Efficient assignment of percentile/rank in MYSQL

问题

I have a couple of very large tables (over 400,000 rows) that look like the following:

+---------+--------+---------------+
| ID      | M1     | M1_Percentile |
+---------+--------+---------------+
| 3684514 | 3.2997 | NULL          |
| 3684515 | 3.0476 | NULL          |
| 3684516 | 2.6499 | NULL          |
| 3684517 | 0.3585 | NULL          |
| 3684518 | 1.6919 | NULL          |
| 3684519 | 2.8515 | NULL          |
| 3684520 | 4.0728 | NULL          |
| 3684521 | 4.0224 | NULL          |
| 3684522 | 5.8207 | NULL          |
| 3684523 | 6.8291 | NULL          |
+---------+--------+---------------+...about 400,000 more

I need to assign each row in the M1_Percentile column a value that represents "the percent of rows with M1 values equal or lower to the current row's M1 value"

In other words, I need:

I implemented this sucessfully, but it is FAR FAR too slow. If anyone could create a more efficient version of the following code, I would really appreciate it!

UPDATE myTable AS X JOIN (
SELECT
  s1.ID, COUNT(s2.ID)/ (SELECT COUNT(*) FROM myTable) * 100 AS percentile
FROM
  myTable s1 JOIN myTable s2 on (s2.M1 <= s1.M1)
GROUP BY s1.ID
ORDER BY s1.ID) AS Z 
ON (X.ID = Z.ID) 
SET X.M1_Percentile = Z.percentile;

This is the (correct but slow) result from the above query if the number of rows is limited to the ones you see (10 rows):

+---------+--------+---------------+
| ID      | M1     | M1_Percentile |
+---------+--------+---------------+
| 3684514 | 3.2997 |            60 |
| 3684515 | 3.0476 |            50 |
| 3684516 | 2.6499 |            30 |
| 3684517 | 0.3585 |            10 |
| 3684518 | 1.6919 |            20 |
| 3684519 | 2.8515 |            40 |
| 3684520 | 4.0728 |            80 |
| 3684521 | 4.0224 |            70 |
| 3684522 | 5.8207 |            90 |
| 3684523 | 6.8291 |           100 |
+---------+--------+---------------+

Producing the same results for the entire 400,000 rows takes magnitudes longer.

回答1:

I cannot test this, but you could try something like:

update table t
set mi_percentile = (
    select count(*)
    from table t1
    where M1 < t.M1 / (
        select count(*)
        from table));

UPDATE:

update test t
set m1_pc = (
    (select count(*) from test t1 where t1.M1 < t.M1) * 100 /
    ( select count(*) from test));

This works in Oracle (the only database I have available). I do remember getting that error in MySQL. It is very annoying.

回答2:

Fair warning: mysql isn't my native environment. However, after a little research, I think the following query should be workable:

UPDATE myTable AS X 
JOIN ( 
  SELECT  X.ID, (
      SELECT  COUNT(*)
      FROM    myTable X1
      WHERE   (X.M1, X.id) >= (X1.M1, X1.id) as Rank)
  FROM myTable as X
) AS RowRank
ON (X.ID = RowRank.ID)
CROSS JOIN (
  SELECT COUNT(*) as TotalCount 
  FROM myTable 
) AS TotalCount
SET X.M1_Percentile = RowRank.Rank / TotalCount.TotalCount;

来源：https://stackoverflow.com/questions/7086751/efficient-assignment-of-percentile-rank-in-mysql

标签

mysql

sql

database

rank