JESS: performance trade-of in updateObject(...)

[Florian Fischer]

Hello,

I was wondering what is the performance trade-off between
    rete.updateObject(myObject);
and
    rete.updateObject(myObject, slot);

When I update an object underlying a shadow fact, I record the list of
updated fields and call updateObject only at the end.  If only one field
was updated I pass the slot name as parameter.  If I updated more than
one, I make a global update.

But for instance, when updating 2 fields out of 91, would it make sense
to do two specific updateObject, one for each slot, instead of one
global update?  If I do a global updateObject, it will have to reread
all 91 fields.  If I do two specific updateObject, it might reorganize
the Rete network twice.  So which is better?

Would it make sense to implement a updateObject(String[] slots) for the
case where you know which fields were updated?

Thanks in advance,
Florian Fischer

--
Florian Fischer
Service d'Informatique Médicale
Hôpitaux Universitaires de Genève
Tél: [phone]
florian.fischer@...



--------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users you@...'
in the BODY of a message to majordomo@..., NOT to the list
(use your own address!) List problems? Notify owner-jess-users@....
--------------------------------------------------------------------

[Ernest Friedman-Hill]

The exact point at which one is better than the other would depend on  
the number of partial matches in memory. To check 91 slots for  
modification, Jess has to call 91 getter functions and do 91 value  
comparisons. If there are very few partial matches, you can easily  
imagine that the amount of computation needed to check all those slots  
would be less than that if you called updateObject twice with specific  
slot names. On the other hand, in the limit as working memory gets  
larger, those 91 comparisons could pale in comparison to what will be  
needed to reevaluate all the matches.

An update function that let you specify a list of property names could  
improve efficiency on the small-working-memory end of the scale; this  
is precisely the regime in which there isn't that much work going on  
anyway, so I don't think it's something to worry about. I will make a  
note though and we can see about adding it to a future release.


On May 15, 2008, at 10:43 AM, Florian Fischer wrote:

> Hello,
>
> I was wondering what is the performance trade-off between
>    rete.updateObject(myObject);
> and
>    rete.updateObject(myObject, slot);
>
> When I update an object underlying a shadow fact, I record the list of
> updated fields and call updateObject only at the end.  If only one  
> field
> was updated I pass the slot name as parameter.  If I updated more than
> one, I make a global update.
>
> But for instance, when updating 2 fields out of 91, would it make  
> sense
> to do two specific updateObject, one for each slot, instead of one
> global update?  If I do a global updateObject, it will have to reread
> all 91 fields.  If I do two specific updateObject, it might reorganize
> the Rete network twice.  So which is better?
>
> Would it make sense to implement a updateObject(String[] slots) for  
> the
> case where you know which fields were updated?
>
> Thanks in advance,
> Florian Fischer
>
> --
> Florian Fischer
> Service d'Informatique Médicale
> Hôpitaux Universitaires de Genève
> Tél: [phone]
> florian.fischer@...
>
>
>
> --------------------------------------------------------------------
> To unsubscribe, send the words 'unsubscribe jess-users  
> you@...'
> in the BODY of a message to majordomo@..., NOT to the list
> (use your own address!) List problems? Notify owner-jess-users@...
> .
> --------------------------------------------------------------------

---------------------------------------------------------
Ernest Friedman-Hill
Informatics & Decision Sciences          Phone: [phone]
Sandia National Labs                FAX:   [phone]
PO Box 969, MS 9012                 ejfried@...
Livermore, CA 94550                 http://www.jessrules.com




--------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users you@...'
in the BODY of a message to majordomo@..., NOT to the list
(use your own address!) List problems? Notify owner-jess-users@....
--------------------------------------------------------------------

[Florian Fischer]

Hello,

I never got a reply on that one.

In the meantime, after doing the change I described, I realized that my
rules don't all fire.  A simple rule like

(defrule MAIN::dpa-unit-reload-on-add
   (care-unit (initialLoad TRUE) (needLoadSermeds FALSE)
(needReloadPatients TRUE) (OBJECT ?ou))
   (service-activity (available TRUE) (service "SAMU"))
   =>
   (printout t "(jess) Initial load for unit " (call ?ou getId) crlf)
   (call ?ou reloadPatients))

doesn't fire for all care units.  After Jess is done running, I wonder
why some care units have not been loaded.  I re-enter the same rule
(copied from a ppdefrule of the rule itself in Jess) and the rule starts
firing.  One time I even had to enter the rule twice in a row before
having all care units processed.  And it is not just reloadPatient that
failed to change the flags.  The text "(jess) Initial load for unit XXX"
does not appear on the standard output.

Removing (slot-specific TRUE) from all my templates fixed that.  
Removing it from only care-unit did not.

Note that I no more use propertyChangeEvents to inform Jess of changes.  
I now use updateObject on one or all slots as I described below.  I
could have made a mistake somewhere but when I request the facts from
Jess, they look correct to me, so it is not an update problem.

I now deal with Jess problems this way.  A rule that acts funny is
removed from Jess and some code is written to supplement it outside of Jess.

Regards,
Florian

Florian Fischer wrote:

> Hello,
>
> I was wondering what is the performance trade-off between
>    rete.updateObject(myObject);
> and
>    rete.updateObject(myObject, slot);
>
> When I update an object underlying a shadow fact, I record the list of
> updated fields and call updateObject only at the end.  If only one
> field was updated I pass the slot name as parameter.  If I updated
> more than one, I make a global update.
>
> But for instance, when updating 2 fields out of 91, would it make
> sense to do two specific updateObject, one for each slot, instead of
> one global update?  If I do a global updateObject, it will have to
> reread all 91 fields.  If I do two specific updateObject, it might
> reorganize the Rete network twice.  So which is better?
>
> Would it make sense to implement a updateObject(String[] slots) for
> the case where you know which fields were updated?
>
> Thanks in advance,
> Florian Fischer
>

--
Florian Fischer
Service d'Informatique Médicale
Hôpitaux Universitaires de Genève
Tél: [phone]
florian.fischer@...



--------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users you@...'
in the BODY of a message to majordomo@..., NOT to the list
(use your own address!) List problems? Notify owner-jess-users@....
--------------------------------------------------------------------