[Pellet-Users] saving the classification/realization state of a knowledgebase

[Bijan Parsia]

On Oct 9, 2006, at 2:41 PM, Alan Ruttenberg wrote:

> On Oct 9, 2006, at 3:48 AM, Bijan Parsia wrote:
>
>> On Oct 9, 2006, at 6:13 AM, Alan Ruttenberg wrote:
>>
>>> I wonder if someone could sketch out some of the issues with saving
>>> the post-classified state of a knowledgebase. If I want to work with
>>> a kb that takes a while to classify/check (e.g. Galen), what
>>> structures need to be saved so that I could restart, repopulate  
>>> them,
>>> and continue querying the KB without reclassifying.
>> [snip]
>> It's not primarily saving the post-classified state of a knowledge  
>> base that's an issue if you intend to update the restored KB. In  
>> vanilla Pellet, everything is thrown away and restarted if you  
>> modify the KB. Christian, of course, is working on incremental  
>> update of the KB, and has some reasonable success...dramatic  
>> success with ABox update (ords of mag). We are working on TBox  
>> update again.
>
> That's very cool - looking forward to it. I've been playing around  
> with productions the consequents of which update the knowledgebase,  
> just toy size right now, but that could form a test case.

Cool.

> However, just being able to save the state, without updating, would  
> be useful - for instance to be able to start up a server with an in  
> memory model to make available for queries only, as in working with  
> an application that uses some fixed ontology.

Well, maybe. Pellet no longer classifies before doing a query. Now,  
it probably would be helpful to force than, then save and reload.

>> He has done some work to incremental query update (in the context  
>> of streaming queries), so that'll be part of the package.
>
> I don't know what you mean by "streaming queries".

Queries over rapidly fluctuating data, e.g., over the output of a  
sensor.

> By query update do you mean delivering incremental answers to a  
> query that is registered, as the kb is updated?

Yes.

>> I think there is little point to serializing the structures until  
>> incremental update is in place (you might save on parsing and  
>> preprocessing, but eh; ok, there are times when parsing dominates,  
>> but that's just morally wrong! ;)).
>
> I agree, but it seems that it's really the time to classify/check  
> consistency, realize that matters with some of the ontologies.

Definitely. It varies quite a bit, sometimes one way, sometimes another.

Cheers,
Bijan.