[Pellet-users] Extremely long inference times

[Evren Sirin]

In summary, the problem you are seeing is due to the structure of the 
axioms in your ontology. The axioms you have do not play well with the 
optimizations in Pellet. Specifically, the subclasses of Report class 
has two axioms like the following ones:

ReportMentionsAnimal-Monkey subClassOf Report
ReportMentionsAnimal-Monkey equivalentTo (mentionsTopic some 
TopicAnimal-Monkey)

Generally, it is desirable to have a single definition (equivalent 
class) axiom for a class with no other subclass or equivalent class 
axioms. This is better both from modeling point of view (write a single 
definition that will infer subclass relations) and from performance 
point of view (otherwise reasoner has to deal with disjunctions which 
makes things harder).

In your ontology, the above two axioms would be semantically equivalent 
to the following single axiom:

ReportMentionsAnimal-Monkey equivalentTo Report and (mentionsTopic some 
TopicAnimal-Monkey)

Making this change to all the subclasses of Report reduces the reasoning 
time two orders of magnitude and requires no additional memory. Attached 
is the ontology I created from 
SafetyClassification_Classes_Rules_001-1000_after_SWRL.owl with a simple 
program (Note that, the SWRL rules and related imports were removed 
during the transformation). Running this version through Pellet command 
line I got the following reasoning times:

Time: 152938 ms (Loading: 2658 Consistency: 220 Classification: 2957 
Realization: 147103 )

FYI, the timings I get with the forthcoming Pellet version looks better 
w.r.t. realization since we improved ABox reasoning considerably:

Time: 34233 ms (Loading: 2788 Consistency: 200 Classification: 2652 
Realization: 28292 )

Cheers,
Evren

PS: The definitions for subclasses of other classes (e.g. SearchTerm) 
seem to have the same issue but since the instances in those cases have 
also explicit type assertions there is no performance problem.


On 6/3/08 8:39 AM, Liu, Richard wrote:
> Hello,
>
>  
>
> Martin O'Connor at Protege suggested that I send this problem and  its ontology 
>
> (attached) to the Pellet list, as he thinks it is a Pellet problem.
>
>  
>
> The project SafetyClassification_Classes_Rules_001-1000.pprj uses 
>
> SafetyClassification_Classes_Rules_001-1000.owl and some ontologies  in 
>
> OCRePublic.  Before running the Jess rule engine on rules  contained within the 
>
> ontology the Pellet 1.5.1 (direct) reasoner  infers in ca. 146 sec.  After 
>
> running Jess using only the rule  Discover_Titles_in_Reports and importing 718 
>
> new "facts" into  ontology  (SafetyClassification_Classes_Rules_001-
>
> 1000_after_SWRL.pprj)  inferencing takes 7332 sec.  Moreover, it requires a heap 
>
> size of  more than 800MB (I succeeded in running with 1GB, but perhaps a bit 
>
> less would also suffice).  I am wondering what I can do to  significantly reduce 
>
> the inferencing times, especially that after  incorporating the new knowledge 
>
> into the ontology) and, if  possible, the required heap size.
>
>  
>
> I am running on a MacBook Pro Core 2 Duo, 2.33GHz, 2GB RAM with OS  X 10.4.11 
>
> (Tiger).
>
>  
>
> Regards,
>
> Richard
>
>  
>
> ____________________________
>
> Richard Liu
>
> Group Research Information
>
> Scientific Information Resources
>
> F. Hoffmann-La Roche Ltd.
>
> PXIBR
>
> Bldg. 092/6.01
>
> CH-4070 Basel, Switzerland
>
> Phone: +41 (0)61 687 92 14
>
> FAX: +41 (0)61 688 95 70
>
> richard.liu at roche.com
>
> Confidentiality Note: This message is intended only for the use of the named recipient(s) and may contain confidential and/or privileged information. If you are not the intended recipient, please contact the sender and delete this message. Any unauthorized use of the information contained in this message is prohibited.
>
>  
>
>   
> ------------------------------------------------------------------------
>
> _______________________________________________
> 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/

-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.owl.zip
Type: application/x-zip-compressed
Size: 24038 bytes
Desc: not available
Url : http://lists.owldl.com/pipermail/pellet-users/attachments/20080604/9245347b/attachment-0001.bin