Optimization
I’ve recovered from last week’s stupidity (and started using version control!), and I believe I’ve got a feature extraction program that works. Yay! Now comes the drudgery: trying to find the optimal combination of speed vs. accuracy. This basically consists of running batches of songs through the feature extractor using different parameters (sampling rates, length of song segment analyzed, analysis window sizes, etc) and then comparing the results.
While it’s tough to get motivated for this relatively boring portion of my project, I’m glad that it’s only taken a couple of weeks to get this far. From my timeline, I’m right on schedule. I’ve still got plenty of problems to solve, like finding the best way to store feature vectors, the playlist generation algorithm, and of course, the actual Rhythmbox plugin. Fortunately, most of the optimization testing stuff is automated, so that can run in the background while I get cracking on the rest.