Question

What would be the most efficient way to select a non previously selected pair of different random values from a column of unique (non repeated) values?

My current approach is (keeping every pair of values already associated in a separate “mytable_associations” table):

SELECT * FROM
(
 SELECT id,count(*) AS associations_count FROM mytable 
 INNER JOIN mytable_associations 
 WHERE (myvalue=myvalue1 OR myvalue=myvalue2) 
 GROUP BY myvalue 
 HAVING associations_count<(SELECT count(*) FROM mytable)-1
 ORDER BY rand() limit 1
) mytable1 
LEFT JOIN 
(SELECT myvalue AS myvalue2 FROM mytable) mytable2
ON mytable1.myvalue1<>mytable2.myvalue2
WHERE
(
 SELECT myvalue1 FROM mytable_associations 
 WHERE
 myvalue1=mytable1.myvalue1 AND myvalue2=mytable2.myvalue2
 OR
 myvalue1=mytable2.myvalue2 AND myvalue2=mytable1.myvalue1
) IS NULL;

(And then of course update mytable_associations with this new association)

Which, as you can see, could hugely benefit from some optimization.

(Sorry about the poor indentation in the code, I really don’t know how to indent mysql commands).

Can you help me?

(P.S. This is my first question ever posted here: Sure I’m doing lots of things wrong and I’d understand the consequent flaming, but please don’t be too hard on me 😉 )

Answer 1

Any solution involving order by rand() is going to be inefficient. For alternatives, see:

• My answer to Selecting random rows with MySQL
• Jan Kneschke on Order by RAND()

To exclude numbers you’ve already picked, here’s how I’d do it (this is pseudocode):

$c1 = SELECT COUNT(DISTINCT myvalue) FROM mytable
$c2 = SELECT COUNT(*) FROM mytable_associations

$offset = ROUND( RAND() * ($c1 * ($c1-1) - $c2) )

SELECT v.* FROM (
  SELECT LEAST(m1.myvalue,my2.myvalue) AS myvalue1,
    GREATEST(m1.myvalue,my2.myvalue) AS myvalue2
  FROM (SELECT DISTINCT myvalue FROM mytable) AS m1 
  INNER JOIN (SELECT DISTINCT myvalue FROM mytable) AS m2
    ON m1.myvalue <> m2.myvalue
) AS v
LEFT OUTER JOIN mytable_associations AS a USING (myvalue1,myvalue2)
WHERE a.myvalue1 IS NULL
LIMIT 1 OFFSET $offset

By ensuring that myvalue1 < myvalue2, and storing them in that order in mytable_associations, you can simplify the join.