||[Nov. 9th, 2007|01:14 am]
I just realised why I don't like continuous assessment in modules.
My rate of work is not constant, nor can I get it to be. I seem to operate on what I call the "Eureka principle". I will know something, but it won't be until it's put another way, or just something else happens that the knowledge integrates in my head and I actually realise what it is that was taught to me a few months back.
When I was studying for my leaving cert, I studied Accounting. It wasn't bad, but for ages I just didn't get it. I mean, I could sorta get it, and if I concentrated I could just about get it done, but just follow the basic ideas and not be able to apply it in all situations because I didn't understand what I was doing. Then when we got to the "tricky" bit of the course, suspense accounts, I saw what was going on. Eureka! I suddenly understood all that was being done. In my mind I massively improved. I think, in reality, I did actually improve. In my leaving I got a B3, which is nothing to be sniffed at.
Frequently this happens to me. When I was starting in Sun, I sat down to learn python. About a week in, without the idea even being mentioned, I realised what polymorphism actually really meant. I always got some idea of it, but sitting down with python just made it all clearer to me. Mel tried to teach this in the second semester of 3rd year, and I just couldn't make it a core idea. But, one day down the line: Eureka! It turned out I did well enough in his course anyway, but that's an extreme case. Other people say they got what he meant when he was teaching it, but it took me a while to pull it all together into my head.
Another example was in Databases - relational calculus, once I realised that it, and some maths I was studying, were using the same notation (sets), the world worked and I could do it. Until then, I just followed rote. But then, Eureka!
So, this is why I don't like continuous assessment. Chances are, by the time you're testing me, the knowledge hasn't had the time to settle. Case in point, I had a Eureka moment this week. I study Data Mining with Tahar, and also Machine Learning. Tahar believes in continuous assessment, the Machine Learning lecturer just has a few labs/tutorials and the rest of the marks (about 60-70%) are from the final paper.
It was mentioned at one point that the stuff we were learning in Machine Learning was also known as Data Mining, but with less of a focus on how the data was gathered/cleaned and so on.
Also, it's probably worth mentioning why I took Data Mining: personal interest. At one point I studied Psychology in the University of Surrey, and had to take statistics. Now, I don't have the book any more, but it mentioned data mining, which it gave a one-sentence summary of something like, "Data Mining allows the acquisition of reliable conclusions/inferences/facts from flawed/noisy data". Back in the day, that pricked my ears and I wanted to learn this magic. So, I bit the bullet, despite not really liking Tahar as a lecturer (although he's not as bad with 4th years - I think it's partially because he likes this subject - although I've not really seen him get passionate about it, but he's more passionate about that than Operating Systems), and it turns out that it's the same as Machine Learning with a bit of explanation as to what data warehouses and data cubes are, and some mention of Data Reduction.
Now, this is the thing. Data Reduction, he just glossed over that. Then he went into some basic statistics, which is what data mining is not - well, it can be based on your standard formulae like standard deviation and error and all that jazz - and there wasn't a lot. That was a few weeks in. He then expected us to understand why reduction was so important. I mean, yes, I got that fewer variables let you clear your view and reduce computation, but it wasn't exactly huge.
Then, in machine learning, just this week, our other lecturer explained "Dimension Reduction" as a machine learning process. He explained that "In higher dimensions, things look more similar" or "the higher the dimension, the more similar everything looks". Eureka!
Today, he explains Dimension transformation and it's truly elegant and beautiful and would scare most arts students hollow (I'm not being deliberately arts-ist - but very few people other than those studying multi-dimensional maths would like this (and yes, there are arts students who study maths... hence "most"), I mean I don't get the finer points but I know enough to get a grasp of what he's doing). Double eureka! I saw how something I read 8 years ago, in a book aimed for psychology students, works. It really does "mine" for hidden data. It really does reveal things that are true, but just aren't apparently obvious (and it seems that the rest of our Data Mining course is just a bunch of cheap machine learning tricks - which Tahar isn't teaching too well).
So, Dimension/Data reduction - I understand it now. I really could explain it now, and it would work in my head, and I could answer it in a final paper. But, I've already dealt with it in continuous assessment - I'm not really going to be revisiting it, except in a very final exam worth at most 20% of my final grade, and there's no assurance that it will be there.
Instead, I've lost marks that I would have gotten if it was a final paper. This isn't about cramming, it's about integration and letting it all snap into place. It's about truly understanding something, which won't always happen immediately. I'm sure there are other things cooking away in a toaster in my head, just ready to pop them out when they're ready. Unfortunately, I just don't know when that will happen.