Select highest 3 scores in each day for every user

Question

I have a MYSQL table like this:

  id |  userid  |  score  |      datestamp      |
-----------------------------------------------------
  1  |    1     |   5         


        

Answer 1

Please take a look at the following code, if your answer to my comment is yes :) Since your data all in 2012, and month of november, I took day.

• SQLFIDDLE sample

Query:

select y.id, y.userid, y.score, y.datestamp 
from (select id, userid, score, datestamp 
      from scores
      group by day(datestamp)) as y    
where (select count(*) 
       from (select id, userid, score, datestamp
             from scores group by day(datestamp)) as x
       where y.score >= x.score
       and y.userid = x.userid
      ) =1 -- Top 3rd, 2nd, 1st    
order by y.score desc
;

Results:

ID  USERID  SCORE   DATESTAMP
8   2       8.5 December, 07 2012 00:00:00+0000
20  3       6   December, 08 2012 00:00:00+0000
1   1       5   December, 06 2012 00:00:00+0000

Based on your latter updates to question. If you need some per user by year/month/day and then find highest, you may simply add aggregation function like sum to the above query. I am reapeating myself, since your sample data is for just one year, there's no point group by year or month. That's why I took day.

select y.id, y.userid, y.score, y.datestamp 
from (select id, userid, sum(score) as score,
      datestamp 
from scores
group by userid, day(datestamp)) as y    
where (select count(*) 
from (select id, userid, sum(score) as score
      , datestamp
from scores
group by userid, day(datestamp)) as x
where y.score >= x.score
and y.userid = x.userid
) =1 -- Top 3rd, 2nd, 1st    
order by y.score desc
;

Results based on sum:

ID  USERID  SCORE   DATESTAMP
1   1       47.5    December, 06 2012 00:00:00+0000
8   2       16      December, 07 2012 00:00:00+0000
20  3       6       December, 08 2012 00:00:00+0000

UPDATED WITH NEW SOURCE DATA SAMPLE

Simon, please take a look at my own sample. As your data was changing, I used mine. Here is the reference. I have used pure ansi style without any over partition or dense_rank. Also note the data I used are getting top 2 not top 3 scores. You can change is accordingly.

Guess what, the answer is 10 times simpler than the first impression your first data gave....

SQLFIDDLE

Query to 1: -- for top 2 sum by user by each day

SELECT userid, sum(Score), datestamp
FROM scores t1
where 2 >=
(SELECT count(*) 
 from scores t2
 where t1.score <= t2.score
 and t1.userid = t2.userid
 and day(t1.datestamp) = day(t2.datestamp)
 order by t2.score desc)
group by userid, datestamp 
;

Results for query 1:

USERID  SUM(SCORE)  DATESTAMP
1       70      December, 06 2012 00:00:00+0000
1       30      December, 07 2012 00:00:00+0000
2       22      December, 06 2012 00:00:00+0000
2       25      December, 07 2012 00:00:00+0000
3       30      December, 06 2012 00:00:00+0000
3       30      December, 07 2012 00:00:00+0000

Final Query: -- for all two days top 2 sum by user

SELECT userid, sum(Score)
FROM scores t1
where 2 >=
(SELECT count(*) 
 from scores t2
 where t1.score <= t2.score
 and t1.userid = t2.userid
 and day(t1.datestamp) = day(t2.datestamp)
 order by t2.score desc)
group by userid
;

Final Results:

USERID  SUM(SCORE)
1      100
2      47
3      60

Here goes a snapshot of direct calculations of data I used.