Friday, October 08, 2004

One of the big plans for Kowari is to introduce the "Resolver" SPI. This is a set of interfaces which is used to wrap any datasource to allow it to be a backend for the Kowari query engine.

Using appropriate implementations of this interface, it will be possible to perform queries on all sorts of different datasources just as we currently perform queries on the Kowari datastore. Some obvious example datasources are MS Exchange servers, Oracle databases, and RDF-XML files. With Kowari's ability for a query to access multiple models at the same time, the Resolver SPI will allow for very powerful realtime data access.

To make this work, the current query code has to be split up from the Kowari datastore, to channel all access through this new interface. This will make the Kowari datastore "just another" backend that can be plugged in. The main advantage of the Kowari datastore over any other is that it should be the fastest possible backend for the kind of data that the resolver interface uses, and it will support all possible features. Other interfaces will not necessarily be capable of supporting every feature. For instance, an RDF-XML file will require access to be restricted to read-only.

Changes like this are far reaching, so we need to momentarily concentrate all our efforts on these modifications. If we worked on anything else in parallel, it would most likely need to be repeated to work with the new changes.

With this in mind, we had a meeting today about the different subsystems which will be affected by the changes. After determining all the major sections, we each picked up a few, and will spend the next couple of days working out just what is involved to make the changes. This should give us enough information to estimate how long it will take.

Node Types
Late in the day TJ discovered that one of the machines did not run the full suite of JXUnit tests correctly. The failing test turned out to be the search for literals in the node type tests.

The test finds the one string and one number found in the Camera ontology. For some reason, the test fails by only returning the number. As with the earlier problems, this only occurs when the full WordNet database is loaded. This indicates that there is some problem with the use or implementation of the HybridTuples class.

There are two places this could go wrong. Either the string is not being returned by the node-type model, or the result is not being joined against correctly. This is easy to determine, as it is possible to just return the literals and not join the result against anything. If both the string and the number are in there, then the problem is not with the node-type data being returned, but with how it gets joined to the other data.

Of course, with WordNet loaded, the result of all literals is huge. Because of this, the appropriate method of query is via the command line interface, with the results being streamed to a "greppable" file. Unfortunately, at this point the network had to go down due to a UPS replacement for the file server (it started beeping at lunch time). Given how late it was on a Friday, we took that as our cue to leave.

At lunch time I went along to UQ as one of the staff members in ITEE was retiring, and they were putting on a free lunch. This gentleman had shown remarkable patience with me when I was finishing my first undergraduate degree, and I've always been particularly grateful to him for it. Besides, who am I to turn down free food? (I skipped the beer though, as I had to return to work) :-(

While there I met one of my old lecturers who asked about my current study. At one point in the conversation he asked when I graduated, and when I responded he made the comment, "Oh, that was before they cut the course." This then led to a conversation about the engineering degree at UQ.

First a little background...

When I went through we had a lot of subjects, and they were really hard. They were harder and required significantly more work than any subjects in my physics degree. Engineering decreed a minimum of 32 hours a week (I forget what that translates to in credit points).

I don't know about the other departments, but when I was first studying I was aware that Electrical Engineering was in a quandary about what to teach the students. Institutions such as IEAust, the IEEE and the IEE lay down a minimum set of requirements for a degree to follow in order to be recognised.

Operating in opposition to this, the university set out a time period for the degree (4 years) and a maximum number of credit points that a student could enroll in. With the restrictions imposed by the university, and the requirements imposed by the institutions, the department created a series of subjects with an enourmous amount of information crammed into them.

There were also further constraints imposed, as mathematics, physics and computing subjects had to be provided by the Science faculty, but these subjects were not as "crammed" as the engineering subjects. Actually, I used to benefit from these subjects, as they did not require much work in comparison to engineering subjects and they allowed me to relax a little.

While Australian universities have been offering fast-tracked engineering degrees in 4 years, other countries take a different approach, taking 7 or more years. This is probably of benefit, as the students would retain more information, and fewer of them would burn out.

Back to more recent times...

In the late 90's, UQ decided that they wanted continuity among all of their courses. A Commerce degree had a different workload to an Arts degree, and this led to differing costs for HECS. Since many faculties did not support an extra workload, faculties such as Engineering were told to cut the work requirements of the students. As a consequence, the hours required of students dropped from 32 hours a week, to 20!

The natural consequence of this, was that subjects had to be dropped. Entire areas of knowledge are not taught to students. Control theory (a difficult topic) has proven to be one of the most valuable subjects I was taught, and yet it is no longer available to students.

If you can't learn these topics at university, then where can you learn them? Research degrees don't cover them. One solution proposed has been to put this information into a coursework Masters degree. This is certainly the approach taken in the US and UK (hence my previous post referring to Masters degrees from those countries not equating to the same qualification here). But this then means that undergraduate students at UQ are not getting the degree they thought they studied and paid for.

My question is now why the IEAust, IEEE and IEE have allowed this. They can't prevent UQ from changing its courses, but they can certainly drop their accreditation of UQ degrees. Having a degree from an unaccredited university is useless.

I'm at a loss as to why none of the institutions have acted on this. There are requirements for an electrical engineering degree, and UQ has suddenly stopped meeting them. At this time the IEE accepts the accreditations provided by IEAust, and I'm guessing that the IEEE do the same. So why haven't IEAust done something about it? Obviously it is political decision (how could they take away accreditation from one of Australia's top "Sandstone" universities), but in the meantime students are missing out, and industry is not getting the professionals it requires.

I will be raising this at the next IEE committee meeting. It has been raised in the past, but at the time a "wait and see" approach was taken. However, last time it was suggested that if things were bad enough then it would be possible to stop accepting IEAust accreditation. Perhaps that would make the IEAust wake up and do their job.

No comments: