I did a some more color matching work and encountered a problem at evaluating the match of the grays. At low saturations the error in hue can become very large (up to 180) and can pull down the CMPI substantially. This can occur even if the matched color is not far away from target color in terms of overall distance in color space. To avoid this one could calculate this distance but i decided to keep it dead simple and come up with a quick and dirty remedy. Where hue error is beyond 50 i set it to the value of 50. To keep the tests comparable always 2 of the 10 squares have a (randomly generated) saturation below 10 and higher saturations for the other 8 squares.
The CMPI of my second test was 61.3. I will do dozens of these trainings in coming weeks and see whether the scatter of these tests results shows a tendency upward.
