mysql: Finding a max in each column of a table - mysql

Question

I have the table below. I'm looking for the both the max value in each column AND it's matching username (all values of NULL to be ignored).
A bunch of mad googling has lead me to believe I need to find the max values and then use a second query to find the matching username?
But is there a query that can return this in one go?
ID username Vale Jorge Andrea
-------------------------------------------
01 John 2 6 NULL
02 Ted NULL 0 0
03 Marcy NULL 2 1
Output would be...
John Jorge 6
John Vale 2
Marcy Andrea 1

Answer 1

There's different ways of looking at it, here's a table that gives a row for each username that has a matching max value:
SELECT
username
, IF (max_vale = t.vale, max_vale, NULL) AS for_vale
, IF (max_jorge = t.jorge, max_jorge, NULL) AS for_jorge
, IF (max_andrea = t.andrea, max_andrea, NULL) AS for_andrea
FROM (
SELECT
MAX(vale) AS max_vale
, MAX(jorge) AS max_jorge
, MAX(andrea) AS max_andrea
FROM t
) y
JOIN t ON (
t.vale = max_vale
OR t.jorge = max_jorge
OR t.andrea = max_andrea
)
http://sqlfiddle.com/#!9/58e37d/5
This gives:
username for_vale for_jorge for_andrea
----------------------------------------------
John 2 6 (null)
Marty (null) (null) 1
Basically, all I'm doing is selecting the specific column max values, then using that query as the source for another query that just looks at the MAX generated columns, and filters (IF()) based on the matches found.

Answer 2

This is one reasonably simple way of doing it... make a table of all the maximum values and join it to the usernames on matching the value with the max. Uses the fact that NULL isn't equal to anything. I haven't attempted to order the results but that is easy enough with adding an ORDER BY clause.
select username, name, COALESCE(mv.vale, mv.jorge, mv.Andrea) as value
from table1
join
(select 'Vale' as name, max(vale) as vale, NULL as jorge, NULL as andrea from table1
union ALL
select 'Jorge', NULL, max(jorge), NULL from table1
union all
select 'Andrea', NULL, NULL, max(andrea) from table1) mv
on table1.vale = mv.vale or table1.jorge = mv.jorge or table1.andrea = mv.andrea
Output
username name value
John Vale 2
John Jorge 6
Marcy Andrea 1
To extrapolate this to more columns is reasonably straightforward (if somewhat painful) e.g. to add a column called fred you would use (changes inside **):
select username, name, COALESCE(mv.vale, mv.jorge, mv.Andrea**, mf.fred**) as value
from table1
join
(select 'Vale' as name, max(vale) as vale, NULL as jorge, NULL as andrea**, NULL as fred** from table1
union ALL
select 'Jorge', NULL, max(jorge), NULL**, NULL** from table1
union all
select 'Andrea', NULL, NULL, max(andrea)**, NULL** from table1) mv
**union all
select 'Fred', NULL, NULL, max(fred), NULL from table1) mf**
on table1.vale = mv.vale or table1.jorge = mv.jorge or table1.andrea = mv.andrea **or table1.fred = mf.fred**
If you have access to stored procedures, you can also do it like this (in a much more flexible way in terms of columns)
DROP PROCEDURE IF EXISTS list_maxes;
DELIMITER //
CREATE PROCEDURE list_maxes(tname VARCHAR(20), column_list VARCHAR(1000))
BEGIN
DECLARE maxv INT DEFAULT 0;
DECLARE cpos INT;
DECLARE colname VARCHAR(20);
-- loop through the column names
WHILE (LENGTH(column_list) > 0)
DO
SET cpos = LOCATE(',', column_list);
IF (cpos > 0) THEN
SET colname = LEFT(column_list, cpos - 1);
SET column_list = SUBSTRING(column_list, cpos + 1);
ELSE
SET colname = column_list;
SET column_list = '';
END IF;
-- find the maximum value of this column
SET #getmax = CONCAT('SELECT MAX(', colname, ') INTO #maxv FROM Table1');
PREPARE s1 FROM #getmax;
EXECUTE s1;
DEALLOCATE PREPARE s1;
-- now find the user with the maximum value
SET #finduser = CONCAT("SELECT username, '", colname, "' AS name, ", colname, ' AS value FROM ', tname,' WHERE ', colname, ' = ', #maxv);
PREPARE s2 FROM #finduser;
EXECUTE s2;
DEALLOCATE PREPARE s2;
END WHILE;
END//
DELIMITER ;
CALL list_maxes('table1', 'Vale,Jorge,Andrea')
Output
John Vale 2
John Jorge 6
Marcy Andrea 1

Answer 3

This is quite long but working. I combine all columns into one table using UNION ALL. Then get the max value per surname. Join this to the original table using the surname and value. Order by value in descending order.
select tv.*
from( select surname, max(val) as maxval
from (
select username,'vale' as surname,vale as val
from tbl
union all
select username,'jorge' as surname,jorge
from tbl
union all
select username,'andrea' as surname,andrea
from tbl) tab
group by surname) tt
join (
select username,'vale' as surname,vale as val
from tbl
union all
select username,'jorge' as surname,jorge
from tbl
union all
select username,'andrea' as surname,andrea
from tbl) tv
on tt.surname=tv.surname and tt.maxval=tv.val
order by tv.val desc;

Answer 4

As a whole other way to resolve the request, we may look at the model, e.g., the fact that what appears to be domain content is represented as columns instead of rows. This is typical of a source aggregate query (pivot or rollup to get aggregated totals or groupings, for instance), but if the underlying data spreads out, should perhaps be based on the transactional integrity of that data source (the "spread out" underlying source).
Basically, I wonder why there are Vale, Jorge and Andrea columns at all in the database. This implies it's already been summarized.
So we may look at an alternate model that is notably easier to navigate for these purposes:
CREATE TABLE IF NOT EXISTS `user` (
`id` MEDIUMINT NOT NULL AUTO_INCREMENT,
`username` varchar(20) NOT NULL,
PRIMARY KEY (`id`)
) DEFAULT CHARSET=utf8;
CREATE TABLE IF NOT EXISTS `prospect` (
`id` MEDIUMINT NOT NULL AUTO_INCREMENT,
`name` varchar(20) NOT NULL,
PRIMARY KEY (`id`)
) DEFAULT CHARSET=utf8;
CREATE TABLE IF NOT EXISTS `aggregate` (
`id` MEDIUMINT NOT NULL AUTO_INCREMENT,
`user_id` int(6) unsigned NOT NULL,
`prospect_id` int(6) unsigned NOT NULL,
`total` int(6) unsigned NOT NULL,
PRIMARY KEY (`id`)
) DEFAULT CHARSET=utf8;
SELECT
user.username
, prospect.name
, MAX(aggregate.total) AS max_aggregate
FROM aggregate
JOIN user ON user_id = user.id
JOIN prospect ON prospect_id = prospect.id
GROUP BY username
This produces:
John Andrea 6
Marty Jorge 2
Ted Jorge 5
http://sqlfiddle.com/#!9/07ba0c/1
This may not be useful to you now, but as your experience grows and your experience with advanced querying evolves, this will make more sense. The main difficulty may be that the core data is already turned, making the querying more difficult because what you want is a different dimension than what you may have already derived.