如何在SQL的另一列中检索与最大值对应的值？

Question

I have the following table, which represents valuations of items.

ITEM             REFERENCEDATE     VALUATION
------------------------------------------------
A                25/01/2012        25.35
A                26/01/2012        51.35
B                25/01/2012        25.00

Edit: (ITEM, REFERENCEDATE) is a unique index.

The goal is to get the latest valuations for a set of item. Which means i'm trying to create a SQL request that would return something like

ITEM             REFERENCEDATE     VALUATION
------------------------------------------------
A                26/01/2012        51.35
B                25/01/2012        25.00

Flowing a tutorial on GROUP BY, I ended up trying

SELECT A.ITEM, A.VALUATION, MAX(A.REFERENCEDATE)
FROM VALUATIONS A
GROUP BY A.ITEM

Full of hope that the SQL server would understand that I need A.VALUATION for the line which realizes the max for A.REFERENCEDATE for the ITEM represented on the current result line.

But instead, I have this unpleasant error message:

Column 'VALUATIONS.VALUATION' is invalid in the select list because it is not contained 
in either an aggregate function or the GROUP BY clause.

How can I indicate that the VALUATION where the maximum of REFERENCEDATE is reached should be used ?

Note: I need a solution that works at least on Oracle and SQL Server

EDIT: Thanks everybody for your help. I was stuck in a hole try to get away with only one single SELECT ... GROUP BY request. Now I see there are two approaches that articulate around the same idea:

1. Making a JOIN with the result of an other independant request that will return all the item/max(date) couples
2. Using a subrequest result in the where clause which will have a different value for each item.

Could anybody provide a reason (or a pointer to a reason) to prefer one to the other ?

Answer 1

Select V.Item, V.ReferenceDate, V.Valuation
From Valuations As V
Where V.ReferenceDate = (
                        Select Max(V1.ReferenceDate)
                        From Valuations As V1
                        Where V1.Item = V.Item
                        )

SQL Fiddle version

In response to your edit, the only way to know for sure which approach will perform better is to evaluate the execution plan on each of the queries. There are many factors that can come into determining the fastest approach and certainly the DBMS itself is one of those factors. A good query engine should be able to deduce the same or similar execution plan regardless of the approach. That said, using a derived table (ie approach #1) may be a bit more explicit to the query engine (even if less explicit to the reader of the query) and thus might perform better. Often it is the case that derived tables perform better than correlated subqueries (my solution and your approach #2). However, I wouldn't alter the approach until I had evidence to support the change. Again, the only way to know which will perform better for certain is to evaluate the execution plan against your data.

Answer 2

If you are using almost any database other than MySQL, then answer is to use ranking functions. In particular, row_number does what you are looking for:

select ITEM, REFERENCEDATE, VALUATION
from (select t.*
             row_number() over (partition by item order by referencedate desc) as seqnum
      from t
     ) t
where seqnum = 1 and
      item in (<your list of items>)

Row number assigns a sequence nubmer to the records for each item. It starts at 1 for the biggest reference date and then 2 for the next biggest and so on (based on the order by clause). You want the first one, where seqnum = 1.

Answer 3

select a.item, a.valuation, a.referencedate
from valuations a
  join (select a2.item, max(referencedate) as max_date
        from valuations a2
        group by a2.item
  ) b ON a.item = b.item and a.referencedate = b.max_date

Answer 4

Try this:

SELECT A.ITEM, MAX(A.VALUATION), A.REFERENCEDATE
FROM VALUATIONS A
JOIN 
(
    SELECT A.ITEM, MAX(A.REFERENCEDATE) AS REFERENCEDATE
    FROM VALUATIONS A
    GROUP BY A.ITEM
) B ON A.ITEM = B.ITEM AND A.REFERENCEDATE = B.REFERENCEDATE
GROUP BY A.ITEM, A.REFERENCEDATE

It will select the MAX value from the columns holding the max(REFERENCEDATE). If you only expect one column to have the max, then it would simply select from the one it can choose from.

Answer 5

This is the code you possibly need:

Select  *
    From    ItemValues      As  A
    Inner   Join
        ItemValues      As  MaxValuedItem
    On  MaxValuedItem.Id    =   (
        Select  Top 1
            B.Id
            From    ItemValues  As  B
            Where   B.Item_Id   =   A.Item_Id
            Order   By      B.Valuation Desc
    )

You need to use a "join" with the table itself that refers to the record that has the maximum value for the same item .