SQL - Distribute Same Values equally across X number of tables - sql

Question

I want to see if someone knows a way to evenly distribute multiple like values across "x" number of temp tables ensuring that the 'like' values (same team name in this example) never get lumped into one particular table. What I am trying to do is create heats for a race and evenly distribute teams across tables. Ex:
**Teams**
-----------
Los Angeles
New York
New York
Los Angeles
Florida
Florida
Arizona
Texas
Alabama
Alaska
New York
New York
New York
I would like the distribution to go end up something like this where all multiple teams are evenly distribute across 2 (or 3 or 4) heats:
**Heat One**
-------------
Los Angeles
New York
Florida
Arizona
Alabama
New York
New York
**Heat Two**
------------
Los Angeles
New York
Florida
Texas
Alaska
New York

Answer 1

Starting with SQL Server 2005, there's a native functionality for bucketing data. NTILE()
The NTILE function is the fourth of four windowing functions introduced in SQL Server 2005. NTILE takes a different approach to paritioning data. ROW_NUMBER, RANK and DENSE_RANK will generate variable sized buckets of data based on the partition key(s). NTILE attempts to split the data into equal, fixed size buckets. BOL has a comprehensive page comparing the ranking functions if
you want a quick visual reference on their effects.
Syntax
The syntax for NTILE differs slightly from the other window functions. It's NTILE(#BUCKET_COUNT) OVER ([PARTITION BY _] ORDER BY _) , where #BUCKET_COUNT is a positive integer or bigint value.
The challenge is ensuring we get a good distribution and that's the part that is subject to the vagueries of the random number generator (newid calls/(SELECT NULL)).
Leveraging Rhys's setup
CREATE table dbo.Teams (TeamId int, TeamName varchar(32));
insert dbo.Teams values
( 1, 'Los Angeles'),
( 2, 'New York'),
( 3, 'New York'),
( 4, 'Los Angeles'),
( 5, 'Florida'),
( 6, 'Florida'),
( 7, 'Arizona'),
( 8, 'Texas'),
( 9, 'Alabama'),
(10, 'Alaska'),
(11, 'New York'),
(12, 'New York'),
(13, 'New York');
SELECT
NTILE(2) OVER (ORDER BY NEWID()) AS Heat
, NTILE(2) OVER (ORDER BY (SELECT NULL)) AS HeatAlternate
, T.TeamName
, T.TeamId
FROM
dbo.Teams AS T
ORDER BY
1,3;
One of the nicer things about this approach is that it can be switched out to make whatever bucketing size you want by simply changing the value passed to ntile. It also ought to scale better as it would only take one pass through the source table.

Answer 2

This approach doesn't sound right (having separate tables called Heat1, Heat2 etc) so you might want to re-think what you're doing, but if your circumstances dictate this is a good approach then how about allocating a random unique (but sequential) number to each team then use MOD to split the teams across heats? In order to get the 'like' teams (same teamname) into different heats they just need to be randomised together and the MOD will separate them.
create table dbo.Teams (TeamId int, TeamName varchar(32))
go
insert dbo.Teams values
( 1, 'Los Angeles'),
( 2, 'New York'),
( 3, 'New York'),
( 4, 'Los Angeles'),
( 5, 'Florida'),
( 6, 'Florida'),
( 7, 'Arizona'),
( 8, 'Texas'),
( 9, 'Alabama'),
(10, 'Alaska'),
(11, 'New York'),
(12, 'New York'),
(13, 'New York')
go
-- First get a random number per unique team name
; with cte as (
select row_number() over (order by newid()) as lrn, t.TeamName
from dbo.Teams t
group by t.TeamName
)
-- Second get a unique random number per team with like teams ordered together
select row_number() over (order by lrn, newid()) - 1 as rn, t.*
into #teams
from dbo.Teams t
join cte c on c.TeamName = t.TeamName
select 'Heat1', *
from #teams
where rn % 4 = 0
select 'Heat2', *
from #teams
where rn % 4 = 1
select 'Heat3', *
from #teams
where rn % 4 = 2
select 'Heat4', *
from #teams
where rn % 4 = 3