Question

I have two tables that looks like this:

Class  Name  Score  Top
1      Amy    90     X
1      Ben    70     X
1      Chu    80     X
2      Don    60     X
2      Elf    65     X
2      Fez    75     X
2      Ges    35     X
2      Han    40     X

Class NumToppers
1      2
2      3

I want to find Top “NumToppers” from each class and accordingly update “Top” field:

Class  Name  Score  Top
1      Amy    90     Y
1      Ben    70     N
1      Chu    80     Y
2      Don    60     Y
2      Elf    65     Y
2      Fez    75     Y
2      Ges    35     N
2      Han    40     N

I have 100s of “Classes” in my real data. So, although data looks toyish, no toy solutions please.

Answer 1

Referring to this excellent page on selecting the ‘greatest-n-per-group’, here’s the query I came up with. In retrospect it is basically identical to @BassamMehanni’s answer, except that there is no ROW_NUMBER() function in MySQL.

This assumes you have tables class and toppers.

Solution:

Note: if you have a primary ID on the class table that isn’t the compound (Class,Name,Score) combination, use that instead for the join condition marked #@@.

set @class='';
set @rank=1;
UPDATE class         
LEFT JOIN 
  (SELECT Class,Name,Score,
       @rank:=if(@class=Class,@rank+1,1) as rank, 
       @class:=Class as dummy 
  FROM class ORDER BY Class,Score DESC) c
ON c.Class=class.Class AND c.Score=class.Score  #@@
   AND c.Name=class.Name                        #@@
LEFT JOIN toppers
ON c.Class=toppers.Class
SET Top = (CASE WHEN rank <= NumToppers THEN 'Y' ELSE 'N' END); 

Explanation

Basically, this query:

1. numbers the rows of class from top-to-bottom by score, within each class. That is, ranks each student within each class.
2. selects rows of class for which the rank is <= NumToppers, for each class.
3. Updates these.

For step 1, see the following (from the link I referred you):

set @class='';
set @rank=1;
SELECT Class,Name,Score, 
       @num:=if(@class=Class,@rank+1,1) as rank, 
       @class:=Class as dummy 
FROM class ORDER BY Class,Score DESC;

This looks through every row of class (after by sorting by class and descending score) and sets the rank to 1 if we’re on to a new class, or to rank+1 if we’re within the same class.

For step 2, we do a JOIN with toppers on class and pick the top NumToppers rows for each class:

set @class='';
set @rank=1;
SELECT *                                          # NEW
FROM toppers                                      # NEW
LEFT JOIN                                         # NEW
  (SELECT Class,Name,Score,                       #\
       @rank:=if(@class=Class,@rank+1,1) as rank, # |(same as step 1)
       @class:=class as dummy                     # |
  FROM class ORDER BY Class,Score DESC) c         #/ 
ON c.Class=toppers.Class                          # NEW
WHERE rank <= NumToppers;                         # NEW

Finally, we update on these conditions (step 3). However we have to do UPDATE class explicitly so we have to add in an extra JOIN of step 2 with class:

set @class='';
set @rank=1;
UPDATE class                                      # NEW
LEFT JOIN 
  (SELECT Class,Name,Score,
       @rank:=if(@class=Class,@rank+1,1) as rank, 
       @class:=class as dummy 
  FROM class ORDER BY Class,Score DESC) c
ON c.Class=class.Class AND c.Score=class.Score    # NEW (join condition)
   AND c.Name=class.Name                          # NEW (join condition)
LEFT JOIN toppers
ON c.Class=toppers.Class
SET top = (CASE WHEN rank <= NumToppers THEN 'Y' ELSE 'N' END); # NEW

If you have a primary ID on class that is preferrably not compound (ie a one-column ID) then join on that instead.