search

owlcs / hermit-reasoner

12 stars 3 forks source link

hermit reasoner: python vs protege #10

Open annagrotius opened 4 years ago

annagrotius commented 4 years ago

I have created an ontology and then used the Hermit reasoner to infer triples. I used the Hermit reasoner via Protege (after uploading the ontology created through owlready module), and I used the Hermit reasoner with the owlready module over the same ontology file. I receive different inferred triples. Why is this? Thanks!

ignazio1977 commented 4 years ago

Hi, we need an example of the different triples you see (and the ontology, or an ontology that reproduces the issue).

annagrotius commented 4 years ago

Sure I have atttached the ontology, and a file that has the shared triples and a file that has the differing triples.

I assumed that the issue was that the Hermit reasoners were simply reasoning differenlty. Or should they reason the same via both platforms (protege vs the python owlready2 module)?

ontology.txt

differing_triples.txt

shared_triples.txt

ignazio1977 commented 4 years ago

Assuming it's the same version of HermiT running in Protege and with owlready2, the inferences will be the same. However there are a number of configuration options that might influence some details.

ignazio1977 commented 4 years ago

The ontology has some peculiarities which may reflect on how the reasoner - and more importantly the APIs around it - might work.

First thing I noticed, the ontology file has two ontology declarations, meaning that the parser will have to arbitrarily decide which one is the ontology IRI. This is probably the reason for the differing triple about the ontology.

Then, the property used_with is symmetric and has domain set to the intersection of Cutlery and Plate, and its domain is Food. I'm not sure if symmetry is intended here, as it makes Food equivalent to the intersection of Cutlery and Plate, which are apparently disjoint (all subclasses of Cutlery seem declared disjoint with Plate).

used_with is inferred to be equivalent to bottom property, and so is spread_on - I assume from the relations between domains and ranges for these properties. Many other properties are inferred to be equivalent to these, it seems (I haven't analysed the whole of the files). This would end up making many classes unsatisfiable. I'm wondering if the ontology is working as intended or if these are bugs in its design.

annagrotius commented 4 years ago

Thank you for looking into this!

So the reason there are two ontolgy declarations is because I link two ontologies together before reasoning over them. I do this with another script and link the ontologies with one triple that contains a 'owl:equivalentClass predicate', connecting a class from both ontologies. This was done using the Rdflib Graph module in Python. Otherwise, do you recommend another way to go around this?

I will look into the other properties defined that you brought up. Bugs are not intended for the design.

annagrotius commented 4 years ago

Also, I noticed that the Hermit reasoner I am using is Hermit 1.4.3 (in Protege version 5.5). What version of the Hermit reasoner is used in owlready?

ignazio1977 commented 4 years ago

That's a question for the owlready team, I can't answer it.

ignazio1977 commented 4 years ago

Don't copy over the ontology declarations, create a new one. Only copy the axioms.