NTILE alternative for non-uniformely distributed data sets - sql

Question

I have a data set and want to display it, but it can be very huge (thousands of points), and I want to filter them. For example here is output for 1000+ points:
Now i use NTILE to get approximation, but it doesn't work as expexted if points are not distributed uniformly. And I get this output (NTILE with parameter 100):
How can I avoid this behaviour? SQL stored procedure is below:
ALTER PROCEDURE [dbo].[usp_GetSystemHealthCheckData]
#DateFrom datetime,
#DateTo datetime,
#EstimatedPointCount int
with recompile
AS
BEGIN
SET NOCOUNT ON;
set arithabort on
if #DateFrom IS NULL
RAISERROR ('#DateFrom cannot be NULL', 16, 1)
if #DateTo IS NULL
RAISERROR ('#DateTo cannot be NULL', 16, 1)
if #EstimatedPointCount IS NULL
RAISERROR ('#EstimatedPointCount cannot be NULL', 16, 1)
;With T as
(
SELECT *, GroupId = NTILE(#EstimatedPointCount) over (order by GeneratedOnUtc)
FROM SystemHealthCheckData
WHERE GeneratedOnUtc between #DateFrom AND #DateTo
)
SELECT CpuPercentPayload = AVG(CpuPercentPayload),
FreeRamMb = AVG(FreeRamMb),
FreeDriveMb = AVG(FreeDriveMb),
GeneratedOnUtc = CAST(AVG(CAST(GeneratedOnUtc AS DECIMAL( 18, 6))) AS DATETIME)
FROM T
GROUP BY GroupId
END

Answer 1

EDIT: new approach
You might split your load with NTILE and then calculate an average for each group? I splitted my set in 4 groups. This lets the query come back with 4 average values. The number of groups could be calculated from the number of points you have or could be done fix.
Something like this:
DECLARE #tbl TABLE(id INT IDENTITY, nmbr FLOAT);
INSERT INTO #tbl VALUES(5),(4.5),(4),(3.5),(3),(2.5),(2),(1.5),(1),(1.5),(1),(0.5),(0),(13),(2),(17),(5),(22),(24),(2),(3),(11);
SELECT tbl2.*
,AVG(nmbr) OVER(PARTITION BY tbl2.tile)
FROM
(
SELECT tbl.*
,NTILE(4) OVER(ORDER BY id) AS tile
FROM #tbl AS tbl
)AS tbl2
If you want it reduced to the group values only you could try this
SELECT AVG(nmbr),tbl2.tile
FROM
(
SELECT tbl.*
,NTILE(4) OVER(ORDER BY id) AS tile
FROM #tbl AS tbl
)AS tbl2
GROUP BY tbl2.tile
--old text
You maybe want to think about a sliding average... In this example I tried to rebuild your values (long linear falling and wild jumping at the end). You can set the #pre and #post variables to set the grade of "flatening".
In short: There is an average calculated for each element and its direct neighbours.
Be aware of the fact that you must add an ORDER BY to avoid random results...
DECLARE #tbl TABLE(id INT IDENTITY, nmbr FLOAT);
INSERT INTO #tbl VALUES(5),(4.5),(4),(3.5),(3),(2.5),(2),(1.5),(1),(1.5),(1),(0.5),(0),(13),(2),(17),(5),(22),(24),(2),(3),(11);
DECLARE #pre INT=3;
DECLARE #post INT=3;
SELECT tbl.*
,AvgBorders.*
,AvgSums.*
,AvgSlide.*
FROM #tbl AS tbl
CROSS APPLY
(
SELECT tbl.id-#pre AS AvgStart
,tbl.id + #post AS AvgEnd
) AS AvgBorders
CROSS APPLY
(
SELECT COUNT(nmbr) AS CountNmbr
,SUM(nmbr) AS SumNmbr
FROM #tbl AS tbl
WHERE tbl.id BETWEEN AvgStart AND AvgEnd
) as AvgSums
CROSS APPLY
(
select AvgSums.SumNmbr / AvgSums.CountNmbr As AvgValue
) As AvgSlide
;