没有已知列名的数据透视

Question

I have the following table with data:

A   B
=== ===
M   2  
M   3
M   5
N   5
N   2
O   6
P   13
P   7
P   9
P   11
P   3

Now I need a PIVOT (?) query to:

• Group by column A.
• Sort on column B.
• Take the lowest 3 values from column B.
• Put these values in the new columns B1, B2, B3

So the result will be:

A   B1   B2   B3
=== ==== ==== ====
M   2    3    5
N   2    5    null
O   6    null null
P   3    7    9

So far I have been trying to create queries with TOP, GROUP BY, PIVOT. I think the best way to go is with the PIVOT, but since I do not have a value I can use as a column name, I am stuck. Also, making a top-3 selection of these values seems also pretty challenging.

* EDIT *

Their is a unique constraint on columns A and B, so the values for B are always unique for the same A.

Answer 1

You can use the PIVOT function to get the result, but I would also implement a windowing function similar to row_number() to get the final result. The row_number() function will create a unique sequenced number for each B value if you partition the data over the A column. This sequenced number will be used as the new column names:

select a, B1, B2, B3
from
(
  select a, b,
    row_number() over(partition by a
                      order by b) seq
  from yourtable
) d
pivot
(
  max(b)
  for seq in ('1' as B1, '2' as B2, '3' as B3)
) piv;

See SQL Fiddle with Demo . This will give a result:

| A | B1 |     B2 |     B3 |
|---|----|--------|--------|
| M |  2 |      3 |      5 |
| N |  2 |      5 | (null) |
| O |  6 | (null) | (null) |
| P |  3 |      7 |      9 |

Answer 2

Here is an approach that works with both Oracle and SQL Server (the original question was tagged with both databases).

You can just enumerate the b values for each a using dense_rank() (or row_number() , but dense_rank() will give three different values if there are duplicates). Then pivot using conditional aggregation:

with ab as (
      select a, b,
             dense_rank() over (partition by a order by b asc) as seqnum
      from t
     )
select a,
       max(case when seqnum = 1 then b end) as b1,
       max(case when seqnum = 2 then b end) as b2,
       max(case when seqnum = 3 then b end) as b3
from ab
group by a;

EDIT:

The question is unclear on what to do if there are duplicates. For instance, if the data in b is 1, 1, 2, 3. Which should be the three columns?

1, 2, 3

or 1, 1, 2

The use of dense_rank() puts three different values into the columns (the first possible result). The use of row_number() would put the three smallest values (the second result). If there are never any duplicate values, then dense_rank() and row_number() produce the same results.