I understand what you say, and I don't disagree with your analysis for the collection of datapoints on nonintersecting spherical surfaces. (Although I do frown at this type of dataset, since distribution densities tend to peak at their peaks, the "in high enough dimensions everything lies on a sphere" is a misconception: it is only the radial histogram that peaks at a nonzero radius...).
Just thought that it might be worth pointing out that there are valid use cases of clustering with real data reference for each class available: consider pictures, to be inspectable it would be helpful to see a real picture instead of some blurry interpolated mess.
Would you mind providing a reference for the bolt vector quantization algorithm? It sounds interesting.
Just thought that it might be worth pointing out that there are valid use cases of clustering with real data reference for each class available: consider pictures, to be inspectable it would be helpful to see a real picture instead of some blurry interpolated mess.
Would you mind providing a reference for the bolt vector quantization algorithm? It sounds interesting.