Funnel steps in SQL query

Question

I have config table as
create table config (sr_no integer, step varchar2(30), step_id integer);
insert into config values (1,'step1', 111);
insert into config values (2,'step8', 222);
and i have detail as
create table user_steps (pk integer,stepsdate date,step_id integer, usern varchar2(30));
insert into user_steps values(11,sysdate,333,'user1');
insert into user_steps values(12,sysdate-1,222,'user1');
insert into user_steps values(13,sysdate-2,111,'user1');
insert into user_steps values(14,sysdate-1,111,'user1');
insert into user_steps values(15,sysdate-2,444,'user1');
insert into user_steps values(16,sysdate-3,333,'user1');
insert into user_steps values(17,sysdate-4,111,'user1');
Now i want to find all the records when
1. user1 did step1 followed by step8 with no step in between
2nd scenario
1. user1 did step1 and then step8 with any step in between
Like steps in a funnel. Any hwlp or sample query would be very helpful. I have spent whole day but just cant get started :(
I have to do this in a SQL query ( no PL/SQL)

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statements; that's very helpful.
You need to post the results you want from that sample daya, as well. Your description could apply equally well to several different things.
Whatever you want, it sounds like analytic functions will be helpful.
For example, problem 1. might be solved like this:
WITH     got_next_step     AS
     SELECT     u.pk
     ,     u.stepsdate
     ,     u.usern
     ,     c.step
     ,     LEAD (c.step) OVER ( PARTITION BY  u.usern     -- See note below
                         ORDER BY        u.stepsdate
                         ,             u.pk
                       )     AS next_step
     FROM          user_steps     u
     LEFT OUTER JOIN     config          c     ON     c.step_id     = u.step_id
     WHERE     usern     = 'user1'
SELECT     *     -- or list columns you want
FROM     got_next_step
WHERE     step          = 'step1'
AND     next_step     = 'step8'
;The analytic PARTITION BY is not needed (and might be wasting a little time) if you're only doing one usern per query, as in your example. It is needed if you do multiple userns in the same query.
I see that stepsdate is not unique for a given usern. When two (or more) rows have exactly the same usern and stepsdate, then the query above considers the rows with higher pks as being later.
"Like steps in a funnel" sounds like a good analogy, but I don't understand it. Could you explain what you mean?
Edited by: Frank Kulash on May 6, 2010 8:38 AM