, MIN (CASE DayOfWeek WHEN 4 THEN Value END) AS [4]
, MIN (CASE DayOfWeek WHEN 5 THEN Value END) AS [5]
FROM (
SELECT DISTINCT c1.LocationId, c1.GroupId, c1.DayOfWeek
, STUFF ( (
SELECT ', ' + c2.Value
FROM cte c2
WHERE c1.LocationId = c2.LocationId AND c1.GroupId = c2.GroupId AND c1.DayOfWeek = c2.DayOfWeek
FOR XML PATH ('')
), 1, 2, '') AS Value
FROM cte c1
) t
GROUP BY LocationId, GroupId