mark nottingham 0.7

Thursday, 17 March 2005

Python Semantic Web

I’m happy to announce that version 0.7 of, a simple API for RDF, is now available. As always, feedback and suggestions are appreciated.

My goal for this release was to clean up cardinality. In particular, I wanted to make Sparta usable with the Atom OWL work that’s been going on. Unfortunately, it’s difficult to show — both because Atom is a moving target, and due to a few flaws in the stylesheet and model — but with a little tweaking, I believe it’ll be pretty straightforward once Atom settles down and they catch up.


The most obvious change in this version is in how cardinality is determined. After actually reading the OWL docs, I think I’ve done it correctly; an object will be considered unique (thereby giving you direct access to it, rather than through a ResourceSet), if:

  1. The predicate is an owl:FunctionalProperty, or
  2. The subject is subclassed to have an owl:Restriction on that property with either an owl:MaxCardinality or an owl:cardinality of “1”.

Practically speaking, this means that where you previously said that a property had owl:maxCardinality of “1” to work with Sparta, you’ll now need to say it has an rdf:type of owl:FunctionalProperty.

Also, __init__ is a little more picky about the properties that you instantiate it with; if something isn’t considered unique (as per above), you’ll need to wrap it in a list; e.g.,

bob = Thing(None, rdf_type=["people_Person"], person_age='23')

assuming that person_age is constrained in cardinality, and rdf_type isn’t.


Assignment can now copy between different stores, so if you’re jugging two rdf datastores with Sparta, you can do things like

bob = StoreOneThing("person_bob")
mary = StoreTwoThing("person_mary")
bob.person_wife = mary

and it will work. This involved adding the copyTo method, which supports directly copying statements from one store to the other.

You can also pass a URI string when working with attributes, which is helpful if you don’t know what prefix to use with it. Although this isn’t too useful with direct access (Python syntax doesn’t allow object properties named bob., it is useful with getattr and friends (e.g., getattr(bob, “”)).

Rich comparison support means you can now compare objects and know if they represent the same underlying RDF node; e.g.,

if bob.wife == Thing("person_mary"):

Finally, one under-the-covers change that should make Sparta a lot more compatible with existing, real-world RDF is that it now does the right thing when nodes have multiple rdf:type and rdfs:range values. The currently implementation is correct if inefficient, but I expect optimisation will be fairly straightforward.

What’s Next

I’m not going to say that this is feature-complete, because I’ve said that too many times before. I still need to dig into RDF and OWL a bit more to see where else I can find nice things to do, and still want to think about typing list members.

After the API and functionality are stable, I’ll look at more documentation, refactoring, optimisation, and documentation, as well as a Redland back end.


Danny said:


Re. the Atom OWL material, I haven’t done anything since those first drafts, but Henry’s* done quite a lot of work trying to get a good mapping, in part for BlogEd **, the Java blogging tool.

I was planning on updating the docs/code once the Atom spec has settled down, based on what Henry had come up with. But having on the case should help a great deal.

Dave Beckett’s now got some Atom support in Redland, so that should help inform choices too.

While you’re digging into OWL, I’d be grateful if you could bear something in mind. I have no idea yet how feasible this would be, but if a significant proportion of the Atom structural rules can been expressed in OWL, then a maybe kind of semantic layer validator might be possible (i.e. bung the OWL ontology into a model along with the instance data in question, then run the whole lot through an inference engine to look for inconsistencies).



Friday, March 18 2005 at 2:31 AM

Henry Story said:

Oh, nice to see about Sparta :-)

I have now summarised my work here as things have moved on quite a lot since the original atom owl model. Since I have a model that is now written in java code, it would be possible to write something that extracts the owl model from the interfaces. The latest BlogEd from CVS is now quite functional.

Sunday, March 20 2005 at 4:42 AM