Split set into uneven percentage buckets - tsql

Question

Every day I am returned a set of x rows (between 5 and 2000).
I need to update a column from this set based on rules. I think this (not exactly working) example demonstrates this
/*
35% a
25% b
30% c
10% null
*/
WITH tally
(vals, updateThis, bucket)
AS
(
SELECT
DATEADD(DAY, - ROW_NUMBER() OVER (ORDER BY (SELECT NULL)), GETDATE())
, NULL
, NTILE(100) OVER (ORDER BY (SELECT NULL))
FROM
(
VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS a(n)
CROSS JOIN (VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS b(n)
CROSS JOIN (VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS c(n)
)
--UPDATE
--SET updateThis
, updated
AS
(
SELECT
t.vals
, CASE
WHEN t.bucket <= 35 THEN 'a'
WHEN t.bucket > 35 AND t.bucket <=60 THEN 'b'
WHEN t.bucket > 60 AND t.bucket <=90 THEN 'c'
WHEN t.bucket > 60 AND t.bucket <=90 THEN 'NULL'
END AS updated
, t.bucket
FROM tally t
)
SELECT
U.updated
, COUNT(1) AS actual
FROM
updated u
GROUP BY U.updated
this solution is not precise and it might not update all the rows even if a + b + c did make up 100%. Also It wouldn't work for sets smaller than 100 rows.
My current working solution is:
Calculate total rows
Calculate actual rows needed (CEILING((#totalRows * ratio) / 100)
Update the final set in a WHILE LOOP, selecting current value and rows needed.
Is there a better - set based solution that would help me get rid of the loop?

Answer 1

Don't know, if I get this correctly...
First of all there seems to be a rather obvious mistake here:
WHEN t.bucket > 60 AND t.bucket <=90 THEN 'NULL'
Shouldn't this be this:
WHEN t.bucket >90 THEN 'NULL'
The function NTILE will spread your sets into rather even buckets. Check my output and find how this behaves in the corner-cases. I suggest to use a computed percentage per row like here:
WITH tally
(vals, bucket)
AS
(
SELECT
DATEADD(DAY, - ROW_NUMBER() OVER (ORDER BY (SELECT NULL)), GETDATE())
,NTILE(100) OVER (ORDER BY (SELECT NULL))
FROM
(
VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS a(n)
CROSS JOIN (VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS b(n)
CROSS JOIN (VALUES (0), (0), (0), (0), (0), (0), (0), (0), (0)) AS c(n)
)
SELECT *
INTO #tmpBuckets
FROM Tally;
--I use this #tmpBuckets-table to get closer to your I have a table scenario
WITH Numbered AS
(
SELECT *
,ROW_NUMBER() OVER(ORDER BY vals DESC) / ((SELECT COUNT(*) FROM #tmpBuckets)/100.0) AS RunningPercentage
FROM #tmpBuckets
)
,ComputeBuckets AS
(
SELECT
t.*
, CASE
WHEN t.RunningPercentage <= 35 THEN 'a'
WHEN t.RunningPercentage > 35 AND t.RunningPercentage <=60 THEN 'b'
WHEN t.RunningPercentage > 60 AND t.RunningPercentage <=90 THEN 'c'
WHEN t.RunningPercentage >90 THEN 'NULL'
END AS ShnugoMethod
, CASE
WHEN t.bucket <= 35 THEN 'a'
WHEN t.bucket > 35 AND t.RunningPercentage <=60 THEN 'b'
WHEN t.bucket > 60 AND t.RunningPercentage <=90 THEN 'c'
WHEN t.bucket > 90 THEN 'NULL'
END AS ZikatoMethod
FROM Numbered t
)
SELECT cb.*
FROM ComputeBuckets cb
ORDER BY cb.vals DESC
GO
DROP TABLE #tmpBuckets;
I think you know, how to use such a cte to update the source table. Otherwise just come back with another question :-)