Wednesday, October 06, 2004

Bowtie Operator
I've just had a look at yesterday's blog using several different browsers on the Mac. IE, Firefox and Camino, all show unicode character 8904 (⋈) as a question mark (?). Safari was the only browser that rendered this unicode character properly. I have no idea what the rendering will look like on the various apps in Windows or Linux.

I keep IE only for those sites which insist upon it (fortunately there aren't many of those). I use Firefox for blog entries, as it provides the full XUL interface that Blogspot offers. Camino seems to offer all the advantages of Firefox, only with the lovely native Aqua widgets. Unfortunately, it does not render the blogger editor correctly, so I rarely use it.

My browser of choice is Safari. This is because it is fast, attractive, and 99.9% compatible with most sites. Importantly, it is fully integrated with the other OSX apps. This is the feature which has me using the OSX Mail program instead of Firebird (even though I feel that Firebird has a better interface, I just can't believe that it does not interface directly into the OSX address book).

It now seems that Safari is the only browser which can render certain unicode characters. This only encourages me to keep using it.

Some of today was spent on owl:differentFrom. This was much easier than owl:sameAs as there is not a lot of entailment that can be made on this predicate.

While documenting the predicate I wanted to keep saying "different to" instead of "different from", so I thought I'd look it up. It seems that "different from" is the common expression in the US, while "different to" is the more common British (and hence, Australian) usage. Since most people reading this will be in the States I opted to stick to "different from". Besides, it matched the OWL predicate as well.

The only consistency check for this predicate was one that had already been done for owl:sameAs. While I was tempted to simply refer to the other section, I wrote the example out again, along with an explanation which was more appropriate for the context.

I spent some time looking at the owl:allDifferent predicate, with some frustrating results. This predicate uses the very lists I hate in RDF. As a result, the only way to select each item in the list is with a walk(...) operator.

Unfortunately, the walk(...) operator requires a fixed end to start from, but the end in this case is a blank node. The only way to "fix" this end is to find it in an outer query and then pass it down via a variable into an inner query which performs the walk. This would seem OK, only the walk function does not accept a variable as its starting point (even one with a fixed value as would happen in a subquery). Late in the day AN said that he could fix this relatively easily, so I am hoping he can have it done early tomorrow.

I had a brief work with AN about the need to find out if a node meets a description as defined by the OWL documentation. He is not keen on the idea of a new function like I proposed, so we discussed some options.

The most prevalent option is to let the rules engine do the work here. The problem is that there are no OWL constructs which can describe that an object meets the conditions of an owl:IntersectionOf or many of the other constructs which can make up a description. AN pointed out that we can use an internally defined predicate to describe this type of thing, which would solve the problem.

If we took this route then we are still left with a couple of options. The first is to define the appropriate rules in the rules engine which can generate these statements. The engine would then be able to derive the entire set of description statements for all the objects in the system. While this would work, I believe that it would be slow, particularly as it would need to traverse the entire iTQL layer for each invocation of the rules.

The second option is to implement the function as I initially proposed it, but to insert the description statements as each recursion returns. This would then permit subsequent invocations to use these statements instead of the original algorithm, providing a significantly reduced workload. Other unrelated operations may also be able to take advantage of knowing which objects meet which descriptions.

AN also mentioned another option which involved performing a lot of intermediate operations with the use of iTQL. While this would obviously work, I do not like the idea of re-entrant iTQL. Besides, it suffers from the inefficiencies of the iTQL that I have already mentioned.

Whatever the outcome, we really need to see some way of letting an inferencing query determine if an object is of a given "type", without that query needing to do any difficult work in that regard. Many of the required inferencing queries are complex enough already without adding to their difficulty.

Another advantage of checking for description compatibility in some internal fashion is that it allows several steps of a rules engine to be skipped. With a check like this it is possible to immediately determine that an instance of a class is also an instance of a superclass (and all superclasses above that). With the purely rules-based approach this information will only be available after executing a number of rules, each of which involves another transition through the entire iTQL layer. While that approach will work, it would be significantly slower than simply giving a rule immediate access to all types for an object.

I suppose we still need to discuss this before any decisions can be made.

No comments: