Querying wide (multi-record case) table data in oracle 10g

Question

I've got a set of "wide" (multi-record case data format) tables similar to the format described in Oracle's "Oracle Data Mining Concepts" book in chapter 2 (http://download-west.oracle.com/docs/cd/B14117_01/datamine.101/b10698/2data.htm#1010394).
I like the flexibility of this format, especially since it can handle thousands of attributes. But the main problem I have with this format is in determining how to write efficient, complex queries against the data.
I've included some sample data below to demonstrate my problem. I've got two tables, one with people and the other with the attributes and values of age, income, height and weight. If I try to generate a complex query against this data, I seem to either have to rely on multisets and functions, or on a large number of self joins. Neither of which seems to scale well or perform well for large amounts of data. I know I could also add a pipelined TABLE function to filter the data more, but none of these options seem like great solutions. Does anyone have any ideas for a better way to form complex well performing queries against this type of data layout?
For example, with my sample data below I'd like to select everyone whose (income = 2000) or (income = 50000 and weight = 210), but the only way I see of doing this is with a query like the following (which won't take advantage of any indexes on the data):
select * from people_attributes_view
where find_eq_attr(AttributeValue('income',2000),attribute_values) is not null
or (find_eq_attr(AttributeValue('income',50000),attribute_values) is not null
and
find_eq_attr(AttributeValue('weight',210),attribute_values) is not null);
Any help is greatly appreciated.
Thanks.
-- My full example starts here ----------------
-- Create the sample tables and data
create table people (id int, name varchar(30));
create table attributes (id int, attribute varchar(10), value NUMBER);
insert into people values (1,'tom');
insert into people values (2,'jerry');
insert into people values (3,'spike');
commit;
insert into attributes values (1,'age',23);
insert into attributes values (1,'income',1000);
insert into attributes values (1,'height',5);
insert into attributes values (1,'weight',120);
insert into attributes values (2,'age',20);
insert into attributes values (2,'income',2000);
insert into attributes values (3,'age',30);
insert into attributes values (3,'income',50000);
insert into attributes values (3,'weight',210);
commit;
-- Create some types, functions and views for the search
CREATE OR REPLACE TYPE AttributeValue AS OBJECT
(attribute varchar(30),
value NUMBER);
CREATE OR REPLACE TYPE AttributeValues AS TABLE OF AttributeValue;
create or replace function find_eq_attr (val AttributeValue, vals AttributeValues) RETURN AttributeValue IS
begin
for i in vals.FIRST .. vals.LAST
LOOP
if ((val.attribute = vals(i).attribute) and
(val.value = vals(i).value)) then
return vals(i);
end if;
END LOOP;
return null;
end;
create or replace view people_attributes_view as
select p.name, p.id,
cast(multiset(select attribute,value from attributes a where a.id = p.id)
as AttributeValues) attribute_values
from people p;
-- Search for everyone whose (income = 2000) or (income = 50000 and weight = 210)
select * from people_attributes_view
where find_eq_attr(AttributeValue('income',2000),attribute_values) is not null
or (find_eq_attr(AttributeValue('income',50000),attribute_values) is not null
and
find_eq_attr(AttributeValue('weight',210),attribute_values) is not null);

Answer 1

I like the flexibility of this format, especially
since it can handle thousands of attributes. But the
main problem I have with this format is in
determining how to write efficient, complex queries
against the data. Can't be done. The flexibility is achieved at the cost of simplicity and performance, it is a trade off.
See this thread for a full discussion on the subject.
http://asktom.oracle.com/pls/ask/f?p=4950:8:::::F4950_P8_DISPLAYID:10678084117056