Possibly stupid thought, but rather than just applying these techniques to the whole of thingspace, shouldn’t one first do something like this: get a bunch of incoming data, create density map (by measuring at various resolutions, perhaps), look for regions with density significantly different from others (ie, first locate some clusters) and then apply these techniques in those regions rather than over the entire whole of the space to help learn the properties of those clusters?
Possibly stupid thought, but rather than just applying these techniques to the whole of thingspace, shouldn’t one first do something like this: get a bunch of incoming data, create density map (by measuring at various resolutions, perhaps), look for regions with density significantly different from others (ie, first locate some clusters) and then apply these techniques in those regions rather than over the entire whole of the space to help learn the properties of those clusters?