[Pellet-users] performance difference between pellet command line and pellet-protege

[Evren Sirin]

On 7/15/07 9:06 PM, howard goldberg wrote:
> 1.5rc1.
> I rebuilt pellet from the zip distribution and pointed p4 to use the 
> newly built version so everything is running from the same build.

It is hard to say something conclusive without seeing the ontology. One 
difference between the command line and Protege-Pellet binding is the 
former uses Jena interface and the latter uses the OWL-API interface. 
Most of the time the reasoning performance you get from either interface 
is very similar. One thing that happens for some OWL-Full ontologies is 
that if you have some datatype properties attached to classes then using 
punning you can assume that the class URI is also used as as individual 
and treat it as an ABox assertion or assume the datatype property URI is 
used as an annotation property and treat them as annotations (which 
means reasoner will ignore them completely). I have seen at least one 
ontology where OWL-API parser gives ABox assertions to Pellet but the 
Jena loader reads the same triples as annotations. But this would only 
cause problems if you have lots of classes and even then it wouldn't 
really be a big problem if you don't have nominals (which SNOMED does 
not contain AFAIK).

Playing with logging outputs [1] would give more detailed information 
about the differences between two cases. Setting

log4j.logger.org.mindswap.pellet.KnowledgeBase=info

would print what is the expressivity of the ontology loaded into Pellet 
and the number of classes, individuals, etc. Setting

log4j.logger.org.mindswap.pellet.taxonomy=debug

would print all the satisfiability and subsumption tests performed and 
how long each test takes. That would help narrow down where the problem 
occurs.

Cheers,
Evren


[1] http://pellet.owldl.com/faq/logging
>
> On 7/15/07, * Ron Alford* <ronwalf at umd.edu <mailto:ronwalf at umd.edu>> 
> wrote:
>
>     howard goldberg wrote:
>     > I have an 18K fragment of SNOMED anatomy with
>     transitivity.  Classifies
>     > in 4 minutes from the pellet command line, takes about an order of
>     > magnitude longer within protege 4--over an hour last time I let
>     it run.
>     > Traced the protege classification within the debugger, spending time
>     > within the phaseone classification loop. It would seem there is
>     either a
>     > representation difference or some missed optimization in p4 that
>     > accounts for the much slower performance.  FACT++ in p4
>     classifies this
>     > ontology in 3 to 4 minutes.  Had posted to the protege4 list;
>     Matthew
>     > Horridge asked me to post over here as well.
>     >
>
>     Can you post which version of pellet you are using?  I can't recall
>     which version of pellet protege uses, but you might try replacing the
>     jar files with your zippier version.  There was a goof-up in which
>     optimizations were enabled by default in 1.5RC1.  See
>     https://cvsdude.com/trac/clark-parsia/pellet-devel/ticket/23
>     for details.
>
>
>     -Ron
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Pellet-users mailing list
> Pellet-users at lists.owldl.com
> http://lists.owldl.com/mailman/listinfo/pellet-users
> _______________________________________________
>
> Sponsored by Clark & Parsia, LLC http://clarkparsia.com/