[Pellet-users] Differences based on Interface (was: RE: Restrict domain inference)

[Michael Smith]

On Tue, 2008-01-29 at 13:37 -0700, Swanson, Tim wrote:
> Michael Smith wrote:
> > 
> > What interface are you using to access Pellet?  This can affect the
> > amount of reasoning performed.

> Would you mind expanding on this point? I'm curious as to precisely how
> the interface used (in particular command-line vs. 3rd party API)
> affects the amount of reasoning.

I was thinking, in particular of some issues with the Jena bindings
where the calling application inadvertently requests more information
than it needs.  See messages [1],[2] for examples.

Generally speaking, the Pellet API aligns more comfortably with the
OWLAPI than with Jena.  Where the APIs do not align directly, there is
the *potential* that additional reasoning is performed.  And, to be
fair, these problems are mostly due to how the interfaces are typically
used, not how it can be optimally used.

> Does this mean that there could be justification for invoking Pellet
> directly (using ATerms, as opposed to via the OWL API or Jena
> interface)?

The Pellet KnowledgeBase interface (using ATerms) is the most direct,
and gives the most control over what reasoning is and is not performed.
Unfortunately, it lacks a clear distinction between public and internal
API elements, is not elegant to work with and the ATerm representation
of data is cumbersome.  In most cases we discourage applications from
using the ATerm based API.

We find working with the OWLAPI to be ver pleasant, and that is
generally where we do application work.  If someone is working in OWL
and doesn't need an RDF view of the data, I discourage using Jena as the
Pellet interface.

> Is there currently any documentation on this issue?

Only that which exists in the mailing list archives.

-- 
Mike Smith

Clark & Parsia

[1] http://lists.owldl.com/pipermail/pellet-users/2008-January/002322.html
[2] http://lists.owldl.com/pipermail/pellet-users/2008-February/002348.html