The spectral clustering was a disappointment. It really didn't find anything that wasn't glaringly obvious and it missed several features that I could see by eye.
I think the problem here is that I'm not very clear on the question that I want answered. I have some data, I see some patterns, but I'm not sure what `answer' I want to get. Some patterns in the data are `interesting', others are not. I can't highlight the interesting ones unless I can formally specify why they are interesting in a quantifiable way.
Programming is hard. Thinking about what you should program is even harder. Maybe I should go shopping.