I'm using the following query to extract a random row from the ones with the minimum date difference within a subgroup (as defined by an ID1,ID2 pair).
It is not working, as the steps that ranks the rows by date difference assigns rank=1 only to the first observation (i.e. even if the second has the same date difference and should receive a rank = 1 too).
Somehow I need to change the WHEN clause to something like...
CASE
WHEN @id1 = t.id1 AND @id2 = t.id2 AND @diff != diff THEN @rownum := @rownum + 1
ELSE @rownum := 1
but when I do that it doesn't find the diff field...
DROP TABLE IF EXISTS temp4;
CREATE TABLE temp4 AS
SELECT x.id1,
x.id2,
x.YEAR,
x.MMDD,
x.id3,
x.id3_YEAR,
x.id3_MMDD,
x.rk
FROM (SELECT t.*,
ABS(DATEDIFF(CONCAT(CAST(t.id3_YEAR AS CHAR(4)),'-', LEFT(t.id3_MMDD,2),'-',RIGHT(t.id3_MMDD,2)),
CONCAT(CAST(t.YEAR AS CHAR(4)),'-', LEFT(t.MMDD,2),'-',RIGHT(t.MMDD,2)))) AS diff,
CASE
WHEN @id1 = t.id1 AND @id2 = t.id2 THEN @rownum := @rownum + 1
ELSE @rownum := 1
END AS rk,
@id1 := t.id1,
@id2 := t.id2
FROM temp3 t
JOIN (SELECT @rownum := 0, @id1 := 0, @id2 := 0) r
ORDER BY t.id1, t.id2, diff, RAND()) x;
I'm pasting the SQL dump
DROP TABLE IF EXISTS `temp3`;
CREATE TABLE IF NOT EXISTS `temp3` (
`id1` char(7) NOT NULL,
`id2` char(7) NOT NULL,
`YEAR` year(4) NOT NULL,
`MMDD` char(4) NOT NULL,
`id3` char(7) NOT NULL,
`id3_YEAR` year(4) NOT NULL,
`id3_MMDD` char(4) NOT NULL
) ENGINE=MyISAM DEFAULT CHARSET=latin1;
INSERT INTO `temp3` VALUES('1', '2', 1992, '0107', '55', 1991, '0528');
INSERT INTO `temp3` VALUES('1', '2', 1992, '0107', '57', 1991, '0701');
INSERT INTO `temp3` VALUES('1', '3', 1992, '0107', '88', 2000, '0101');
INSERT INTO `temp3` VALUES('1', '3', 1992, '0107', '44', 2000, '0101');
This version is actually really close to what I need
DROP TABLE IF EXISTS temp4;
CREATE TABLE temp4 AS
SELECT
x.id1,
x.id2,
x.YEAR,
x.MMDD,
x.id3,
x.id3_YEAR,
x.id3_MMDD
FROM
( SELECT
t.*,
@rownum := CASE
WHEN @id1 = t.id1 AND @id2 = t.id2 AND @diff != diff THEN @rownum + 1
ELSE 1
END AS rk,
@id1 := t.id1,
@id2 := t.id2,
@diff := t.diff
FROM
( SELECT
t.*,
ABS(DATEDIFF(CONCAT(CAST(t.id3_YEAR AS CHAR(4)),'-', LEFT(t.id3_MMDD,2),'-',RIGHT(t.id3_MMDD,2)),
CONCAT(CAST(t.YEAR AS CHAR(4)),'-', LEFT(t.MMDD,2),'-',RIGHT(t.MMDD,2)))) AS diff
FROM temp3 t
ORDER BY t.id1, t.id2, diff, RAND()
) t,
( SELECT @rownum := 0, @id1 := null, @id2 := null, @diff := null ) r
) x
WHERE x.rk = 1;