A short follow up to the earlier post on image analysis. We can cram another principal component into the visualisation by going 3D. This animation depicts the first three components as spatial coordinates, and the second three as colour coordinates.
(There is a little POV-Ray sphere for each data point.)
There is an obvious shape and colour distribution. For instance, one end of the main body is blue, and the other is pink. I have to wonder if that implies a bug in my PCA algorithm. It is supposed to select orthogonal components.
How can I test it? It’s a bit hard with 13 dimensions for each of 136742 data points. A smaller test case is necessary.