Rows into matrix

Question

Lets say I have a table defined as
create table factors
( datasource varchar(20),
v1 varchar2(10) ,
v2 varchar2(10) ,
value integer
The value column holds co-variance data. In other words the number of distinct values in v1 and v2 for given datasource is always the same. Lets say the distinct values of v1 and v2 are f1 ... f10. The number of rows would be ( 10* 9) / 2 ) + 10 = 55
I want to return a 10 x 10 matrix with all possible values filled in. What is the best way to do this with only one scan of data? Preferably done in PLSQL. It's pretty trivial in perl but I need to to hook this to some tool the can just run sql stmt or execute a stored proc.
Thanks!

Answer 1

That's why I was trying to write something a little more generic.
The problems you are going to face are:
a) What determines the order of the v1, v2 values in the matrix?. You'll see in my code, as far as I got, that I had to explicitly define the distinct v1/v2 values in order to get them out in the order you had specified. If the v1/v2 values can be different, how can a program or SQL know what order they should be in?
b) As you'll see in my example, I was dealing with 4 distinct v1/v2 values explicitly. If the requirement is to have anything from 2 upwards of these values then you're getting into the realms of dynamically creating your output. This is most likely going to be done in PL/SQL, however if you are specifically requiring the output to come out as queried rows and columns (i.e. to use in some other SQL) then you are going to struggle to get this dynamically for varying columns without writing multiple pipeline functions for the different number of columns (the number of rows doesn't matter so much as SQL's very flexible in that respect) or by having the output always producing the maximum possible columns with nulls in those that aren't used. I think there may be some possible means by using Objects (see How to pipeline a function with a dynamic number of columns? for an example of dynamically pipelining multiple columns), but that's outside the scope of my time to investigate for you)
Edit...
Ok here goes something dynamic (up to 4 distinct v1/v2 values, but easily expandable):-
SQL> CREATE TABLE t AS SELECT 'US' as pid, 'value' as v1, 'value' as v2, 1123 as val from dual union all
  2        select 'US', 'value', 'banks', 0.10 from dual union all
  3        select 'US', 'value', 'internet', 0.02 from dual union all
  4        select 'US', 'value', 'biotech', 0.055 from dual union all
  5        select 'US', 'banks', 'banks', 456 from dual union all
  6        select 'US', 'banks', 'internet', 2.33 from dual union all
  7        select 'US', 'banks', 'biotech', 4.33 from dual union all
  8        select 'US', 'internet', 'internet', 1.1111 from dual union all
  9        select 'US', 'internet', 'biotech', 0.9099 from dual union all
10        select 'US', 'biotech', 'biotech', 7.7777 from dual union all
11        select 'UK', 'value', 'value', 1 from dual union all
12        select 'UK', 'value', 'internet', 2 from dual union all
13        select 'UK', 'value', 'agents', 3 from dual union all
14        select 'UK', 'internet', 'internet', 4 from dual union all
15        select 'UK', 'internet', 'agents', 5 from dual union all
16        select 'UK', 'agents', 'agents', 6 from dual
17  /
Table created.
SQL>
SQL> create type matrix_rec as object (lbl varchar2(10)
  2                                   ,val1 varchar2(10)
  3                                   ,val2 varchar2(10)
  4                                   ,val3 varchar2(10)
  5                                   ,val4 varchar2(10))
  6  /
Type created.
SQL>
SQL> create type matrix_row as table of matrix_rec
  2  /
Type created.
SQL>
SQL> create or replace function show_matrix(p_id IN VARCHAR2) RETURN matrix_row PIPELINED IS
  2    CURSOR val(y IN VARCHAR2,x IN VARCHAR2) IS
  3      SELECT DISTINCT val
  4      FROM t
  5                             WHERE pid = p_id
  6      AND   ((v1 = y AND v2 = x) OR (v1 = x AND v2 = y));
  7
  8    v   SYS_REFCURSOR;
  9    x   SYS_REFCURSOR;
10    y   SYS_REFCURSOR;
11    o   matrix_rec := matrix_rec(null, null, null, null, null);
12    l   VARCHAR2(10);
13    r   NUMBER;
14    ly  VARCHAR2(10);
15    lx  VARCHAR2(10);
16    xyv NUMBER;
17  begin
18    -- put out a heading
19    o.lbl := p_id;
20    OPEN v FOR 'SELECT rownum rn, v1 FROM (SELECT DISTINCT v1 FROM t WHERE pid = :p_id ORDER BY v1)' USING p_id;
21    LOOP
22      FETCH v INTO r, l;
23      EXIT WHEN v%NOTFOUND;
24      CASE r
25        WHEN 1 THEN o.val1 := l;
26        WHEN 2 THEN o.val2 := l;
27        WHEN 3 THEN o.val3 := l;
28        WHEN 4 THEN o.val4 := l;
29      ELSE NULL;
30      END CASE;
31    END LOOP;
32    CLOSE v;
33    PIPE ROW (o);
34
35    -- now output the rows of data
36    OPEN y FOR 'SELECT DISTINCT v1 FROM t WHERE pid = :p_id ORDER BY v1' USING p_id;
37    LOOP
38      FETCH y INTO ly;
39      EXIT WHEN y%NOTFOUND;
40      OPEN x FOR 'SELECT rownum rn, v1 FROM (SELECT DISTINCT v1 FROM t WHERE pid = :p_id ORDER BY v1)' USING p_id;
41      LOOP
42        FETCH x INTO r, lx;
43        EXIT WHEN x%NOTFOUND;
44        SELECT DISTINCT val
45        INTO xyv
46        FROM t
47                             WHERE pid = p_id
48        AND   ((v1 = ly AND v2 = lx) OR (v1 = lx AND v2 = ly));
49        o.lbl := ly;
50        CASE r
51                   WHEN 1 THEN o.val1 := xyv;
52                   WHEN 2 THEN o.val2 := xyv;
53                   WHEN 3 THEN o.val3 := xyv;
54                   WHEN 4 THEN o.val4 := xyv;
55                 ELSE NULL;
56                 END CASE;
57      END LOOP;
58      CLOSE x;
59      PIPE ROW (o);
60    END LOOP;
61    CLOSE y;
62  end;
63  /
Function created.
SQL> select * from table(show_matrix('US'));
LBL        VAL1       VAL2       VAL3       VAL4
US         banks      biotech    internet   value
banks      456        4.33       2.33       .1
biotech    4.33       7.7777     .9099      .055
internet   2.33       .9099      1.1111     .02
value      .1         .055       .02        1123
SQL> select * from table(show_matrix('UK'));
LBL        VAL1       VAL2       VAL3       VAL4
UK         agents     internet   value
agents     6          5          3
internet   5          4          2
value      3          2          1
SQL>Message was edited by:
blushadow