The Data Mining assignment clearly stated that a simple implementation was sufficient. Still, just throwing together the DBSCAN algorithm from pseudo code in the paper didn't seem like a really great learning experience. So, I've gone ahead and done the things you do when you're serious about programming. Like, full unit test coverage, proper segmentation of the problem into model, data, and controller classes (no view, since it's a command line app), and interfaces so you could hook up a high-performance spatial database to the data layer without modifying the algorithm. I'll give it a nice writeup, too, along with some pretty graphs our of R. I'd say I've probably tripled the work necessary.
I'm doing fine and know I could get by with less, but looking for the minimum successful path is what got me in trouble at Cornell. This go round I'm focusing on maximum output. It may be sub-optimal, but it's the surest road to success. And, if this try doesn't work out, there won't be a third.
No comments:
Post a Comment