[rules-users] Pattern aggregation

[Tim 4076]

What do you guys think of this: my facts are all the same type, so I didn't
have anything like a consumer object to break things down with. But what I
did was:

rule "group 1"
when
  $a : Trans( $groupValue : prodCat, $date : date )
  not Trans( this != $a, prodCat == $groupValue, date < $date )
  $b : LinkedList( size >= 1 ) from collect ( Trans( this != $a, prodCat ==
$groupValue, date > $date ) )
then
  //do something
end

This is quite quick. The rule grabs all the Trans objects that have the same
prodCat, in a single firing, without the need to iteractively retract things
or fire multiple times for the same group value. As the first condition only
matches a single fact (the one with the oldest date), there is only ever a
single permutation that can fulfill the conditions.

This is only aggregating on a single attribute, but the principal should
work with more.



On 18 October 2010 16:12, Wolfgang Laun <wolfgang.laun at gmail.com> wrote:

> Doing s.th. like
>   $t1 : Trans( $id : id, $pc : pc, $tt : tt )
>   $t2 : Trans( this != $t1, id == $id, pc == $pc, tt < ($tt + 3600) )
> is bound to produce poor performanc.
>
> Divide and conquer!
>
> You might start with a Consumer record
>   Consumer( $id : id )
>   $t1 : Trans( id == $id: , $pc : pc, $tt : tt )
>   $t2 : Trans( this != $t1, id == $id, pc == $pc, tt < ($tt + 3600) )
>
> You might run an (external) sort on the Transaction records and
> process it in batches of identical id+pc.
>
> If transaction times don't go around the clock, you might sort by
> date, and process day by day.
>
> You may have to create a Domain Specific Language for the
> non-programmers, putting a firm rein on how they combine the basic
> facts. Processing large batches is bound to require skills they just
> dont have.
>
> -W
>
>
> On 18 October 2010 16:14, Greg Barton <greg_barton at yahoo.com> wrote:
> > It would be nice if we had an example of some rules.  That way we can
> rule out obvious performance killers like cartesian products and multiple
> "from" clauses in one rule.
> >
> > GreG
> >
> > On Oct 18, 2010, at 5:19, Tim 4076 <tm4076 at gmail.com> wrote:
> >
> > Hi,
> > I'm trying to use drools to do grouping of data according to patterns
> defined in my rules, but I'm having issues creating something that works in
> a reasonable amount of time (seconds). I've tried all sorts of permutations
> without much luck and would like to hear how others would do the same thing.
> >
> > To give an example: I've got a big batch of transaction records and I
> want to aggregate all the records where the consumer id and product category
> are the same and the purchases were made within an hour of each other.
> >
> > The fact that its matching the same values between facts, rather than
> against constants seems to scupper it somewhat.
> >
> > I would go down the ETL route, but the idea is for non-techies to define
> their own aggregations using rules.
> >
> > -Cheers. Tim
> >
> >
> > _______________________________________________
> > rules-users mailing list
> > rules-users at lists.jboss.org
> > https://lists.jboss.org/mailman/listinfo/rules-users
> >
> >
> >
> >
> > _______________________________________________
> > rules-users mailing list
> > rules-users at lists.jboss.org
> > https://lists.jboss.org/mailman/listinfo/rules-users
> >
>
> _______________________________________________
> rules-users mailing list
> rules-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/rules-users/attachments/20101019/e1663534/attachment.html