K-means clustering is a classic clustering algorithm. So what? It’s seduced me to draw the conclusion that never rely on classics!
And this time I’ve got 400,000 72-dimensional (and 90) vectors to be clustered! I’m running an executable from the web. Then? It’s been a plight for me, as well as for the computer (which is decent with duo core 2.8GHz processor and 3G RAM). The fan runs, NOISILY, for several days, but there’s no output. As a skeptic myself, I searched for couples of other implementations. Unluckily, none turns out to be really swift!! And I started to feel the ”itches” introduced by something that’s theoretically feasible but practically intractable!
Perhaps it’s time to relook into the problem and redesign the algorithm! The moral of the story: Don’t be taken by the beauty of theoretical analysis, put it into practice!