Tuesday, May 25, 2004

Testing Trans
The trans statement had some bugs this morning, but they were all found and fixed. On the verge of checking in, and a completely unrelated JXUnit test failed. Will have to chase this down before anything gets committed.

More correspondence. Yes AN, some of that was you. :-)

Finally reading AN's plan for implementing inferencing.

I'll confess that inferencing has me intimidated. I know we can do it, but we need it fast and scalable, and that is the tough part. While some ways of implementing will be fast for some situations, they will suffer on other occasions. eg. Inferring from a large model that all gets loaded in one hit would be most efficient if done during the commit phase of the load. A regularly changing model will suffer badly from that strategy, and would do better to be inferred from when queried.

AN is intent on getting it working, and then making it scalable, even if that means reimplementing. That eases my mind somewhat, as it lets me concentrate on what is involved in inferencing without being distracted by scalability issues.

The other thing that I've been considering has been how an ontology model can be used as the inferencing rules. Up until now I've been just considering the RDFS rules, and these have been best expressed with the entailment-rdf-mt-20030123.xml file. However, that won't work for ontologies. Sure we can hard code in the OWL rules (subClassOf is transitive, what minCardinality means, etc) but then the application of the ontology will be the tough part. I think I need to sit down with AN for some quality time on this. Good thing he's back next week. Hope he's not too jet-lagged!

No comments: